TwinMarket: A Scalable Behavioral and Social Simulation for Financial Markets

💡 Update

• 09/2025: TwinMarket was accepted to NeurIPS 2025. See you in San Diego! 🌊
• 04/2025: TwinMarket won the Best Paper Award 🏆 at the Advances in Financial AI Workshop @ ICLR 2025.

📖 Overview

TwinMarket is an innovative stock market simulation system powered by Large Language Models (LLMs). It simulates realistic trading environments through multi-agent collaboration, covering personalized trading strategies, social network interactions, and news/information analysis for an end-to-end market simulation.

🎯 Key Features

• 🤖 Intelligent Trading Agents: LLM-driven, personalized decision-making
• 🌐 Social Network Simulation: Forum-style interactions and user relationship graphs
• 📊 Multi-dimensional Analytics: Technical indicators, news, and market sentiment
• 🎲 Behavioral Finance Modeling: Includes disposition effect, lottery preference, and more
• ⚡ High-performance Concurrency: Scalable simulation for large user populations
• 📈 Real-time Matching Engine: Full order matching and execution

🚀 Quick Start

# Configure your API and embedding models
cp config/api_example.yaml config/api.yaml
cp config/embedding_example.yaml config/embedding.yaml

# Run the demo
bash script/run.sh

📝 Development Guide

Extend Trading Strategies

Implement new strategies in trader/trading_agent.py:

def custom_strategy(self, market_data):
    """Custom trading strategy"""
    # Implement your strategy logic here
    pass

Add New Evaluation Metrics

Add metrics in trader/utility.py:

def calculate_custom_metric(trades):
    """Compute custom metric"""
    # Implement metric calculation here
    pass

🧾 Citation

@inproceedings{yang2025twinmarket,
  title     = {TwinMarket: A Scalable Behavioral and Social Simulation for Financial Markets},
  author    = {Yuzhe Yang and Yifei Zhang and Minghao Wu and Kaidi Zhang and
               Yunmiao Zhang and Honghai Yu and Yan Hu and Benyou Wang},
  booktitle = {Proceedings of the 39th Annual Conference on Neural Information Processing Systems (NeurIPS)},
  year      = {2025},
  url       = {https://arxiv.org/abs/2502.01506}
}

Star History