The Turtle Investment Framework is an AI-assisted fundamental analysis system for A-shares, Hong Kong, and US stocks. It merges deterministic Python data collection with LLM-driven qualitative analysis and multi-method valuation. Targeting sophisticated investors and researchers, it automates and deepens fundamental analysis via a hybrid architecture for actionable insights.

How It Works

A hybrid architecture combines Python scripts for reliable data acquisition (Tushare, web search) with LLMs for nuanced qualitative assessments. Version v2.0-beta introduces a PDF-first, single-agent approach for qualitative analysis, directly processing annual report PDFs to minimize information silos and enhance cross-validation. Quantitative analysis features a four-factor model with "penetration return rate" calculation and a valuation module supporting multiple methods (DCF, DDM, PE Band, PEG, PS).

Quick Start & Requirements

• Installation: Clone repository (git clone https://github.com/terancejiang/Turtle_investment_framework.git), cd Turtle_investment_framework, then bash init.sh for environment setup and dependency installation.
• Prerequisites: Python >= 3.10, Tushare Pro account/API Token (via .env or environment variable), pdfplumber (built-in).
• Setup: init.sh automates virtual environment creation, dependency installation, and environment verification.
• Documentation: CHANGELOG_V2.md details version updates.

Highlighted Details

• PDF-First Qualitative Analysis: A single LLM agent directly processes annual report PDFs for a 6-dimensional qualitative assessment: business model, moat (Greenwald framework), external environment, management, MD&A, and ownership structure.
• Hybrid Architecture: Seamless integration of deterministic Python data collection (Tushare, web search) with LLM qualitative insights and quantitative analysis.
• Four-Factor Model & Valuation: Detailed four-factor model includes "penetration return rate" calculation and a robust valuation module that auto-classifies companies and applies multiple methods (DCF, DDM, PE Band, PEG, PS).
• Two-Tier Screener: Sophisticated stock screener with rapid Tier 1 batch filtering (~5 seconds) and deep Tier 2 analysis, capable of processing over 5000 stocks.

Maintenance & Community

The project includes CHANGELOG_V2.md for version updates. No explicit community links (Discord, Slack) or contributor information beyond the repository owner are detailed in the README.

Licensing & Compatibility

MIT License, generally permitting commercial use and integration into closed-source projects, subject to license terms.

Limitations & Caveats

The system is in v2.0-beta, indicating ongoing development. It requires a Tushare Pro API token for core data collection. Some advanced features or strategies (e.g., "烟蒂策略") are marked as planned for future implementation.