Trading Strategy Enhancements¶

Based on: "We Backtested a Viral Trading Strategy" article insights Date: October 10, 2025 Status: ✅ Phase 1 & 2 Complete

🎯 What We Implemented¶

Phase 1: Enhanced Strategy Features ✅¶

1. ATR (Average True Range) Calculations¶

Added calculate_atr() - Measures stock volatility
Added calculate_atr_percent() - Normalizes ATR as % of price
Purpose: Adapt risk management to each stock's unique volatility

2. ATR-Based Stop Losses (Replaces Fixed %)¶

Before:

stop_loss = current_price * 0.95  # Fixed -5% for all stocks
take_profit = current_price * 1.10  # Fixed +10%

After:

stop_loss = current_price - (3 * atr)  # Adapts to volatility
take_profit = current_price + (3 * atr)  # Symmetric risk/reward

Why This Matters: - Volatile stocks (CABA, biotech) get wider stops → less likely to get stopped out on noise - Stable stocks (NVDA, big cap) get tighter stops → better capital preservation - 3x ATR = article's recommendation for robust risk/reward

3. 200-Day SMA Trend Filter ✨¶

New Entry Requirement:

if current_price > sma_200:
    # Long-term uptrend ✅
else:
    # SKIP - Don't fight long-term downtrend ❌

Impact: - Article showed this was #2 most important feature for ML model - Prevents buying stocks in long-term bear markets - Simple, proven, works across all market conditions

4. Volatility Entry Filter (ATR%)✨¶

New Entry Requirement:

atr_pct = (atr / current_price) * 100
if atr_pct < 5.0:
    # Calm market ✅ - Good time to enter
else:
    # Too volatile ❌ - Wait for stability

Impact: - Article showed ATR% was #1 most important feature (highest importance!) - Avoids entering during chaotic, unpredictable periods - Waits for "calm before the breakout"

Phase 2: Walk-Forward Backtesting ✅¶

New Tool: `backtest_walk_forward.py`¶

The Problem (From Article): - Original strategy claimed 5,394% profit - Turned out to be curve-fitted to historical data - When tested properly → Near-zero profit

The Solution: Walk-Forward Analysis

Traditional Backtest (WRONG):
  Train on ALL data (2020-2025) → Optimize → Test on SAME data
  Result: Overfitted, unrealistic

Walk-Forward (CORRECT):
  Period 1: Train on 2020-2021 → Test on Jan-Mar 2022 ✅
  Period 2: Train on 2020-2022 → Test on Apr-Jun 2022 ✅
  Period 3: Train on 2021-2022 → Test on Jul-Sep 2022 ✅
  ... rolls forward through time
  Result: Tests on UNSEEN future data → Realistic

Key Features: - Configurable train/test windows (default: 12 months train, 3 months test) - Triple-Barrier Method for trade exits (3x ATR profit/stop, 30-day time limit) - Tracks win rate, average return, exit reasons - Prevents overfitting by design

Usage:

# Backtest QQQ from 2020-2025
uv run python3 tools/backtest_walk_forward.py QQQ --start-year 2020 --end-year 2025

# Backtest SPY with custom windows
uv run python3 tools/backtest_walk_forward.py SPY --train-months 6 --test-months 2

📊 Enhanced Momentum Strategy (Example)¶

Entry Criteria (All Must Be True)¶

Criterion	Threshold	Purpose	Source
Price > SMA(200)	Long-term uptrend	Don't fight trend	Article #2 feature
ATR% < 5.0%	Not too volatile	Enter during calm	Article #1 feature
RSI > 60	Strong momentum	Ride the wave	Classic indicator
Price > SMA(20)	Short-term uptrend	Confirm direction	Classic indicator
7-Day Gain > 5%	Recent strength	Hot stock filter	Momentum signal

Exit Rules (ATR-Based)¶

Stop Loss: Entry - (3 × ATR)
Take Profit: Entry + (3 × ATR)
Risk/Reward: Always 1:1 (symmetric)

Example: CABA @ $3.00, ATR = $0.20 - Entry: $3.00 - Stop: $2.40 (-20%, adapts to biotech volatility) - Target: $3.60 (+20%)

Example: NVDA @ $190, ATR = $5.00 - Entry: $190 - Stop: $175 (-7.9%, tighter for stable stock) - Target: $205 (+7.9%)

