About | HeinOnline Law Journal Library | HeinOnline Law Journal Library | HeinOnline

LOG IN
LOG IN


6 Stan. Computational Antitrust 1 (2026)

handle is hein.journals/stfdcmp6 and id is 1 raw text is: ARTICLE Smart Agent-Based Modelling with LLMs: Leveraging Large Language Models for a Better Understanding of Algorithmic Collusion Carlos Eduardo Veras Neves*& Tanise Brandao Bussmann** Abstract. This paper introduces a Smart Agent-Based Modelling (SABM) within computational antitrust framework to simulate and detect conditions fostering algorithmic collusion. Using SABM, we document how Large Language Model (LLM)-driven agents achieve tacit collusion in a Bertrand duopoly, stabilizing prices above competitive levels without being explicitly instructed to do so. Simulations in English and Portuguese reveal that linguistic context influences outcomes, and communication between agents potentializes emergent behaviors, such as mimicking concerns about collusion. These findings highlight SABM's potential to enhance regulatory oversight, offering an accessible tool for antitrust authorities to help address autonomous algorithmic collusion of pricing agents in digital markets. KEYWORDS: Algorithmic collusion; Computational antitrust; Smart agent-based modelling; Large Language Models; Tacit collusion; Bertrand duopoly JEL: C63, C73, D43, K21, L13, 033 * PhD, Cerebro Project, CADE, Brasil, carlos.neves@cade.gov.br. The authors declare no conflict of interest. This research received no institutional funding. Maritaca AI provided a limited number of API credits to support the computational experiments; no financial compensation was involved. * * PhD, Cerebro Project, CADE, Brasil