DeepFund: AI-Driven Buffett-Style Quantitative Investment Analysis & Optimization

A comprehensive full-stack solution integrating Warren Buffett’s timeless investment principles with cutting-edge quantitative techniques and AI-driven market sentiment analysis — delivering optimized portfolios for mid- to long-term value investing in U.S. equities.

💌 DeepFund: Join Our Free Stock Insights Newsletter

Get Buffett-style stock picks, AI-driven market news, and optimized portfolios—delivered automatically.

🔧 Key Features

• DCF intrinsic value evaluation based on 10,000 Monte Carlo simulations per ticker
• AI-powered macroeconomic and news sentiment analysis using Gemini 2.5 Flash
• Vue.js frontend for DeepFund that shows top 10 tickers, live market indices, and a subscription banner
• Flask API backend running on Railway
• Financial data crawling using FinancialModelingPrep API and yFinance
• Original algorithm that implements Buffett-style scoring of value investing (B-Score)
• Momentum scoring based on technical indicators (RSI, MACD, moving averages - although Buffett himself didn't really care about these!)
• Portfolio optimization using Modern Portfolio Theory (Sharpe, Sortino, CVaR, variance)
• Queries Marketaux API for the latest news
• Results saved to Excel and automatically emailed to subscribers

---

📁 Directory Structure

. (root)
├── README.md
├── backend
│   ├── deep_fund.xlsx
│   ├── dist
│   ├── recipients.json # Auto-updated subscriber email list
│   ├── requirements.txt
│   ├── server.py # Flask backend for managing email subscriptions
│   └── yf_cache_multi.csv
├── frontend
│   ├── README.md
│   ├── dist
│   ├── index.html
│   ├── node_modules
│   ├── package-lock.json
│   ├── package.json
│   ├── public
│   ├── src
│   │    ├── App.vue # frontend web for displaying top tickers of the month and receiving subscription requests
│   └── vite.config.js
└── src
    ├── __pycache__
    ├── backtest_buffett_us.py
    ├── buffett_krx.py
    ├── buffett_us.py # Main analysis script for the US market
    ├── buffett_us_old.py
    └── yf_cache_downloader.py

---

1. Algorithm Logic

① Valuation Metrics

• Indicators: PBR, PER, DCF, etc.
• Considers relative undervaluation against industry averages
• Some metrics are normalized into rank scores for evaluation

② Financial Health Assessment

• Indicators: D/E, CR, ICR, ROE, ROA, EPS growth, dividend growth, operating margin, etc.

• Scoring criteria defined in the buffett_score() function

• Example:

  • D/E ≤ 0.5 → +1 point
  • EPS < 0 → deduction

③ Momentum Scoring

• Points accumulated when technical conditions met:

  • 20-day MA > 60-day MA (medium-term uptrend)
  • 50-day MA > 200-day MA (long-term breakout)
  • 20-day/60-day returns ≥ 10%
  • RSI rebound confirmation
  • MACD golden cross

④ AI-Powered Sentiment & Macroeconomic Summary

• Uses Gemini 2.5 Flash Google Search to:

  • Summarize recent macroeconomic trends
  • Interpret news articles with sentiment scoring
  • Generate stock picks based on quantitative + narrative alignment

---

2. Scoring and Stock Selection

• B-Score = Valuation score + Financial health score
• Total score = B-Score + Momentum score
• Select top X stocks by total score to build the portfolio

---

3. DCF Intrinsic Value Calculation

• Uses past 5 years of Free Cash Flow (FCF) data
• Intrinsic value calculated via DCF model based on FCF growth and discount rate
• Assesses undervaluation compared to current market price

---

4. Portfolio Optimization

• Calculates portfolio weights based on top stocks under four optimization criteria:

Criterion	Objective Function
Sharpe	Maximize annual return / volatility
Sortino	Maximize annual return / downside volatility
CVaR	Minimize average loss in worst 5% scenarios
Variance	Minimize overall return variance

• Results saved in separate Excel sheets (e.g., Weights_CVaR, Weights_Sharpe, etc.)

---

📤 Automation & Deployment

✅ GitHub Actions

• Configured in .github/workflows/buffett.yml to run every Mondays and Fridays at 23:00 UTC
• Auto-downloads prices, computes scores, optimizes weights, and sends results via email

✅ Railway Deployment

• Flask backend + Vue frontend deployed via Railway
• Public-facing email subscription system with auto-update to recipients.json
• Auto commits/pushes email of a new subscriber to GitHub using REST API

---

Disclaimer

• This analysis and all provided information are for reference purposes only.
• The project owner and affiliates disclaim all responsibility for any investment losses or damages.

---

📜 License

MIT License (c) 2025 Hyungsuk Choi (chs_3411@naver.com)