Reproduce “FinMem: A Performance-Enhanced LLM Trading Agent with Layered Memory and Character Design” on OpenResearch

Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.

Universal app for Apple silicon and Intel. Requires macOS 11 or later.

About this paper

Yangyang Yu, Haohang Li, Zhi Chen, Yuechen Jiang, Yang Li, Denghui Zhang, Rong Liu, Jordan W. Suchow and 1 more

FINMEM, an LLM-based trading agent developed at Stevens Institute of Technology, introduces a layered memory system and dynamic character design to autonomously process multi-source financial information. It consistently outperformed traditional DRL agents and other LLM models in cumulative return and Sharpe Ratio, achieving robust results with significantly less training data than conventional methods.

Key insights

  • LLM-based agents can achieve superior trading performance with significantly less historical training data compared to traditional Deep Reinforcement Learning (DRL) agents when designed with specialized financial intelligence.
  • A layered memory system, emulating human cognition but extending its capacity, is critical for effectively processing and prioritizing time-sensitive, hierarchical financial information from diverse sources.
  • Dynamic, self-adaptive character designs (e.g., risk inclination) enhance an LLM agent's ability to adjust trading strategies in response to real-time market conditions, improving robustness and profitability.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.