Obstacles to algorithm assessments
Authority's previous position on algorithms
When there is a possibility of serious and irreparable damages while the adoption of a final decision on a case is pending, the Turkish Competition Board (the Board) is entitled to apply interim measures(1) to preserve the context that existed before the violation took place and to prevent irreparable damage to competition in the relevant market until the Turkish Competition Authority (the Authority) has concluded its investigation.
On 30 September 2021, the Authority announced its decision to issue interim measures against DSM Grup Danışmanlık İletişim ve Satış Ticaret AŞ (Trendyol) for its practices in the multi-category online marketplaces market. This is the second interim measure decision in 2021 that the Authority has issued regarding the digital markets sector. Whether it is a coincidence that the Board utilised interim measures in digital markets consecutively, or whether it utilised interim measures as a tool to keep up with the rapid developments in digital markets, will become more apparent in the near future.
Nonetheless, the Trendyol decision is highly remarkable as it might become a milestone in terms of determining the Board's approach going forward, given that it was the first instance in which the Board decided to impose interim measures in an investigation conducted on algorithm-based competition law violations.
Following the Board's decision, a preliminary investigation was initiated against Trendyol on whether it had violated article 4 (Agreements, Concerted Practices and Decisions Limiting Competition) and article 6 (Abuse of Dominant Position) of Law 4054 on the Protection of Competition (Law 4054).
As a result of an on-site inspection and an analysis that was conducted during the preliminary investigation phase based on the data in the algorithms and information systems, the Authority found that Trendyol had:
- acted as the intermediary for the third-party sellers, and conducted the sales of its own brands, such as TrendyolMilla, TrendyolMan and TrendyolKids;
- interfered with the listing algorithm in a way that gave its own products an unfair advantage;
- used data on marketplace activities in the creation of the marketing strategy of its own brands; and,
- discriminated between sellers in the marketplace by interfering with the algorithms.
In light of the above, and considering that Trendyol had gained a significant market share in recent years in all categories within the market for multi-category marketplaces (particularly fashion), the Board decided to apply interim measures in the context of article 9 of Law 4054, as such violations had the potential to cause serious and irreparable damages before the final decision is rendered.
Within this scope, the Board decided that Trendyol shall:
- end all kinds of actions, behaviour and practices that provide an advantage against its competitors, including the interventions made through algorithms and coding, for other products and services within the context of the marketplace activity. Further, it shall avoid such actions during the investigation;
- stop sharing and using any data that was obtained and produced from marketplace activities for other products and services under its economic unity, and avoid such actions during the investigation;
- end all kinds of actions, behaviour and practices that may discriminate among sellers in the marketplace, including interventions made through algorithms and coding, and avoid such actions during the investigation;
- take all necessary technical, administrative and organisational measures to ensure the auditability of the above interim measures;
- retain data on the parametric and structural changes that were made on all algorithm models used for product searches, seller listing and seller score calculation for at least eight years, with all versions and irrefutable accuracy as regards Trendyol;
- retain the source codes of all software that was specifically developed for use within Trendyol for at least eight years, with all versions and irrefutable accuracy; and
- retain user access and authorisation records and manager audit records for all software used within the scope of the business activities being conducted within Trendyol for at least eight years, with irrefutable accuracy.
Obstacles to algorithm assessments
Algorithms, like most technology-based mechanisms, are used in a variety of ways, some of which are more advanced and complex than others. In many cases, it is not fully possible for an outsider to understand how the mathematical processes of algorithms work or what role undertakings play in driving their algorithms towards a particular pricing strategy, especially when it comes to artificial intelligence.(2)
While investigating a certain violation, the burden of proof must be satisfied in order to turn an allegation into a fact. In principle, the burden of proof lies with the authority that is conducting the investigation. Therefore, in cases where the Authority alleges that a violation was made through algorithms, it must first analyse:
- the algorithm's object, implementation and changes over time;
- the undertaking's responsibility from the algorithmic behaviour;
- the scope of a suspected violation; and
- the intent or negligence of the undertaking.
The Authority can also analyse the information from the input data that was used by the algorithm when assessing whether there is a restriction by object.(3)
Further, information on the output of the algorithm and the decision-making process connected with it could be useful in detecting collusions through a pricing algorithm, by assessing whether a potential infringement can be attributed to an undertaking and determining whether the algorithmic behaviour was intended or foreseeable by such undertaking. Therefore, linking these algorithmic processes to illegal behaviour or holding undertakings accountable for using algorithms in a way that restricts competition is not always an easy task.(4)
Even if the steps mentioned above are taken, it is debatable whether there is an actual theory of harm when it comes to undertakings' use of their algorithms.(5) In cases where the algorithm is independently and autonomously learning from itself or cooperating with other algorithms and adjusting itself accordingly, which results in profit maximisations through price coordination, is it really possible for an authority to prove such action is attributable to the undertaking at hand?(6)
Assessing liability for algorithmic actions for undertakings can result in two outcomes:
- holding an undertaking liable for anticompetitive conduct through developing/using an algorithm that takes actions which lead to anticompetitive results; or
- holding an undertaking liable for not complying with reasonable care and foreseeability regarding this conduct.(7)
Authority's previous position on algorithms
Before the Trendyol investigation, the Authority had not inspected algorithmic commercial behaviours. Therefore, such examination constitutes a milestone for on-site investigations, as the Authority has analysed the algorithms of an undertaking in detail for the first time. However, the Trendyol investigation is not the first time the Board has faced algorithms as a tool for infringement.
