← Back to Reviews
20260115_020246_micro_mr_eth_rsi_vwap.md
# AI Review: micro_mr_eth_rsi_vwap (1-Minute)
**Date:** Thu Jan 15 02:02:46 CET 2026
**Strategy File:** /root/trade_1m/strategies_rejected/micro_mr_eth_rsi_vwap.py
---
```json
{
"score": 72,
"flags": [
"cost_sensitive",
"execution",
"inconsistent"
],
"review": "## Microstructure Mean Reversion Strategy Review\n\n### Strengths\n\n**Clear Microstructure Edge**: The strategy targets a legitimate 1-minute friction pattern - RSI extremes with VWAP deviation. The \"bounce confirmation\" approach (waiting for RSI to cross back from <5 to >15) is intelligent and addresses the \"falling knife\" problem that plagues naive mean reversion.\n\n**Round, Sensible Parameters**: RSI(5), 60-min VWAP, 20-period ATR, and thresholds like 5/95/15/85 are all round numbers that suggest principle-based selection rather than optimization artifacts.\n\n**Cost-Aware Design**: The 2-hour cooldown (120 bars) explicitly targets quality over quantity. The 0.35% TP vs 0.25% SL provides asymmetric reward:risk (1.4:1) which is appropriate for mean reversion at this timeframe.\n\n**Volatility Context**: ATR z-score filter (<1.5 for entry, >2.5 emergency exit) shows awareness that microstructure edges collapse in chaos.\n\n### Critical Concerns\n\n**1. Execution Reality Gap (Major)**\n\nThe strategy uses next-bar execution after detecting signal conditions. However, the logic tracks RSI extremes and bounce confirmations within the same processing loop:\n\n- When `current_rsi < 5`, it marks `recent_oversold = True`\n- The entry trigger checks if `prev_rsi < 15 and current_rsi >= 15`\n- This means the signal bar has already seen the bounce\n\nAt 1-minute timeframes, this creates **material slippage risk**. If RSI crosses from 4 → 17 in a single bar (common in volatile micro moves), your \"next bar open\" entry is already late to the reversion. The optimal entry would be intra-bar as RSI crosses 15, which you cannot achieve.\n\n**Impact**: Real execution will be materially worse than backtest. The 0.06% avg trade barely covers 0.05% slippage, leaving almost no margin for this timing gap.\n\n**2. Validation Failure (Critical)**\n\nValidation shows **9 trades vs 91 training trades** - an 89% reduction in opportunity. This is a massive red flag for:\n\n- **Regime dependency**: The extreme RSI<5 / >95 conditions may have been prevalent in training months but rare in validation\n- **Market microstructure shift**: Volatility patterns or market-making behavior may have changed\n\nThe validation return of 0.61% on 9 trades means:\n- Each trade needs to survive 0.10% round-trip costs\n- Only 0.51% left after costs = 5.7% of gross return eaten\n- No statistical significance with n=9\n\n**3. Cost Sensitivity**\n\nWith 91 training trades over 5 months (~18/month) and 0.06% avg trade:\n- Gross edge per trade: 0.06%\n- Slippage buffer: 0.06% - 0.05% = **0.01% cushion**\n- This is a **17% margin** - extremely tight\n\nAny real-world execution variance (wider spreads during RSI extremes, partial fills, delayed detection) will consume this edge entirely.\n\n**4. Setup Window Rigidity**\n\nThe 15-bar setup window for bounce confirmation is clever but creates edge cases:\n- If RSI hits <5 at bar 100, you only accept bounces until bar 115\n- If the bounce happens at bar 116, you miss it despite valid setup\n- This arbitrary cutoff may filter out legitimate signals\n\n### Minor Issues\n\n**Indicator Pre-Computation**: The strategy pre-computes RSI/ATR arrays but doesn't update them incrementally. This means on bar 1000, it recalculates all 1000 bars of RSI. Not a logical flaw, but inefficient.\n\n**VWAP Deviation Storage**: When marking extreme states, it stores `oversold_vwap_dev` but doesn't validate this is the *worst* deviation if multiple extremes occur within the window.\n\n### Month-over-Month Consistency\n\nWhile training shows 5/5 profitable months (excellent), the validation collapse suggests the backtest period may have had unusually favorable microstructure conditions. The strategy needs to prove it works across:\n- Low and high volatility regimes \n- Trending and ranging markets\n- Different liquidity conditions\n\nWith only 9 validation trades, we cannot assess this.\n\n### Verdict\n\nThis is a **conceptually sound microstructure strategy** with legitimate edge logic, but it suffers from:\n1. **Execution timing gaps** that will degrade real performance\n2. **Razor-thin cost margins** that leave no room for implementation variance \n3. **Severe validation failure** indicating regime brittleness\n\nThe 0.06% avg trade is not sufficient for 1-minute execution reality. A robust 1-minute mean reversion strategy should target ≥0.10% avg trade to provide meaningful cost buffer.\n\n**Score: 72/100** - Good concept, acceptable training results, but execution concerns and validation failure prevent higher rating. Not ready for live deployment without addressing timing gaps and cost sensitivity.",
"rule_suggestion": "For mean reversion strategies using RSI extremes: require average trade P&L ≥ 0.10% (2x slippage assumption) to provide adequate cost buffer. Strategies with <0.08% avg trade should be flagged as cost-sensitive. Additionally, validation trade count should be at least 40% of training trade count to confirm opportunity consistency across regimes."
}
```