← Back to Reviews
20260114_124030_defensive_vol_calm.md
# AI Review: defensive_vol_calm (1-Minute)
**Date:** Wed Jan 14 12:40:30 CET 2026
**Strategy File:** /root/trade_1m/strategies/defensive_vol_calm.py
---
```json
{
"score": 35,
"flags": [
"overfitting",
"concentration",
"cost_sensitive",
"inconsistent"
],
"review": "## Critical Issues\n\n### 1. Severe Overfitting to Specific Market Regime\n\nThe strategy shows classic signs of overfitting to a specific market period:\n- **Extreme concentration**: 10 out of 23 total trades (43%) occur in a single month\n- **Wildly inconsistent monthly performance**: Ranges from -0.5% to +2.3%, with most months near-flat\n- **Most return from one period**: The 2.14% total return is almost entirely driven by one month's 2.3% return\n\nThis is a **RED FLAG** for 1-minute strategies. A truly robust microstructure edge should work consistently across different market conditions, not suddenly \"wake up\" in one specific month.\n\n### 2. Overly Specific Parameter Combinations\n\nThe strategy stacks multiple highly-tuned conditions:\n- ATR z-score threshold of **exactly -2.0** (why not -1.5 or -2.5?)\n- Volume surge of **exactly 3.0x** (why not 2.5x or 3.5x?)\n- **Exactly 3** consecutive bars (why not 2 or 4?)\n- 360-bar cooldown (why exactly 6 hours?)\n- 15-bar hold period (why exactly 15 minutes?)\n\nThese precise values suggest curve-fitting to historical noise rather than capturing a genuine microstructure phenomenon. Round numbers like 20, 120, 360 are good, but the **combination** of multiple specific thresholds is suspicious.\n\n### 3. Cost Sensitivity at Low Trade Frequency\n\nWith only 23 trades over 5 months (average 0.15 trades/day):\n- **0.10% slippage cost** = 23 trades × 0.10% = **2.3% total cost**\n- **Return after costs**: 2.14% - 2.3% = **-0.16%** (nearly breakeven, but negative)\n- Average trade profit: 0.093%, which is barely 1x the round-trip cost\n\nThe \"defensive\" label doesn't excuse being cost-sensitive. A defensive strategy should either:\n- Trade even LESS with higher conviction\n- Have wider edges per trade to survive costs\n\nThis strategy is stuck in the middle: too few trades to diversify risk, too thin edges to survive execution reality.\n\n### 4. Statistical Reliability: Not Enough Data\n\n23 trades is **insufficient sample size** to validate any strategy:\n- With 60% win rate and 23 trades, even a coin-flip strategy could produce similar results\n- The one high-performing month (10 trades, 60% WR) could easily be luck\n- Need 100+ trades minimum to distinguish skill from noise at 1-minute frequency\n\n### 5. The \"Extreme Calm Before Volume Surge\" Thesis is Questionable\n\nThe core hypothesis: \"ATR < -2 std + 3x volume surge = edge\" is theoretically weak:\n- **Why would extreme calm + sudden volume be predictive?** \n - If it's a breakout, the move already happened (you're late)\n - If it's a false breakout, you're catching a knife\n- The 3-consecutive-bar filter tries to add directionality, but this is **reactive**, not predictive\n- 15-minute hold is arbitrary—no microstructure edge has a fixed time decay\n\n### 6. Month-Over-Month Validation Failure\n\nBreaking down the monthly Sharpe ratios:\n- Two negative months (-0.47, -1.07)\n- One excellent month (+2.04)\n- Two mediocre months (+0.58, +1.59)\n\nThis is **NOT consistent**. A real microstructure edge (VWAP reversion, RSI extremes, vol spike mean reversion) should show positive Sharpe in 4 out of 5 months, not this wild variance.\n\n## What Would Make This Strategy Better?\n\n1. **Simplify conditions**: Remove at least 2 of the 5 entry filters. Too many ANDs = curve-fitting\n2. **Widen the edge**: If you're only trading 0.15x/day, each trade should target 0.3%+ profit (3x cost buffer)\n3. **Dynamic exits**: Use price-based exits (e.g., VWAP reversion complete) instead of arbitrary 15-bar timer\n4. **Prove consistency**: Should work in 4+ out of 5 months with Sharpe > 0.3 in each\n5. **Reduce parameters**: Try just ONE condition: \"ATR z-score < -2 AND volume > 3x\" with 3-bar direction. See if that alone works. If not, the edge doesn't exist.\n\n## Verdict\n\nThis is a **curve-fit defensive strategy** that accidentally caught one good month. The \"defensive\" framing is misleading—being flat most of the time isn't defensive if your rare trades are cost-sensitive and inconsistent.\n\n**Score: 35/100** - The strategy has some structure (reasonable warmup, no obvious lookahead), but the concentration risk, cost sensitivity, and month-to-month inconsistency make it unviable for live 1-minute trading.",
"rule_suggestion": "Defensive strategies with <50 trades over 5 months must demonstrate: (1) Positive Sharpe in at least 4 out of 5 months, (2) Average trade profit >3x round-trip cost (0.3%), (3) No single month contributing >40% of total return. Ultra-low-frequency strategies magnify the impact of overfitting and must show stronger consistency to compensate for small sample size."
}
```