Cooperation between independent market makers

Artificial Collusion: Examining Supracompetitive Pricing by Q-Learning Algorithms

“…An influential contribution to this debate is Calvano et al (2020a), in which a particular pricing algorithm is said to 'autonomously' learn to collude when competing firms adopt it to set their prices. Our analysis focuses on claims made in this paper, but algorithmic collusion has also been examined in various related settings, including sequential-move games (Klein, 2021), sellers on a platform (Johnson, Rhodes and Wildenbeest, 2021;Sánchez-Cartas and Katsamakas, 2022), settings with more advanced reinforcement learning methods (Hettich, 2021;Kastius and Schlosser, 2021;Wang, 2022), dealer markets with multiple market makers (Han, 2022;Xiong and Cont, 2021), and continuous-time models (Cartea et al, 2022). Most of these papers make wider contributions to the literature, yet to the extent that they rely on similar simple Q-learning algorithms our findings may be relevant to them as well.…”

Can maker-taker fees prevent algorithmic cooperation in market making?

“…For the market-making problem, [23] finds collusive prices are possible under decentralized multi-agent reinforcement learning without price information sharing. [14] considers Q-learning with Boltzmann selection and proves the convergence to supra-competitive spread levels when agents have no memory.…”