A new preprint on arXiv examines how market makers can be trained with cooperative multi-agent reinforcement learning. The abstract defines a market maker's core task: quoting prices to buy and sell an asset, earning the spread while keeping inventory risk under control.
The study specifically targets prediction-market order books, where prices reflect probabilities of future events. By using multiple agents that cooperate, the research aims to improve liquidity provision in this setting. The abstract does not yet disclose results or experimental details, as it is cut off after the introduction.
Because only the abstract is available, the article notes the research direction rather than findings. It highlights the intersection of multi-agent RL and market microstructure, suggesting a possible path toward more adaptive automated market making.