🆚 Before vs. After Comparison¶

Component	Before	After	Improvement
Stop Loss	Fixed -5%	3× ATR (dynamic)	Adapts to volatility
Take Profit	Fixed +10%	3× ATR (dynamic)	Symmetric risk/reward
Trend Filter	20-day SMA only	200-day SMA required	Avoid long-term downtrends
Volatility Check	None	ATR% < 5%	Enter during calm markets
Backtesting	None	Walk-forward	Prevent overfitting
Win Definition	Hope for +10%	Triple-barrier method	Systematic exits

🚀 How to Use the Enhanced System¶

1. Interactive Testing (Paper Account)¶

uv run python3 tools/strategy_tester.py

Select option 1 (Momentum Strategy) → You'll now see: - 200-SMA filter check ✨ - ATR% volatility filter ✨ - ATR-based stop/target prices ✨ - Risk/reward ratio

2. Walk-Forward Backtest (Validate Strategy)¶

# Test enhanced momentum on QQQ (2020-2025)
uv run python3 tools/backtest_walk_forward.py QQQ

# Test on your favorite stock
uv run python3 tools/backtest_walk_forward.py NVDA --start-year 2022

This shows TRUE out-of-sample performance (not curve-fitted!).

3. Check CABA Trailing Stop¶

Your CABA trailing stop monitor is still running:

# View live log
tail -f ~/caba_trailing_stop.log

# Check status
ps aux | grep trailing_stop_monitor

📈 Expected Improvements¶

Based on the article's findings:

Risk Management¶

Before: One-size-fits-all stops → frequent unnecessary stop-outs
After: ATR-based stops → fewer false stops, better risk/reward

Win Rate¶

Before: No trend filter → bought stocks in downtrends → low win rate
After: 200-SMA filter → only uptrends → higher win rate expected

Drawdowns¶

Before: Entered during volatile spikes → caught in whipsaws
After: ATR% filter → entered during calm → smoother equity curve

Realistic Performance¶

Before: No backtesting → guessing if strategy works
After: Walk-forward testing → KNOW it works (or doesn't!)

🧪 Next Steps (Optional)¶

Phase 3: Machine Learning (If Walk-Forward Shows Promise)¶

If backtests show positive results, we can:

Feature Engineering
Normalize all indicators (MACD/price, RSI/100, etc.)
Create distance metrics (price vs SMA200)
Add MACD histogram, volume indicators
ML Model Training
Use LightGBM (article's choice)
Train on historical walk-forward data
Learn complex patterns human rules miss
Triple-Barrier Labeling
Systematically label "good" vs "bad" trades
Model learns what defines a winning setup

But: Only do this if Phase 2 backtests show the rule-based strategy has an edge!

🎓 Key Lessons from the Article¶

1. Always Walk-Forward Test¶

"Optimizing on full dataset is curve-fitting. It's a lie."

2. Volatility Is King¶

"ATR% was the #1 most important feature. The model learned to be selective."

3. Don't Fight the Trend¶

"200-SMA filter was #2. Never trade against the primary trend."

4. Accept Lower Returns for Lower Risk¶

"376% out-of-sample beats 791% in-sample because it's REAL."

5. Systematic Beats Discretionary¶

"Triple-barrier method removes emotion from exits."

📋 Files Modified/Created¶

Modified:¶

tools/strategy_tester.py - Enhanced with ATR, SMA200, volatility filters

Created:¶

tools/backtest_walk_forward.py - Walk-forward backtesting framework
STRATEGY_ENHANCEMENTS.md - This document
TRAILING_STOP_GUIDE.md - CABA trailing stop documentation (earlier today)

Running:¶

CABA 15% trailing stop monitor (background process)

⚙️ Configuration¶

All enhancements use these defaults (can be customized):

ATR Period: 14 days
ATR Multiplier for Stops: 3x
ATR% Volatility Threshold: 5.0%
Long-term Trend: 200-day SMA
Short-term Trend: 20-day SMA
Walk-Forward Train: 12 months
Walk-Forward Test: 3 months
Triple-Barrier Time Limit: 30 days

Status: ✅ Ready to test! Recommendation: Run walk-forward backtest on QQQ or SPY to validate the enhanced strategy works historically.

# Let's see if it has an edge!
uv run python3 tools/backtest_walk_forward.py QQQ