This paper introduces a multi-agent reinforcement learning framework designed to optimize market making, specifically applied to the Bitcoin cryptocurrency market. The framework employs two agents: a macro-agent that makes high-level decisions on whether to buy, sell, or hold an asset, and a micro-agent that focuses on placing limit orders within the order book to execute trades efficiently. By leveraging reinforcement learning, the study addresses the complexity of market making in dynamic and volatile environments, demonstrating how this approach can adapt to real-time market conditions and improve trading performance.

The research highlights the viability of reinforcement learning for tackling intricate financial problems, with the Bitcoin market serving as a test case due to its high volatility and liquidity. The results indicate that the dual-agent system enhances decision-making processes, leading to more successful trades and optimized order placement. This work contributes to the growing field of algorithmic trading by providing a scalable and adaptive solution that can be extended to other financial markets, showcasing the potential of AI-driven strategies in quantitative finance.