Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Large Language Model Agent in Financial Trading: A Survey

Han Ding, Yinheng Li, Junhao Wang, Hang Chen

TL;DR

The paper surveys the growing use of large language model agents in financial trading, categorizing architectures into Trader and Alpha Miner and detailing data inputs from numerical, textual, visual, and simulated sources. It synthesizes findings from 27 papers, highlighting how news sentiment, memory/reflection, debate, and RL-based techniques enable LLMs to generate trading signals or alpha factors. The review reports that backtesting often shows promising returns (15-30% annualized) but notes methodological gaps, such as reliance on closed-source models, short horizons, limited trading costs consideration, and narrow market coverage. The work maps current capabilities and outlines directions for future research, including openness, multimodal data integration, benchmark rigor, and robust evaluation in diverse markets.

Abstract

Trading is a highly competitive task that requires a combination of strategy, knowledge, and psychological fortitude. With the recent success of large language models(LLMs), it is appealing to apply the emerging intelligence of LLM agents in this competitive arena and understanding if they can outperform professional traders. In this survey, we provide a comprehensive review of the current research on using LLMs as agents in financial trading. We summarize the common architecture used in the agent, the data inputs, and the performance of LLM trading agents in backtesting as well as the challenges presented in these research. This survey aims to provide insights into the current state of LLM-based financial trading agents and outline future research directions in this field.

Large Language Model Agent in Financial Trading: A Survey

TL;DR

The paper surveys the growing use of large language model agents in financial trading, categorizing architectures into Trader and Alpha Miner and detailing data inputs from numerical, textual, visual, and simulated sources. It synthesizes findings from 27 papers, highlighting how news sentiment, memory/reflection, debate, and RL-based techniques enable LLMs to generate trading signals or alpha factors. The review reports that backtesting often shows promising returns (15-30% annualized) but notes methodological gaps, such as reliance on closed-source models, short horizons, limited trading costs consideration, and narrow market coverage. The work maps current capabilities and outlines directions for future research, including openness, multimodal data integration, benchmark rigor, and robust evaluation in diverse markets.

Abstract

Trading is a highly competitive task that requires a combination of strategy, knowledge, and psychological fortitude. With the recent success of large language models(LLMs), it is appealing to apply the emerging intelligence of LLM agents in this competitive arena and understanding if they can outperform professional traders. In this survey, we provide a comprehensive review of the current research on using LLMs as agents in financial trading. We summarize the common architecture used in the agent, the data inputs, and the performance of LLM trading agents in backtesting as well as the challenges presented in these research. This survey aims to provide insights into the current state of LLM-based financial trading agents and outline future research directions in this field.
Paper Structure (30 sections, 2 figures, 1 table)

This paper contains 30 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Architecture
  3. LLM as a Trader
  4. News-Driven
  5. Reflection-Driven
  6. Debate-Driven
  7. Reinforcement Learning Driven
  8. LLM as an Alpha Miner
  9. LLM Selection in Agent
  10. Dataset
  11. Numerical Data
  12. Textual Data
  13. Fundamental Data
  14. Financial Reports
  15. Analyst Reports
  16. ...and 15 more sections

Figures (2)

  • Figure 1: Overview of architectures of finance LLM agents.
  • Figure 2: Histogram of base LLM used by Finance Agent (one paper may contain multiple agent)