project_journal.md

AI Hedge Fund Roadmap – Project Journal

Current Status

Completed: Days 1–17

---

Days 1–7: Foundations

• Python fundamentals
• NumPy
• Pandas
• Data manipulation
• Visualization
• Financial returns basics

---

Days 8–11: First Strategy

Built:

• SMA crossover strategy
• Buy/Hold benchmark
• Position tracking
• Proper use of shift(1)
• Long-only framework

Key lesson:

• Position and signal are different concepts.

---

Days 12–15: Research Layer

Added:

• Sharpe ratio
• Exposure
• Trade statistics
• Transaction costs
• Slippage
• Stop loss
• Take profit
• Position sizing experiments

Findings:

• TP often reduced performance.
• Stop loss improved robustness.
• Random strategy behaved as expected.

---

Day 16: Refactor

Created modular pipeline:

• load_data()
• calculate_indicators()
• generate_signals()
• run_backtest()
• calculate_metrics()
• plot_results()

Key lesson:

• Strategy != System

---

Day 17: Multi-Asset Validation

Assets tested:

• TSLA
• AAPL
• MSFT
• BTC-USD

Findings:

• SMA strategy generally underperformed Buy & Hold.
• BTC showed the strongest Sharpe.
• Correlation:
  • BTC vs equities: low (~0.17)
  • MSFT vs AAPL: high (~0.66)

Regime Analysis

Observed:

• Bullish regimes generally positive.
• Bearish regimes sometimes positive.
• Sideways regimes consistently weak.

Conclusion:

• Strategy appears regime-dependent.
• Sideways markets are the main weakness.

---

Next Focus (Day 18+)

Automatic Regime Classifier

Using:

• SMA100 slope
• Volatility
• Price relative to SMA100

Regimes:

• Bullish
• Sideways
• Bearish

Goals:

1. Validate classifier visually.
2. Compare performance by regime.
3. Test filtering sideways markets.
4. Determine whether edge exists only in specific regimes.

---

Long-Term Roadmap

Upcoming topics:

• Regime filters
• Walk-forward validation
• Parameter search
• Portfolio construction
• Multi-asset backtesting
• Factor research
• ML-based signals (later)
• Quant research infrastructure