Baseline Without LLM: Raw GEX Strategy Performance #58
Description
🎯 Objective
Create baseline strategy that trades every negative GEX day without LLM filtering to prove LLM adds value through selective filtering.
📊 Strategy Definition
Raw Negative GEX Strategy
- Entry Trigger: Every day when GEX < 0 (negative gamma regime)
- Direction: Contrarian approach (same as GAMMA_TRAP)
- Exit Rules: Same risk management as LLM strategy (1% stop, 1.5% target)
- Position Size: Same sizing methodology
- No Intelligence: Pure mechanical execution, no pattern recognition
Performance Comparison Framework
- Baseline Strategy: Raw negative GEX entries (every occurrence)
- LLM Strategy: Selective negative GEX entries (only high-confidence patterns)
- Key Metrics: Win rate, expected value, Sharpe ratio, drawdown
📋 Implementation Requirements
Baseline Strategy Components
```python
class BaselineGEXStrategy:
def generate_signals(self, gex_data):
# Simple rule: enter contrarian when GEX < 0
signals = []
for date, gex_value in gex_data.items():
if gex_value < 0:
signals.append({
'date': date,
'direction': 'contrarian', # Buy if usually bearish
'confidence': 1.0, # No filtering
'reason': 'negative_gex_regime'
})
return signals
```
Performance Analysis
- Sample Size: Compare same date range as LLM strategy
- Statistical Tests: Validate if LLM filtering improves performance
- Risk Metrics: Ensure fair comparison with same risk parameters
🎯 Success Criteria
- Prove LLM Value: LLM-filtered strategy outperforms raw baseline
- Statistical Significance: Measurable improvement in key metrics
- Risk-Adjusted Performance: Better Sharpe ratio, lower drawdown
- Documentation: Clear evidence that intelligence adds value over mechanical rules
🔗 Dependencies
- GEX historical data for same periods as LLM testing
- Risk management framework (stops, targets, sizing)
- Performance comparison utilities
📈 Expected Outcome
Demonstrate that LLM pattern recognition and selective entry filtering provides measurable improvement over naive "trade every negative GEX day" approach.
This baseline proves the intelligence layer adds value beyond simple GEX regime detection.