GifariKemalandClaude Opus 4.5
7af9183af3
- XGBoost ML model with 37 features for market direction prediction - Smart Money Concepts (SMC): Order Blocks, FVG, BOS, CHoCH - HMM market regime detection (trending/ranging/volatile) - ATR-based stop loss with 1.5 ATR minimum distance - Broker-level SL protection with fallback - Time-based exit (max 6 hours per trade) - Session-aware trading optimized for London/NY overlap - Auto-retraining based on market conditions - Telegram notifications and web dashboard - Backtest results: 63.9% win rate, 2.64 profit factor, 4.83 Sharpe Backtest period: Jan 2025 - Feb 2026, 654 trades, $4,189 net P/L Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
80 lines
2.2 KiB
Markdown
80 lines
2.2 KiB
Markdown
# Backtest Tuning Report
|
|
**Date**: February 6, 2026
|
|
**Period**: January 2, 2025 - February 4, 2026
|
|
**Data**: 25,807 bars (M15 timeframe)
|
|
|
|
## Threshold Tuning Results
|
|
|
|
| ML Threshold | Total Trades | Win Rate | Net P/L | Profit Factor |
|
|
|--------------|--------------|----------|---------|---------------|
|
|
| **50%** | **485** | **61.6%** | **$3,120.55** | **2.02** |
|
|
| 52% | 463 | 59.0% | $1,868.08 | 1.55 |
|
|
| 55% | 306 | 59.5% | $1,443.56 | 1.74 |
|
|
|
|
## Optimal Configuration
|
|
|
|
```python
|
|
ML_THRESHOLD = 0.50 # Optimal from tuning
|
|
SIGNAL_CONFIRMATION = 2 # Consecutive signals
|
|
PULLBACK_FILTER = True # Enabled
|
|
TRADE_COOLDOWN = 300s # 5 minutes
|
|
```
|
|
|
|
## Performance Metrics (50% Threshold)
|
|
|
|
### Overall
|
|
- **Total Trades**: 485
|
|
- **Wins**: 299 (61.6%)
|
|
- **Losses**: 186 (38.4%)
|
|
- **Net P/L**: $3,120.55
|
|
- **Profit Factor**: 2.02
|
|
|
|
### Risk Metrics
|
|
- **Max Drawdown**: 2.4% ($163.74)
|
|
- **Avg Win**: $20.69
|
|
- **Avg Loss**: $16.48
|
|
- **Expectancy**: $6.43 per trade
|
|
- **Sharpe Ratio**: 3.69 (Excellent)
|
|
|
|
### Exit Reasons
|
|
| Reason | Count | Percentage |
|
|
|--------|-------|------------|
|
|
| Take Profit | 271 | 55.9% |
|
|
| Trend Reversal | 181 | 37.3% |
|
|
| Timeout | 31 | 6.4% |
|
|
| Max Loss | 2 | 0.4% |
|
|
|
|
### Session Performance
|
|
| Session | Trades | Win Rate | Net P/L |
|
|
|---------|--------|----------|---------|
|
|
| **Golden Time (London-NY)** | 103 | **68.0%** | $1,012.72 |
|
|
| Tokyo-London Overlap | 25 | **72.0%** | $248.24 |
|
|
| Sydney-Tokyo | 215 | 61.4% | $1,233.13 |
|
|
| NY Session | 73 | 53.4% | $398.98 |
|
|
| London Early | 69 | 58.0% | $227.47 |
|
|
|
|
## Key Findings
|
|
|
|
1. **Lower threshold = Better performance**: 50% threshold outperforms 55% significantly
|
|
- 58% more trades (485 vs 306)
|
|
- 2.1% higher win rate (61.6% vs 59.5%)
|
|
- 116% more profit ($3,120 vs $1,443)
|
|
|
|
2. **Golden Time is still best**: 68% WR with significant profits
|
|
|
|
3. **Smart exit is effective**:
|
|
- 55.9% take profit (good!)
|
|
- Only 0.4% max loss exits (risk well managed)
|
|
|
|
4. **Excellent risk-adjusted returns**:
|
|
- Sharpe Ratio 3.69 (>2 is excellent)
|
|
- Max drawdown only 2.4%
|
|
|
|
## Recommendation
|
|
|
|
Update main_live.py with:
|
|
- ML Threshold: 50% (changed from 55%)
|
|
- Keep other filters (pullback, confirmation, session)
|
|
|
|
**Expected monthly profit**: ~$240 (based on 13-month backtest)
|