TradingAgents-CN - Chinese Multi-Agent Financial Trading Decision Framework

What is TradingAgents-CN?

TradingAgents-CN is a Chinese financial trading decision-making framework based on a multi-agent model. Developed on top of TauricResearch/TradingAgents, it provides Chinese users with a complete documentation system and localized support. The framework simulates the professional division of labor and collaborative decision-making processes of real trading firms, using multiple specialized AI agents to collaboratively evaluate market conditions. This includes a multi-agent collaborative architecture involving analyst teams, research teams, trader agents, risk management, and management teams. It supports various large-scale language models, such as Alibaba Cloud, Google AI, OpenAI, and Anthropic, and can be intuitively operated through a web interface. TradingAgents-CN supports Chinese financial markets such as A-shares and Hong Kong stocks, and integrates Chinese financial data from Tushare, AkShare, and other sources, providing Chinese users with convenient and intelligent financial trading decision support.

Main functions of TradingAgents-CN

• Multi-agent collaborative architecture
  • Analyst TeamThis includes fundamental analysts, technical analysts, news analysts, and social media analysts, who are responsible for analyzing the market from different perspectives.
  • Research teamComposed of bullish and bearish analysts, they engage in structured debates and provide in-depth market insights.
  • Trader AI AgentBased on all the information, make the final trading decision.
  • Risk ManagementTo ensure the security of transactions, we conduct multi-level risk assessments and management.
  • ManagementCoordinate the work of various teams to ensure the quality and efficiency of decision-making.
• Multiple LLM model support
  • Domestic LLM IntegrationIt fully supports Alibaba Cloud's 1000 Refinements (qwen-turbo, qwen-plus-latest, qwen-max) and Google AI (gemini-2.0-flash, gemini-1.5-pro, gemini-1.5-flash).
  • International LLM supportSupports models such as OpenAI (GPT-4o, GPT-4o-mini, GPT-3.5-turbo) and Anthropic (Claude-3-Opus, Claude-3-Sonnet, Claude-3-Haiku).
• Intuitive operationA modern web interface based on Streamlit, allowing users to access it directly through a browser without needing to operate via the command line.
• Real-time progress displayThe analysis process is visualized, and the progress is displayed in real time, avoiding user anxiety while waiting.
• Intelligent configurationSupports 5 levels of research depth selection, from quick analysis (2-4 minutes) to comprehensive analysis (15-25 minutes).
• Results DisplayThe structured display includes investment recommendations, target prices, confidence levels, and risk assessments.
• Chinese interfaceThe fully localized Chinese user interface and analysis results lower the barrier to entry for Chinese users.
• Comprehensive data integration
  • A-share data supportProvides real-time A-share quotes and historical data via the Tongdaxin API.
  • US stock data supportSupports real-time market data from sources such as FinnHub and Yahoo Finance.
  • News Data IntegrationIt integrates Google News, financial news, and other sources to provide real-time news data.
  • Social data supportSupports sentiment analysis on social media platforms such as Reddit and Twitter.
  • Database supportSupports MongoDB data persistence and Redis high-speed caching, improving data access efficiency.
• Parallel processingParallel analysis of multiple agents improves analysis efficiency.
• Smart cachingMulti-layered caching strategies reduce API call costs.
• Real-time analysisIt supports real-time market data analysis and enables rapid response to market changes.
• Flexible configurationHighly customizable agent behavior and model selection to meet the needs of different users.
• Cost controlReal-time statistics on token usage help users understand costs. Cost estimates are provided for different configurations to help users optimize usage costs.
• Chinese document systemIt provides over 50,000 words of detailed Chinese documentation, covering the complete path from beginner to expert.
• Example TutorialIncludes 100+ practical code examples to help users get started quickly and learn in depth.

Technical Principles of TradingAgents-CN

• Role specialization and division of laborAssign clear and specific roles and objectives to LLM agents, break down complex transaction tasks into smaller, manageable subtasks, and allow each agent to focus on its area of expertise.
• Multi-agent cooperation mechanismThe system employs a multi-agent architecture to simulate the organizational structure and collaborative processes of real-world trading companies. Through information sharing, debate, and feedback, the agents achieve comprehensive analysis of market information and optimize decision-making.
• Model Selection and ApplicationChoose the appropriate LLM model based on the needs of different tasks. For example, use fast thinking models (such as gpt-4o-mini) for data retrieval and information summarization, and use deep thinking models (such as gpt-4o) for complex reasoning and decision support.
• LLM and Natural Language ProcessingLeveraging the powerful natural language processing capabilities of LLM, it performs in-depth understanding and analysis of text data, extracting key information and knowledge. Intelligent agents can engage in natural language dialogue and debate, explaining the decision-making process to users.
• Fusion of structured and unstructured dataThis approach combines structured data (such as financial statements and transaction data) with unstructured data (such as news articles and social media sentiment). Through LLM analysis and processing, it uncovers the correlations and potential value between data points, providing richer information support for trading decisions.
• Data pipeline designThrough efficient data pipeline design, processes such as data acquisition, cleaning, and feature engineering are implemented. For example, after raw data is acquired from the FinnHub API, it undergoes cleaning and feature engineering before being finally incorporated into the agent's knowledge-sharing pool.

The project address for TradingAgents-CN

• Github repositoryhttps://github.com/hsliuping/TradingAgents-CN

Application Scenarios of TradingAgents-CN

• Individual Stock Analysis: Conduct in-depth analysis of individual stock fundamentals, sentiment, news, and technical indicators to formulate buy, sell, or hold trading strategies.
• Portfolio risk assessment: Conduct a comprehensive risk assessment of the portfolio or individual assets, identify potential risk factors, and develop corresponding risk control measures.
• Market risk warningIt monitors market risks in real time, issues early warning signals in a timely manner, and helps investors avoid or reduce investment losses.
• Market Trend AnalysisIt provides market research tools for financial institutions and researchers to conduct in-depth analysis of market trends, industry dynamics, and macroeconomic factors.
• Investment Strategy ResearchThrough debate and collaboration among multiple agents, we explore and research new investment strategies to provide theoretical support for investment decisions.