AI Hedge Fund

AI Hedge Fund: An Educational AI-Powered Trading System

The AI Hedge Fund project is an innovative proof-of-concept (POC) that demonstrates how artificial intelligence, particularly large language models (LLMs), can be leveraged to simulate a hedge fund's decision-making process. Created by developer Virat, this open-source GitHub repository explores the integration of AI agents in financial analysis and trading strategy formulation. Importantly, it is strictly for educational and research purposes, with explicit disclaimers against using it for real-world trading or investment decisions. No actual trades are executed, and users are advised to consult financial professionals for any investment-related activities.

Core Components and Agents

At the heart of the system is a collaborative team of AI agents, each inspired by legendary investors and analysts. These agents work together to evaluate stocks from various perspectives, ensuring a multifaceted approach to investment analysis:

• Investor Agents:

  • Aswath Damodaran Agent: Focuses on disciplined valuation combining narrative, numbers, and fundamentals.
  • Ben Graham Agent: Emphasizes value investing, seeking undervalued 'hidden gems' with a strong margin of safety.
  • Bill Ackman Agent: Adopts an activist stance, identifying opportunities for bold interventions and change.
  • Cathie Wood Agent: Champions growth investing, prioritizing disruptive innovation and long-term potential.
  • Charlie Munger Agent: Seeks high-quality businesses at reasonable prices, drawing from multidisciplinary wisdom.
  • Michael Burry Agent: Hunts for contrarian deep-value plays, inspired by 'The Big Short'.
  • Mohnish Pabrai Agent: Applies Dhandho principles for low-risk, high-reward 'doubles'.
  • Peter Lynch Agent: Looks for 'ten-baggers' in accessible, everyday companies.
  • Phil Fisher Agent: Conducts in-depth 'scuttlebutt' research for meticulous growth opportunities.
  • Rakesh Jhunjhunwala Agent: Represents bold Indian market investing.
  • Stanley Druckenmiller Agent: Targets macro trends and asymmetric growth opportunities.
  • Warren Buffett Agent: The core value investor, buying wonderful companies at fair prices.

• Analysis Agents:

  • Valuation Agent: Computes intrinsic stock values and issues buy/sell signals.
  • Sentiment Agent: Gauges market sentiment from news and social data.
  • Fundamentals Agent: Dives into financial statements and key metrics.
  • Technicals Agent: Applies chart patterns and indicators for timing.

• Management Agents:

  • Risk Manager: Monitors portfolio risks, sets limits, and calculates metrics like VaR.
  • Portfolio Manager: Synthesizes inputs to make final allocation and order decisions.

This multi-agent architecture mimics a real hedge fund team, where diverse viewpoints converge to reduce biases and enhance decision quality. The system processes data for popular tickers like AAPL, MSFT, NVDA, and TSLA (free data), with API support for broader coverage.

Setup and Usage

Installation

To get started:

1. Clone the repo: git clone https://github.com/virattt/ai-hedge-fund.git.
2. Set up a .env file with API keys (e.g., OPENAI_API_KEY for GPT models; FINANCIAL_DATASETS_API_KEY for additional data).
3. Install dependencies via Poetry: poetry install.

Running the System

• Command Line Interface (CLI): Run analysis on tickers with poetry run python src/main.py --ticker AAPL,MSFT,NVDA. Options include --ollama for local LLMs, --start-date, and --end-date for historical periods. A backtester is available: poetry run python src/backtester.py --ticker AAPL,MSFT,NVDA to simulate performance.

• Web Application: For a user-friendly GUI, follow instructions in the /app directory to launch a full-stack interface, ideal for visual exploration of agent outputs and portfolios.

The system outputs detailed reports, including agent rationales, signals, and simulated portfolio metrics, often visualized in screenshots provided in the README.

Key Features and Educational Value

• AI Integration: Powered by LLMs like GPT-4o or local models via Ollama, agents use prompt engineering to emulate investor philosophies.
• Data Sources: Free for major tech stocks; extensible via APIs for fundamentals, sentiment, and technicals.
• Risk and Portfolio Management: Includes position sizing, diversification, and backtesting to teach practical finance concepts.
• Customization: Users can add agents, tweak prompts, or integrate new data sources.

With over 42,000 stars on GitHub, it has gained significant traction in the AI and finance communities. The project encourages contributions through focused pull requests and features requests via issues. Licensed under MIT, it's freely available for learning how AI can transform quantitative finance.

Limitations and Disclaimers

As emphasized, this is not financial advice. Past simulations do not predict future results, and the creator disclaims liability. It's a sandbox for experimenting with AI in investing, highlighting both potentials (e.g., scalable analysis) and pitfalls (e.g., LLM hallucinations in financial contexts).