Finding a holistic solution to the problem of algorithmic tacit collusion

42 already being, perhaps in a nonconscious manner, shown by some scholars to be complementary in nature. Moreover, this example also shows that the use of ex ante regulation can help the CJEU to avoid coming up with over extensive and hard to justify interpretations of the rules of EU competition law for the purposes of covering gaps in its scope of application.

Furthermore, another evidence for the complementarity of competition law and regulation is the fact that they can still reinforce each other even in cases where there are two rules (one derived from each regime) that may at first glance appear to be completely different. For

example, let us imagine that the proposal by Calzolari of categorizing tacit collusion as a form of concerted practice becomes reality, together with the proposal of Bernhardt and Dewenter of limiting the speed at which algorithms can perform their operations. Even though these two rules appear to be unrelated they can produce a common cumulative effect. The categorisation of algorithmic tacit collusion as a form of concerted practice would deter undertakings from engaging in it. Nevertheless, undertakings are still going to be tempted to do so since colluding in such a way while remaining undetected will result in higher dividends. However, if at the same time, a rule limiting the speed at which algorithms can perform their operations is in place, this would make achieving collusion harder and perhaps even increase the likelihood of the undertakings being caught. Consequently, even though the two rules may seem too distinct from each other they can still contribute to the attainment of the same end-goal and do so more effectively together than they could have done individually.

7.2. Finding a holistic solution to the problem of algorithmic tacit

43 simultaneously and if yes how they will interact with each other. As a result, this section seeks to follow this line of reasoning present a proposal for a solution to the problem that is in line with it.

To start with, as has already been discussed in chapter 5 there is a plethora of remedies that have been proposed. However, in order to be successful, a holistic strategy for tackling algorithmic tacit collusion has to do at least three things. First, it has to implement remedies that can be enforced in a fast and efficient enough way. Second, these remedies have to be theoretically coherent with the legal regime from which it is derived (e.g. competition law remedies created through judicial interpretation have to be well justified). Third, each of the remedies must not impair the effectiveness of the others. Having in mind these three requirements, this paper now turns to propose a solution.

Since the problem of algorithmic tacit collusion is by its very nature borne out of a gap in EU competition law, it makes sense for it to be able to at least partially address it by using its own tools. One of the best ways to do so, is by incorporating algorithmic tacit collusion under the concept of concerted practices. This can be done in accordance with the proposal of Calzolari that has been discussed in section 5.3. According to him, in order to treat algorithmic tacit collusion as a form of concerted practice, it is necessary to first establish liability of companies for the decisions taken by their algorithms. In this regard, Calzolari’s reasoning of using strict corporate liability for the actions of their subsidiaries demonstrates is a good illustration that under competition law liability can be assigned to a company even for conduct that it is not aware of. Furthermore, this conclusion is also corroborated by more recent case-law of the CJEU. More specifically, in the case of Scania the General Court has ruled that undertakings can be held liable for a single and continuous infringement caused by actions of their employees on multiple levels and even if said levels were not aware of each other’s role in the infringement. ¹³⁵ As a result, establishing liability of companies for the actions of their algorithms is entirely consistent with the principles of EU competition law.

Following on, the second step towards treating algorithmic tacit collusion as a form of concerted practices is to establish the way in which it harms the competitive process. Here, Calzolari

135 Case T-799/17 Scania and Others v Commission EU:T:2022:48, paragraphs 470-477.

44 provides a convincing argument by stating that when companies adopt pricing algorithms

because they are better at monitoring their competitors and extracting information about the market, they knowingly substitute the risks of competition with knowledge. Furthermore, for the purpose of easier enforcement, this can be coupled with a rebuttable presumption that if the Commission uncovers a case of algorithmic tacit collusion, it is to be deemed illegal unless the undertakings involved in it can justify their conduct. As a result, it is possible to justify the treatment of algorithmic tacit collusion as a form of concerted practice. Moreover, doing so is also desirable since it is likely to deter firms from engaging in it and will force them to focus on making sure that the algorithms which they purchase are compliant with EU competition law.

Additionally, another competition law remedy that has to be deployed is the establishment of producer liability for companies that create algorithms capable of reaching a tacitly collusive outcome. This can be done in line with Beneke and Mackenrodt’s proposal discussed in section 5.2 and will serve to address the source of the problem and provide for stronger deterrence.

However, by itself, prohibiting algorithmic tacit collusion by virtue of the framework of EU competition law is not enough to deter undertakings from participating in it as effectively as possible. This is mainly due to two reasons. First, as exemplified by Harrington’s proposals for an evidentiary method for analysing algorithms may prove to be a complex and challenging task that requires a high amount of expertise and resources. Second, even though something has been categorized as illegal anticompetitive conduct, this does not mean that undertakings will not try to participate in it. Because of this, to achieve a maximum deterrence effect, ex post competition law remedies need to be supplemented by ex ante regulation.

The main ex ante remedy that can used is that proposed by Bertrand and Dewenter of adopting a lag in the pricing activities of algorithms. This is because it is relatively easy to enforce and as they themselves have stated, a similar version of it has already been applied in practice. ¹³⁶ If such a remedy is indeed adopted, it can have a great debilitating effect by striking at one of the core characteristics that give algorithms their ability to outperform humans. However,

constructing such a remedy comes with a major challenge. It is hard to determine how much

136 Bernhardt and Dewenter (n 79) 339 and 340.

45 exactly should the activities of algorithms be slowed down in order to preserve competition and at the same time not discourage innovation. This issue leads us to the final consideration that will be explored in this chapter. Namely, the fact that it is currently hard to determine the exact manner in which many ex ante remedies (such as lags in pricing activities) can be implemented because, as has been shown throughout the first half of this thesis, there is a great deal of

uncertainty of when and how algorithmic tacit collusion will materialize. However, this does not mean that at this point in time ex ante remedies are useless. On the contrary, they are amongst the best tools to deal with this uncertainty. To do this, ex ante remedies have to be used to accumulate as much knowledge on the development of pricing algorithms and their role in the operation of markets as possible. This can be done through the creation of new regulatory agencies that specialize in working with algorithmic technologies or the delegation of additional tasks and resources to currently existing ones. These agencies would then proceed to monitor and report on new developments in pricing algorithms and their corresponding markets and inform the relevant institutions about them. Furthermore, such agencies would also be in the best position to propose new policies that are devised on the basis of their experience and expertise, and which are capable of striking the right balance between protecting the competitive process and not impairing innovation. In this sense through the use of ex ante regulation, the EU and its member-states can create new institutions that will allow them to learn quickly and act in an appropriate and expert manner when the threat of algorithmic tacit collusion materializes.

Because of this, in the current situation, ex ante regulation can prove to be an indispensable element of the overall strategies of public institutions to deal with the challenges presented by algorithmic tacit collusion.

However, even though regular monitoring by specially designated regulatory authorities is an effective way of collecting information, it is not the only one. Another instrument that can be used for monitoring the current state of pricing algorithms and their effects on markets are sectoral inquiries. Currently, such inquiries are one of the main tools that the European Commission uses in order to gather data on important sectors of the economy. As has been shown in chapter 4 it is indeed the 2017 sector inquiry of the Commission into the E-commerce sector that served a key role in highlighting the diffusion of pricing algorithms amongst online retailers. Because of this, going forwards in the future, specialized sectoral inquiries need to be

46 conducted into the use of pricing algorithms in various markets. These inquiries, like the work of the abovementioned monitoring institutions, would be a source of information for policymakers and when combined with them, they would serve to paint the most robust possible picture of the state of development of pricing algorithms at any given point in time.