I. Introduction
Pricing algorithms are computer models used to predict the optimal price. Said optimal price depends on various inputs, such as: demand and supply conditions, and prices charged by competitors in substitutable and/or complimentary goods. In a City of the Future where most companies rely on this type of algorithms to define prices, such technology may lead to collusive outcomes.
There are four types of possible computer algorithm collusion, each one raising specific challenges: i) Messenger; ii) Hub and Spoke; iii) Predictable Agent and iv) Autonomous Machine.
The Messenger, where the algorithm is directly used as a mean for humans to put into practice an agreed collusive decision – which does not trigger new challenges to Competition Law. The Hub and Spoke, where a single algorithm is used by numerous players in the market (cluster of similar vertical agreements). The Predictable Agent, where algorithms are unilaterally designed to react to changing market conditions and with the awareness of possible outcomes from other players’ algorithms. And, finally, the Autonomous Machine, where the algorithm will achieve a target, like profit maximization, based on strategies created from the everchanging conditions of the market.
This Insight will particularly focus on the Hub and Spoke category.
II. Hub and Spoke
A Hub and Spoke can be defined as “the establishment of horizontal collaboration through the medium of vertical discussions with relevant hub”, in other words, it uses “a vertical input to facilitate horizontal collusion”.
In the case of algorithmic Hub and Spokes, the hub may be the IT company that supplies the algorithm that multiple companies (spokes) use to define their prices.
Even though the European Commission has not yet decided about Hub and Spoke arrangements, it had a preliminary conclusion about the e-books Case, where it was considered that the joint switch to an agency model (instead of a wholesale model) amounted to a concerted practice between five publishers and Apple, since it had the purpose of raising e-books retail prices, within the meaning of Article 101. (1) TFUE . According to the UNITED States District Court Southern District of New York, where the case was also discussed, Apple intentionally joined the conspiracy and it was considered that, without Apple’s collaboration, such price-fixing conspiracy would not have succeeded. Similar were the preliminary conclusions of the European Commission. However, it should be noted the European Commission did not reach a decision, since the case was settled, and the parties involved assumed a series of commitments.
In addition to the above, it should be noted that third-party price-fixing collusion can be achieved by i) code level alignment, where “algorithms with a shared purpose, for example the calculation of prices” and “a similar implemented methodology”; ii) data level alignment (through data polls or possible use of confidential client’s data in algorithm’s calibration) or iii) a combination of both.
III. Liability
In light of the above, who should answer in case of liability is limited to: i) the supplier of the algorithm, ii) the humans who deploy it, iii) or “ no one”. Even though the last option does not seem viable, the other solutions are clear – especially when we are talking about autonomously acting black-box algorithms which can be beyond the control of humans.
Firstly, the following requisites should be analysed to identify if liability is foreseen: i) the context of exchange of information between spokes and hub; ii) the link between the information exchanges and the market conduct of the players; and iii) if the players have distanced themselves publicly.[1]
For the first requisite, we must analyse the level of awareness of the spokes regarding the information exchanges between the other spokes and the hub. However, full knowledge is not necessary – the undertaking can either be aware of the anticompetitive objectives or in a position where they could reasonably foresee those acts and have been prepared to accept the risk entailed – as explained by the ECJ in the Eturas case.[2]
Regarding the third requirement, there is a rebuttable presumption that the undertaking participating in collusive arrangements that have continuously acted in par with its competitor, knew about the information exchanged. In order to rebut that presumption, the undertaking must publicly distance itself from the practice, report it to the competent authorities or present other proofs.
Moreover, the relevant literature on the matter varies on the specifics of accountability: some authors believe that liability should be only possible when i) a reasonable standard of care and foreseeability is breached, considering human limitations;[3] ii) such accountability would also exist if the actions were carried by an employee;[4] iii) the undertaking omitted a necessary intervention after becoming aware of a coordinated behaviour and, in a broader understanding, iv) the software users discover that a pricing algorithm is also used by competitors, because allegedly they become aware of the risk.
