20260114_061011_eth_vol_spike_hedge.md

# AI Review: eth_vol_spike_hedge **Date:** Wed Jan 14 06:10:11 CET 2026 **Strategy File:** /root/trade_rules/strategies/eth_vol_spike_hedge.py --- ```json { "score": 25, "flags": ["overfitting", "inconsistent", "concentration", "complexity"], "review": "## Critical Issues\n\n### 1. Validation Period Failure (Fatal)\nThe strategy produces **0% return with 0 trades** in the validation period, while showing 16.85% total return in training. This is a complete failure - the pattern it learned doesn't generalize to unseen data.\n\n### 2. Severe Year-to-Year Inconsistency\nThe year-by-year breakdown reveals extreme instability:\n- **2022**: -26.4% return, 20% win rate, -2.95 Sharpe (catastrophic failure)\n- **2023**: 0% return, 0 trades (pattern didn't exist)\n- **2020-2021, 2024-2025**: Positive returns with reasonable Sharpe\n\nThis is classic overfitting to specific market conditions. A robust hedge should provide some protection consistently, not collapse in certain years.\n\n### 3. Over-Engineered Entry Conditions\nThe strategy requires **5 simultaneous conditions** for entry:\n1. Volume > 3x average\n2. Bar down > 5%\n3. RSI < 30\n4. Price < EMA50\n5. Lower wick > 20% of range\n\nThis is excessive filtering that creates a brittle pattern. The more conditions you stack, the more you're curve-fitting to historical coincidences rather than robust market dynamics.\n\n### 4. Suspiciously Specific Thresholds\nMultiple parameters appear curve-fitted:\n- **3.0x volume threshold**: Why exactly 3.0? Not a round standard value\n- **5% drop requirement**: Too precise\n- **20% wick ratio**: Oddly specific\n- **RSI exit at 55**: Not a standard level (30/70 are standard)\n\nThese look optimized to historical data rather than based on market theory.\n\n### 5. Low Trade Count\nWith only **49 total trades** across 5.5 years of training data (average 9 trades/year), there's insufficient statistical significance. The strategy is too selective, and results are highly dependent on a few specific setups.\n\n### 6. Execution Concerns\nThe strategy assumes:\n- Perfect detection of \"capitulation\" on a 4h bar close\n- Ability to enter at next bar open after extreme volatility\n- No slippage consideration during panic selling events\n\nIn reality, entering during volume spikes involves significant slippage and execution risk.\n\n## Why It Failed Validation\n\nThe strategy learned a very specific pattern that occurred in certain historical periods but doesn't represent a persistent market inefficiency. The combination of 5 conditions creates a \"phantom pattern\" - it looks good in backtest but doesn't exist going forward.\n\n## Positive Notes\n\n- Uses standard indicators (RSI, EMA) rather than proprietary metrics\n- No lookahead bias detected\n- Time-based exits show risk awareness\n- Code quality is clean and readable\n\n## Recommendation\n\n**REJECT** - This strategy fails validation completely and shows severe overfitting. The 2022 collapse and 2023 absence demonstrate it doesn't work across market regimes. Starting from scratch with a simpler, more robust approach would be better than trying to salvage this.", "rule_suggestion": "**Maximum Entry Conditions Rule**: Strategies with more than 3 simultaneous entry conditions must demonstrate consistent performance across ALL years in the training period (no single year with negative Sharpe below -1.0 or zero trades). Complex filters often create phantom patterns that don't generalize." } ```