mean-reversion-engine

# Mean Reversion Engine ## CRITICAL: Only use in RANGING regimes. Mean reversion in trends = catching knives. ```python import pandas as pd import numpy as np class MeanReversionEngine: @staticmethod def bollinger_bounce(df: pd.DataFrame, period: int = 20, std_mult: float = 2.0) -> dict: """Buy at lower band, sell at upper band. Classic range strategy.""" close = df["close"] mid = close.rolling(period).mean() std = close.rolling(period).std() upper = mid + std_mult * std lower = mid - std_mult * std pct_b = (close - lower) / (upper - lower) current = df.iloc[-1] return { "strategy": "bollinger_bounce", "upper": round(upper.iloc[-1], 5), "middle": round(mid.iloc[-1], 5), "lower": round(lower.iloc[-1], 5), "pct_b": round(pct_b.iloc[-1], 3), "signal": "BUY (at lower band)" if pct_b.iloc[-1] < 0.05 else "SELL (at upper band)" if pct_b.iloc[-1] > 0.95 else "WAIT", "target": round(mid.iloc[-1], 5), "stop": round(lower.iloc[-1] - (upper.iloc[-1] - lower.iloc[-1]) * 0.25, 5) if pct_b.iloc[-1] < 0.05 else round(upper.iloc[-1] + (upper.iloc[-1] - lower.iloc[-1]) * 0.25, 5), } @staticmethod def rsi_extreme_fade(df: pd.DataFrame, period: int = 14, oversold: float = 25, overbought: float = 75) -> dict: """Fade RSI extremes with divergence confirmation.""" close = df["close"] delta = close.diff() gain = delta.where(delta > 0, 0).rolling(period).mean() loss = (-delta.where(delta < 0, 0)).rolling(period).mean() rsi = 100 - (100 / (1 + gain / loss.replace(0, np.nan))) # Divergence: price makes new low but RSI makes higher low (bullish) price_lower = close.iloc[-1] < close.rolling(20).min().iloc[-5] rsi_higher = rsi.iloc[-1] > rsi.rolling(20).min().iloc[-5] bull_divergence = price_lower and rsi_higher and rsi.iloc[-1] < 40 price_higher = close.iloc[-1] > close.rolling(20).max().iloc[-5] rsi_lower = rsi.iloc[-1] < rsi.rolling(20).max().iloc[-5] bear_divergence = price_higher and rsi_lower and rsi.iloc[-1] > 60 return { "strategy": "rsi_extreme_fade", "rsi": round(rsi.iloc[-1], 1), "oversold": rsi.iloc[-1] < oversold, "overbought": rsi.iloc[-1] > overbought, "bullish_divergence": bull_divergence, "bearish_divergence": bear_divergence, "signal": "BUY (oversold + divergence)" if rsi.iloc[-1] < oversold and bull_divergence else "BUY (oversold)" if rsi.iloc[-1] < oversold else "SELL (overbought + divergence)" if rsi.iloc[-1] > overbought and bear_divergence else "SELL (overbought)" if rsi.iloc[-1] > overbought else "WAIT", "quality": "A+" if bull_divergence or bear_divergence else "B", } @staticmethod def zscore_reversion(df: pd.DataFrame, lookback: int = 60, entry_z: float = 2.0, exit_z: float = 0.5) -> dict: """Z-score based reversion with configurable thresholds.""" close = df["close"] mean = close.rolling(lookback).mean() std = close.rolling(lookback).std() z = (close - mean) / std.replace(0, np.nan) return { "strategy": "zscore_reversion", "z_score": round(z.iloc[-1], 3), "mean": round(mean.iloc[-1], 5), "signal": "BUY (z < -2)" if z.iloc[-1] < -entry_z else "SELL (z > 2)" if z.iloc[-1] > entry_z else "EXIT" if abs(z.iloc[-1]) < exit_z and abs(z.iloc[-2]) > exit_z else "WAIT", "target": round(mean.iloc[-1], 5), "distance_to_mean_pct": round((close.iloc[-1] / mean.iloc[-1] - 1) * 100, 2), } @staticmethod def scan_all(df: pd.DataFrame, symbol: str = "") -> dict: return { "symbol": symbol, "bollinger": MeanReversionEngine.bollinger_bounce(df), "rsi_fade": MeanReversionEngine.rsi_extreme_fade(df), "zscore": MeanReversionEngine.zscore_reversion(df), "WARNING": "Mean reversion ONLY works in ranging markets. Check regime first.", } ``` ## Practical Mean-Reversion Scalping: EMA + Bollinger Band (CodeTrading) > Source: "Trading with Python: Simple Scalping Strategy" by CodeTrading (Jan 2024) A practical M5 scalping implementation of mean-reversion with trend filter: - **Entry:** Price touches lower BB (long) or upper BB (short) while dual EMA confirms trend direction - **Logic:** Mean-reversion (BB bounce to center) filtered by trend-following (EMA 30/50 alignment for 6+ candles) - **Results:** 25% return in 3 months on EURUSD M5, 1,671 trades, 44% win rate - **Key insight:** Combining trend-following with mean-reversion produces steadily increasing equity — the trend filter prevents trading BB bounces against the trend - SL = ATR * 1.1, TP = SL * 1.5 For full implementation details and Python code, see `scalping-framework` skill.