From 2015 to 2020, the Authority started investigating online platforms with dominant positions in the market such as Yemeksepeti(8) and Booking.com.(9) Even though the Authority dealt with online platforms in the digital sector in its earlier decisions, it abstained from examining the algorithms that the platforms used. In the 2017 Booking.com decision, the Board evaluated Booking.com's "best-price" guarantee practices by examining the "most favoured customer" clause in Booking.com's agreements and assessed whether its effects on the market were anticompetitive.
Algorithm-related allegations were also assessed in the Board's Google AdWords decision(10) in 2020, where it evidently became more interested in digital platform algorithms by investigating whether Google had violated Law 4054 by making changes to its algorithm. The investigation in this case covered algorithm-related allegations, as well as others.
Nonetheless, the Board first concluded that:
based on the findings reached within the scope of the case at hand, it is not possible to come to a conclusion that Google causes a violation of competition through changing the algorithms and giving incomplete information regarding these changes.
Secondly, it concluded that
at this stage, no determination was made that would require intervention as per Law No. 4054, within the scope of the allegations that Google changed the algorithm with the intention of deliberately excluding organic search results from the market, and the allegations that the text advertising of the websites affected their ranking in the organic results.
In addition, according to the 2020 Organisation for Economic Co-operation and Development (OECD) policy notes, the Authority empowered its Strategy Development Department to become more familiar with digital developments in the economy in order to monitor the effects of algorithms on both consumers and markets.(11) Further, the OECD notes specify that:
It is stated by TCA that in recent years, there have been significant developments in the digital economy both in national and international level, which requires competition authorities to closely monitor the effects of multi-sided platforms and the use of algorithms on both consumers and markets.
In April 2021, the Authority published its preliminary report on the e-marketplace platforms sector. In the report, the Authority highlighted its concerns on algorithms and their effect on the marketplace, and signalled that it would dive deeper into the world of algorithms.
With all their different types, levels of development and outcomes, algorithms may be too complex to tackle at once. It is an ongoing debate among competition law experts as to whether competition enforcers have the necessary tools to address concerns in digital markets(12) or whether new measures are needed for certain types of misconduct.(13)(14) Either way, interim measures are among the tools that certain scholars recommend(15) (either after the legal bar on their use is lowered or as they are regulated already in competition law policies).
For now, it seems that the Authority has become aware of algorithms and codes that may lead to competition infringements and it is not hesitant to investigate technical aspects of algorithms more thoroughly to assess such infringement. Nonetheless, whether interim measures are considered by the Board as an effective tool against rapidly evolving digital markets is expected to become more apparent in the near future.
For further information on this topic please contact Gönenç Gürkaynak, Berfu Akgün or Can Baran Beder at ELIG Gürkaynak Attorneys-at-Law by telephone (+90 212 327 17 24) or email ([email protected], [email protected] or [email protected]). The ELIG Gürkaynak Attorneys-at-Law website can be accessed at www.elig.com.
(1) Article 9 of Law 4054 on the Protection of Competition.
(2) Gönenç Gürkaynak, Burcu Can and Sinem Ugur, "Algorithmic Collusion: Fear of the Unknown or too Smart to Catch?", 1 November 2020. THE EVOLUTION OF ANTITRUST IN THE DIGITAL ERA: Essays on Competition Policy, Volume 1, Competition Policy International, November 2020, Available at SSRN.
(4) Avigdor Gal, It's a Feature, not a Bug: On Learning Algorithms and What They Teach Us, Roundtable on Algorithms and Collusion, 21-23 June, 2017, DAF/COMP/WD(2017)50.
(5) Salil Mehra, Antitrust and the Robo-seller; Competition in the Time of Algorithms, 100 MINN L REV 1323 – 1375 (2016).
(6) Crandall, J W, Oudah, M, Tennom et al, Cooperating with machines. Nat Commun 9, 233 (2018).
(7) Autorité de la concurrence & Bundeskartellamt, Working Paper – Algorithms and Competition, November 2019.
(8) Yemeksepeti (28 January 2021, 21-05/64-28).
(9) Booking.com (5 January 2017, 17-01/12-4).
(10) Google AdWords (12 November 2020, 20-49/675-295).
(11) OECD's Consumer data rights and competition – Note by Turkey dated 25 May 2020 and numbered DAF/COMP/WD(2020)55.
(12) Michal S Gal & Niva Elkin-Koren, Algorithmic Consumers, 30 Harv J L & Tech 38 (2017), at 38.
(13) Salil K Mehra, De-Humanizing Antitrust: The Rise of the Machines and the Regulation of Competition (21 August 2014), Temple University Legal Studies Research Paper No. 2014-43, at 2.
(14) Gönenç Gürkaynak, Burcu Can and Sinem Ugur, "Algorithmic Collusion: Fear of the Unknown or too Smart to Catch?", 1 November 2020. THE EVOLUTION OF ANTITRUST IN THE DIGITAL ERA: Essays on Competition Policy, Volume 1, Competition Policy International, November 2020, Available at SSRN.
(15) "A new competition framework for the digital economy", report by the Commission "Competition Law 4.0", (September 2019), Federal Ministry for Economic Affairs and Energy (BMWi), at 7 & 71. Available here.
This article was first published in Legal Insights Quarterly.