This survey paper provides a comprehensive examination of the emerging field of large language model (LLM) agents in financial trading. The authors systematically review current research exploring whether LLM agents can outperform professional traders in this highly competitive domain that requires strategic thinking, domain knowledge, and psychological resilience. The paper synthesizes findings from multiple studies to understand the architectures used for LLM trading agents, the types of data inputs they process, and their performance in backtesting scenarios.

The survey identifies key challenges in LLM-based trading research, including data integration issues, the need for reliable performance evaluation frameworks, and the difficulty of capturing complex market dynamics. The authors analyze how different studies have approached these challenges and compare various agent architectures that incorporate LLMs for decision-making in financial markets. The paper concludes by outlining promising future research directions, highlighting both the potential of LLM agents to transform trading practices and the significant hurdles that must be overcome before they can reliably compete with human professionals in real-world trading environments.