 cbazza
|
ce197c06f4
|
Implement session-specific confidence thresholds (NY Fine-Tuning)
FEATURE: Session-Specific Confidence Thresholds
- Asian: >=95% Confidence (unchanged, 97.8% WR)
- NY: >=97% Confidence (NEW, improves WR from 43.3% to 56.5%!)
- London/Overlap: Blocked (as before)
EXPECTED IMPACT:
- Eliminates 7 poor NY trades (all <97% confidence)
- NY Win-Rate: 43.3% → 56.5% (+13.2 pp)
- NY Profit: $1,418 → $1,655 (+$237)
- Total Profit: $8,306 → $8,598 (+$292)
- Overall Win-Rate: 67.8% → ~71%
IMPLEMENTATION:
1. session_filter_patch.py
- Added session_confidence_thresholds config
- New function: get_session_confidence_threshold()
- New function: is_confidence_sufficient()
2. session_confidence_filter.py (NEW)
- Wrapper for execute_trade_v2_adaptive
- Session-specific confidence checks
- Test suite (6/6 tests passed ✅)
3. analyze_ny_session.py (NEW)
- Detailed NY session analysis
- Simulations for different thresholds
- Data shows 97-98% trades had 100% WR
TESTING:
All 6 test cases passed:
- Asian 96%: ALLOWED ✅
- Asian 94%: BLOCKED ✅
- NY 98%: ALLOWED ✅
- NY 96%: BLOCKED ✅
- London 99%: BLOCKED ✅
- Overlap 99%: BLOCKED ✅
NEXT STEPS:
1. Integrate wrapper into notebook
2. Restart kernel
3. Monitor for 1 week
4. Review performance improvement
FILES:
- session_filter_patch.py: Updated config + new functions
- session_confidence_filter.py: Wrapper implementation
- analyze_ny_session.py: Analysis tool
- NY_SESSION_FINETUNING.md: Complete documentation
|
2025-12-26 17:46:58 +01:00 |
|