Algorithmic Collusion in E-commerce: Dynamic Pricing and Antitrust Mechanism Design via Multi-Agent Reinforcement Learning

Abstract

Currently, the increasing prevalence of algorithmic dynamic pricing in the e-commerce ecosystem has blurred the boundaries of antitrust enforcement. Therefore, this study constructs a multi-agent reinforcement learning framework, given the limitations of traditional intention-based legal frameworks in interpreting these autonomous behaviors. Furthermore, this study explores proactive regulatory intervention measures, proposing and evaluating the effectiveness of dynamic penalty fees and introducing local differential privacy to disrupt algorithmic market observations. This discussion elevates the focus from simply identifying the theoretical harms of algorithms to the architectural design of algorithmic governance, demonstrating that the future antitrust paradigm can shift towards continuous, behavior-based structural interventions.