ContestTrade is a multi-agent trading framework designed for event-driven stock selection in the A-share market. It aims to automate the process of identifying, evaluating, and tracking investment opportunities triggered by market events, ultimately providing actionable asset allocation recommendations without human intervention. The system is targeted at quantitative traders and researchers looking to build AI-powered trading teams.

How It Works

ContestTrade employs a two-stage pipeline that simulates a company's decision-making process. Initially, multiple data analysis agents process raw market data, transforming it into structured "text factors." An internal contest mechanism then evaluates these factors, creating an optimal "factor investment portfolio." This portfolio is subsequently analyzed by research agents, each with unique "trading beliefs," who generate trading proposals. A second internal contest refines these proposals into a unified and reliable asset allocation strategy. This dual-contest approach ensures that decisions are driven by robust insights, enhancing adaptability and resilience in complex market conditions.

Quick Start & Requirements

• Installation: Clone the repository, create and activate a Conda environment (conda create -n contesttrade python=3.10, conda activate contesttrade), and install dependencies (pip install -r requirements.txt).
• Configuration: Requires API keys for data sources (Tushare, Bocha, SerpAPI) and LLMs (LLM, LLM_THINKING, VLM) to be configured in config.yaml.
• Dependencies: Python 3.10, Conda, and API keys for various services.
• Usage: Launch via CLI (python -m cli.main run) and interactively select analysis times. Results are saved in Markdown format in contest_trade/agents_workspace/results.

Highlighted Details

• Supports customizable "trading beliefs" for research agents to tailor investment strategies.
• The framework is designed for event-driven opportunities, focusing on catalysts like news, announcements, and policy changes.
• Currently supports the A-share market, with plans to expand to US and Hong Kong markets.

Maintenance & Community

• The project is community-driven, welcoming contributions via its GitHub repository.
• Bug reports and feature suggestions can be submitted through the Issues page.
• A roadmap indicates ongoing development, including V2.0 focusing on market and feature expansion.

Licensing & Compatibility

• Licensed under the Apache 2.0 License.
• The license permits commercial use and linking with closed-source projects.

Limitations & Caveats

• The framework is primarily for academic research and educational purposes; outputs do not constitute investment advice.
• Relies on external API keys and data sources, which may have associated costs or limitations.
• AI models used may be subject to "hallucinations" or inaccuracies, and data sources may have latency or completeness issues.