Another matter greatly discussed is if a facilitator/intermediary can be liable. As stated in a Note by the European Union “outsourcing of IT services (…) can also create hub-and-spoke situations when competitors adopt the same algorithm or exchange sensitive commercial information through a common provider”. Furthermore, according with the AC-Treuhand case,[5] facilitators can be sanctioned even if they do not operate on the affected market.[6] Therefore, this can apply if the facilitator is an online platform operator or a third-party algorithm.
IV. Possible solutions
Until now, governments have adopted a market-oriented approach which some believe has contributed to the growth of the digital economy – especially taking into account the fact that excessive regulatory interventions could reduce innovation and prove to be unsuitable. In light of the above, in this section, some non-exhaustive suggestions will be presented.
In terms of algorithm supervision, we can follow three approaches: i) ex-ante – companies will have to report the use of certain algorithms, under certain market conditions, which could result in prohibition; ii) ex-post – intervention is triggered when markets seem to operate in concerted way – this might include the need of simulating the behaviour of algorithms to compare the output of such simulation and the market.
The first approach might be harder to implement because of the costs associated, however, both solutions present the problem of defining the adequate level of intervention and defining what is “tendency to collude”; for instance, a bad judgment can similarly harm consumers.
Some advocate for a third option: iii) non-intervention and possibly relying on self-organization of the companies which would commit to good principles to improve their reputation. However such approach may lead to some lacunas that can be exploited by market players.
Other interesting approaches might entail iv) compliance by design – in which regulation would prohibit certain features.[7] Although it may be easier to implement than the previous suggestions, such approach may limit algorithm-based innovation and add an extra burden for supervisory agencies; v) price regulation – such as setting maximum prices, baseline prices and competitive benchmark prices. Nevertheless, this could reduce incentives for high-quality products and may set unfair/wrong prices; vi) algorithm combat – aimed to destabilize the transparency through mixed signaling. This, however, may generate inefficiencies and harm competition.
Finally, it has been proposed vii) introducing policies to make tacit collusion unstable. Such approach would change the structural characteristics of digital markets that most facilitate collusion. At the same time, this approach can be implemented in various ways, such as: through systems of secret discounts or imposition of restrictions on online information (which could harm the consumers); measure like deceleration (which would reduce the speed and frequency that sellers can adjust prices),[8] and time lag (that may lead to prices being decreased immediately and sellers having to respect time lags to raise the prices. This would force companies to take into account the possible reactions of other players and would act as a punishment mechanism for defecting from the supra-competitive price).
IV. Conclusion
In sum, some Competition Law concepts may still appear unclear and insufficient to face the threats of digital economy. Although some authors are of the opinion that redefining concepts like “agreement” and “intent” might not be the answer, policymakers may need to introduce checks and balances into the original pricing algorithm and monitoring functions. On the contrary, others do not believe in the flexibility of the Article 101 TFUE. For instance, David Currie, former CMA Chairman in the UK and former “believer”, faced with the possibility of machine learning algorithms conclude that the best way to maximise long-term profit is through collusion, even if constraints are built to stop such practice, turned him into an advocate for a new framework.
Notwithstanding, it should be considered that competition law may not be the only or the most suitable vehicle for addressing the problem. For instance, the issue may be reframed by considering these problems as a market manipulation or unfair commercial practice. This way, the focus shifts to the analysis of the impact of algorithms on the market and the consumers.
[1] In AC-Treuhand, those requirements were satisfied because the defendant (a consultancy firm), with knowledge, provided administrative services, such as organising cartel meetings and collecting data, with the intent of implementing a cartel.
[2] Similarly in Case C-542/14 VM Remonts.
[3] JANKA, UHSLER, Antitrust 4.0, European Competition Law Review 2018, p. 112, regarding AC-Treuhand.
[4] DOHRN, HUCK, Der Algorithmus als „Kartellgehilfe“?, Der Betrieb 2018, p. 173 in Autorité de la concurrence, Bundeskartellamt, Algorithms and Competition, 2019, p. 58.
[5] Case C-194/14 P, AC-Treuhand v Commission, ECLI:EU:C:2015:717.
[6] Ibid., Paragraph 36.
[7] i.e. being “programmed to not react to most recent changes in prices; (…) ignore price variations”.
[8] This is used in the fuel sector in Western Australia and Austria, where the seller can only change their prices once a day.
