Algorithmic Collusion in EU Competition Law: Decoding the Puzzle

This thesis explores the phenomenon of algorithmic collusion, a term used to describe scenarios where pricing algorithms facilitate, enable, or even autonomously engage in explicit or tacit collusive behavior. In recent years, this issue has drawn significant attention within the global antitrust community, as highlighted by the extensive body of research stemming from Ezrachi and Stucke’s seminal work, Virtual Competition (2016). The debate surrounding algorithmic collusion is driven by three main concerns. First, pricing algorithms may indirectly foster human collusion by enhancing market transparency, simplifying coordination, and easing enforcement. Second, algorithms can act as tools for firms to implement collusive strategies, making such behavior more stable and less detectable. Third, and most critically, AI-driven algorithms might independently learn to collude without explicit programming, raising the question of whether firms should bear responsibility for unintended collusive outcomes. Against this backdrop, this research seeks to “decode the puzzle” of algorithmic collusion, focusing on its compatibility with the EU antitrust legal framework, specifically Article 101 TFEU on agreements and concerted practices. The study is structured around three objectives: (1) categorizing algorithmic collusion based on existing literature, (2) assessing the applicability of Article 101 TFEU to each category, and (3) proposing alternative remedies where Article 101 proves insufficient. The findings reveal that when pricing algorithms act as tools to support collusive strategies, Article 101 TFEU should apply, and the use of such algorithms may influence the severity of fines. For autonomous algorithmic collusion, which often mimics human behavior, explicit collusion can be addressed under Article 101 through an expansive interpretation of “meeting of minds”. Even when firms do not directly program algorithms to collude, they can still be held liable based on principles of vicarious liability. However, tacit collusion, whether human or algorithmic, remains outside the scope of Article 101 due to the legal acceptance of outcomes driven by interdependence. This thesis challenges the exclusion of algorithmic tacit collusion from antitrust scrutiny, arguing that algorithms could make tacit collusion more pervasive and effective than human-driven interdependence. As such, the legal treatment of algorithmic tacit collusion warrants reevaluation. To address this, the thesis maps and evaluates existing remedies within and beyond the antitrust framework, while also proposing new approaches. Central to this is the concept of ‘outcome visibility,’ which holds firms accountable for the observable market effects of their algorithms, even when unintentional. By advancing this framework, the thesis contributes to bridging regulatory gaps and ensuring that antitrust law keeps pace with technological advancements.