Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FinMem: A Performance-Enhanced LLM Trading Agent with Layered Memory and Character Design

Yangyang Yu, Haohang Li, Zhi Chen, Yuechen Jiang, Yang Li, Denghui Zhang, Rong Liu, Jordan W. Suchow, Khaldoun Khashanah

TL;DR

The paper tackles the challenge of processing vast, time-sensitive financial information for automated trading. It introduces FinMem, a memory-augmented LLM agent with Profiling, Memory (Working + Layered Long-Term), and Decision-making modules, enabling adaptive, interpretable trading decisions. Empirical results on real-world data show FinMem outperforms DRL and open-source LLM baselines, with robustness to limited training data and market volatility. The layered memory and dynamic profiling contribute to superior decision quality and demonstrate potential for broader, multi-agent financial systems. This work advances practical, memory-aware AI agents for finance and beyond, enabling faster adaptation to evolving markets.

Abstract

Recent advancements in Large Language Models (LLMs) have exhibited notable efficacy in question-answering (QA) tasks across diverse domains. Their prowess in integrating extensive web knowledge has fueled interest in developing LLM-based autonomous agents. While LLMs are efficient in decoding human instructions and deriving solutions by holistically processing historical inputs, transitioning to purpose-driven agents requires a supplementary rational architecture to process multi-source information, establish reasoning chains, and prioritize critical tasks. Addressing this, we introduce \textsc{FinMem}, a novel LLM-based agent framework devised for financial decision-making. It encompasses three core modules: Profiling, to customize the agent's characteristics; Memory, with layered message processing, to aid the agent in assimilating hierarchical financial data; and Decision-making, to convert insights gained from memories into investment decisions. Notably, \textsc{FinMem}'s memory module aligns closely with the cognitive structure of human traders, offering robust interpretability and real-time tuning. Its adjustable cognitive span allows for the retention of critical information beyond human perceptual limits, thereby enhancing trading outcomes. This framework enables the agent to self-evolve its professional knowledge, react agilely to new investment cues, and continuously refine trading decisions in the volatile financial environment. We first compare \textsc{FinMem} with various algorithmic agents on a scalable real-world financial dataset, underscoring its leading trading performance in stocks. We then fine-tuned the agent's perceptual span and character setting to achieve a significantly enhanced trading performance. Collectively, \textsc{FinMem} presents a cutting-edge LLM agent framework for automated trading, boosting cumulative investment returns.

FinMem: A Performance-Enhanced LLM Trading Agent with Layered Memory and Character Design

TL;DR

The paper tackles the challenge of processing vast, time-sensitive financial information for automated trading. It introduces FinMem, a memory-augmented LLM agent with Profiling, Memory (Working + Layered Long-Term), and Decision-making modules, enabling adaptive, interpretable trading decisions. Empirical results on real-world data show FinMem outperforms DRL and open-source LLM baselines, with robustness to limited training data and market volatility. The layered memory and dynamic profiling contribute to superior decision quality and demonstrate potential for broader, multi-agent financial systems. This work advances practical, memory-aware AI agents for finance and beyond, enabling faster adaptation to evolving markets.

Abstract

Recent advancements in Large Language Models (LLMs) have exhibited notable efficacy in question-answering (QA) tasks across diverse domains. Their prowess in integrating extensive web knowledge has fueled interest in developing LLM-based autonomous agents. While LLMs are efficient in decoding human instructions and deriving solutions by holistically processing historical inputs, transitioning to purpose-driven agents requires a supplementary rational architecture to process multi-source information, establish reasoning chains, and prioritize critical tasks. Addressing this, we introduce \textsc{FinMem}, a novel LLM-based agent framework devised for financial decision-making. It encompasses three core modules: Profiling, to customize the agent's characteristics; Memory, with layered message processing, to aid the agent in assimilating hierarchical financial data; and Decision-making, to convert insights gained from memories into investment decisions. Notably, \textsc{FinMem}'s memory module aligns closely with the cognitive structure of human traders, offering robust interpretability and real-time tuning. Its adjustable cognitive span allows for the retention of critical information beyond human perceptual limits, thereby enhancing trading outcomes. This framework enables the agent to self-evolve its professional knowledge, react agilely to new investment cues, and continuously refine trading decisions in the volatile financial environment. We first compare \textsc{FinMem} with various algorithmic agents on a scalable real-world financial dataset, underscoring its leading trading performance in stocks. We then fine-tuned the agent's perceptual span and character setting to achieve a significantly enhanced trading performance. Collectively, \textsc{FinMem} presents a cutting-edge LLM agent framework for automated trading, boosting cumulative investment returns.
Paper Structure (23 sections, 6 equations, 11 figures, 5 tables)

This paper contains 23 sections, 6 equations, 11 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Backbone Algorithms of Contemporary Autonomous Trading Agents
  4. Advancements from LLMs to LLM Autonomous Agents
  5. Architecture Design of LLM Autonomous Agent
  6. Architecture of FinMem
  7. Profiling Module
  8. Memory Module
  9. Working memory
  10. Layered long-term memory
  11. Decision-making Module
  12. Experiments Setups
  13. Datasets And Database Structure:
  14. Baseline And Comparative models:
  15. Evaluation Metrics:
  16. ...and 8 more sections

Figures (11)

  • Figure 1: The prompt template for FinMem's profiling module. It includes two key elements of its character setting: professional background knowledge and three distinct investment risk inclinations. In the self-adaptive risk inclination option, the omitted texts align with the detailed descriptions provided for the risk-seeking and risk-averse inclinations.
  • Figure 2: Memory module structure of FinMem with a detailed view of components, operations, and workflow. The cognitive architectures of FinMem's memory module have two core components -- Working Memory and Layered Long-term Memory.
  • Figure 3: (1) The decision-making module workflow of the FinMem trading agent retrieves critical memory events to inform specific decisions. (2) LLM prompt template used by FinMem to interact with incoming financial information.
  • Figure 4: The distribution of news in scraped from Alpaca News API for the five stocks in the experiments
  • Figure 5: Cumulative return comparison over time between FinMem and other algorithmic agents across five stocks.
  • ...and 6 more figures