A Deep Reinforcement Learning Approach to Enhancing Liquidity in the U.S. Municipal Bond Market: An Intelligent Agent-based Trading System

Abstract

This paper presents a new approach to improve revenue in the US bond market using deep learning (DRL) as an artificial intelligence-based market. This study addresses the persistent lack of capacity in this critical business by combining advanced machine learning techniques with the specialised knowledge of financial institutions in the city. A comprehensive multi-agent simulation environment is developed, incorporating key market microstructure features and risk management constraints. The DRL agent is trained using historical trading data from 2018 to 2022, sourced from the Municipal Securities Rulemaking Board's EMMA system. Experimental results demonstrate the agent's superior performance compared to benchmark strategies across various market conditions. The DRL agent consistently improves key liquidity metrics, including bid-ask spreads and market depth, while maintaining robust risk-adjusted returns. The study finds that the proposed approach enhances market efficiency and exhibits adaptability during periods of market stress. Potential impacts on municipal finances were discussed, including reducing the cost of borrowing for local governments and improving cost discovery. Although limitations such as activation capabilities and real-world challenges are recognised, research has yielded positive results for using AI in the financial industry. It is an excellent way to develop the urban economy in the future.