An LLM-Based Multi-Agent System for Financial Trading Simulation and Strategy Backtesting
Project Description
Financial trading requires synthesizing data from diverse sources—news, financial reports, technical indicators, and macroeconomic trends. Traditional rule-based systems struggle with such heterogeneity. Large Language Models (LLMs) have shown promise in understanding unstructured data and performing complex reasoning. This project aims to explore how multiple LLM agents can collaborate to simulate a more adaptive and human-like trading decision system.
Supervisor
GUO, Song
Co-Supervisor
ZHANG, Jie
Quota
2
Course type
UROP1000
UROP1100
Applicant's Roles
Develop a multi-agent framework that supports Financial Trading Simulation and Strategy Backtesting. Required knowledge Python (pandas, LLM APIs), Machine Learning.
Applicant's Learning Objectives
- Develop a multi-agent framework based on LLMs for simulating financial trading decisions.
- Explore how dividing different roles among agents (e.g., sentiment analysis, technical analysis, risk management) affects decision quality.
- Evaluate the effectiveness of this collaborative system against traditional strategies through backtesting.
- Explore how dividing different roles among agents (e.g., sentiment analysis, technical analysis, risk management) affects decision quality.
- Evaluate the effectiveness of this collaborative system against traditional strategies through backtesting.
Complexity of the project
Moderate