← Back to Reviews
20260115_004046_micro_momentum_sol_rsi.md
# AI Review: micro_momentum_sol_rsi (1-Minute)
**Date:** Thu Jan 15 00:40:46 CET 2026
**Strategy File:** /root/trade_1m/strategies_rejected/micro_momentum_sol_rsi.py
---
```json
{
"score": 45,
"flags": [
"overfitting",
"cost_sensitive",
"concentration",
"execution"
],
"review": "## Critical Issues\n\n### 1. Severe Overfitting to Historical Data\n\n**Problem**: This strategy shows classic signs of curve-fitting with 144 trades over 5 months (28.8 trades/month) yet achieves an unrealistic 78% win rate and 13.20 Sharpe ratio. The training metrics are dramatically better than validation performance (train: +23.73%, val: +2.6%), indicating the parameters were optimized to historical data.\n\n**Evidence**:\n- RSI(3) with thresholds of 75/25 are suspiciously precise\n- Volume multiple of exactly 1.5x seems fitted\n- Min bar body of 0.15% is oddly specific (not round like 0.1% or 0.2%)\n- ATR z-score threshold of exactly 2.0 combined with 60-bar lookback\n- The strategy has 10+ parameters all working together perfectly in training\n\n**Why this matters at 1-minute**: With thousands of bars available, it's trivial to find parameter combinations that worked historically. Real microstructure edges should be simpler and more robust.\n\n### 2. Cost Sensitivity Failure\n\n**The math doesn't work**:\n- Claims +9.33% after 0.10% round-trip slippage\n- BUT validation shows only +2.6% return\n- With 17 validation trades × 0.10% = -1.7% in costs\n- This leaves ~4.3% of gross edge, meaning net is barely positive\n\n**At realistic 1-minute execution**:\n- 0.05% per side is optimistic for 1-minute timeframes\n- Market impact on momentum breakouts (entering WITH the move) is higher\n- Real slippage likely 0.10-0.15% per side = 0.20-0.30% round-trip\n- This would completely eliminate the validation edge\n\n### 3. Trade Concentration Risk\n\n**Low sample size**: \n- Only 144 trades over 5 months in training\n- Only 17 trades in validation (below minimum threshold of 20)\n- With 1-minute data, this is suspiciously low frequency\n- One bad trade = -0.25% stop loss vs +0.30% take profit (1.2:1 R:R)\n- At 17 trades, variance is massive - results are noise\n\n**Monthly contribution analysis missing**: The strategy reports max month contribution of 29%, but with only ~29 trades/month average, this means results could be dominated by a handful of lucky trades.\n\n### 4. Execution Reality Problems\n\n**Same-bar entry issue**:\n- Strategy enters on RSI crossover of 75/25 on the current bar\n- At 1-minute timeframe, you see the close AFTER the bar completes\n- By then, the momentum impulse may have already reversed\n- The strategy assumes you can act on the RSI extreme simultaneously with its formation\n\n**Momentum entry problem**:\n- Entering WITH momentum (long when RSI>75) means chasing\n- You're buying after price already moved significantly\n- Slippage will be worse because you're hitting the offer when others are too\n- The 78% win rate likely evaporates with +1 bar entry delay\n\n**Volume confirmation lag**:\n- Requires 1.5x average volume on entry bar\n- By the time you confirm volume, the bar is closed\n- Next bar entry faces worse prices\n\n### 5. Parameter Over-Specification\n\n**Non-round parameters suggest fitting**:\n- 0.15% min bar body (why not 0.10% or 0.20%?)\n- 1.5x volume multiple (why not 1.5x is oddly precise)\n- RSI(3) with 75/25 thresholds\n- 60-bar ATR lookback with 20-bar ATR period\n- 30-bar volume lookback\n\nA robust microstructure edge should use rounder, more intuitive thresholds.\n\n## Minor Issues\n\n### 6. Exit Logic Concerns\n\n**RSI mean reversion exit**: Exiting when RSI crosses 50 is reasonable, but combined with 10-bar max hold, 0.30% TP, and 0.25% SL creates four competing exit mechanisms. This complexity usually indicates parameter fitting.\n\n**Short holding period**: 10-bar max hold = 10 minutes. Combined with RSI(3) reversion, actual hold times are likely 3-7 bars. This is legitimately micro-timeframe, but increases cost sensitivity.\n\n### 7. Cooldown Period Inconsistency\n\n**60-bar cooldown** = 1 hour between trades seems excessive for a \"micro momentum\" strategy. This reduces trade frequency significantly and may be fitted to avoid losing periods.\n\n## What Would Make This Strategy Credible\n\n1. **Simpler parameterization**: Use RSI(5) or RSI(10) with round thresholds like 70/30 or 80/20\n2. **Higher trade frequency**: Micro momentum should generate 200+ trades over 5 months\n3. **+1 bar entry delay testing**: Show results entering the bar AFTER signal\n4. **Higher cost assumptions**: Test with 0.20% round-trip minimum\n5. **Remove fitted parameters**: Eliminate volume multiple, specific bar body, ATR filters\n6. **Walk-forward validation**: Show strategy works in subsequent periods without reoptimization\n\n## Bottom Line\n\nThis strategy achieved good training results by fitting multiple parameters to historical 1-minute data. The validation failure (17 trades, +2.6% return) reveals the overfitting. The claimed edge doesn't survive realistic execution costs at 1-minute timeframes.\n\nThe core concept (RSI momentum continuation) isn't inherently flawed, but the implementation is over-engineered and fitted. Real microstructure edges are simpler and more robust to parameter changes.",
"rule_suggestion": "For momentum-based entries WITH the trend direction: Require demonstration that edge survives +1 bar entry delay with at least 50% of training period return remaining, since momentum entries suffer higher slippage when chasing moves. Strategies entering on momentum breakouts must show profitability with minimum 0.15% round-trip execution costs."
}
```