File size: 11,030 Bytes
1cdd6ba ff98dba 33a396a 1cdd6ba 33a396a ff98dba 33a396a ff98dba 1cdd6ba ff98dba 1cdd6ba 33a396a 1cdd6ba 33a396a 1cdd6ba 33a396a ff98dba 1cdd6ba 33a396a 1cdd6ba 33a396a 1cdd6ba 33a396a 1cdd6ba 33a396a ff98dba 1cdd6ba ff98dba 33a396a 1cdd6ba 33a396a 1cdd6ba 33a396a 1cdd6ba ff98dba 1cdd6ba 33a396a ff98dba 1cdd6ba 33a396a ff98dba 33a396a ff98dba 33a396a ff98dba 33a396a 1cdd6ba 33a396a ff98dba 33a396a 1cdd6ba ff98dba 1cdd6ba ff98dba 1cdd6ba 33a396a 1cdd6ba 33a396a 1cdd6ba 33a396a 1cdd6ba ff98dba | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 | """
regime.py — Market regime detection with ADX, volatility compression,
distance-from-mean filter, and regime confidence scoring.
Key fixes vs prior version:
- STRUCTURE_LOOKBACK halved (10) to reduce entry lag
- True ATR (not EWM-only) with percentile-based compression detection
- ADX for objective trend strength (replaces pure HH/HL heuristic)
- Regime confidence: composite of trend + structure + vol alignment
- Distance-from-mean filter to avoid entering extended moves
"""
from typing import Dict, Any
import numpy as np
import pandas as pd
from config import (
ATR_PERIOD,
STRUCTURE_LOOKBACK,
STRUCTURE_CONFIRM_BARS,
VOLATILITY_EXPANSION_MULT,
VOLATILITY_CONTRACTION_MULT,
VOL_COMPRESSION_LOOKBACK,
VOL_COMPRESSION_PERCENTILE,
VOL_EXPANSION_CONFIRM_MULT,
ADX_PERIOD,
ADX_TREND_THRESHOLD,
ADX_STRONG_THRESHOLD,
DIST_FROM_MEAN_MA,
DIST_FROM_MEAN_ATR_MAX,
REGIME_CONFIDENCE_MIN,
)
def compute_atr(df: pd.DataFrame, period: int = ATR_PERIOD) -> pd.Series:
high, low, prev_close = df["high"], df["low"], df["close"].shift(1)
tr = pd.concat(
[high - low, (high - prev_close).abs(), (low - prev_close).abs()],
axis=1,
).max(axis=1)
# Use Wilder's smoothing (RMA) — matches TradingView / industry standard
return tr.ewm(alpha=1.0 / period, adjust=False).mean()
def compute_adx(df: pd.DataFrame, period: int = ADX_PERIOD) -> pd.DataFrame:
"""
Returns DataFrame with columns: adx, di_plus, di_minus.
Uses Wilder smoothing throughout to match standard ADX definition.
"""
high, low, close = df["high"], df["low"], df["close"]
prev_high = high.shift(1)
prev_low = low.shift(1)
prev_close = close.shift(1)
dm_plus = (high - prev_high).clip(lower=0)
dm_minus = (prev_low - low).clip(lower=0)
# Zero out when the other direction is larger
mask = dm_plus >= dm_minus
dm_plus = dm_plus.where(mask, 0.0)
dm_minus = dm_minus.where(~mask, 0.0)
tr = pd.concat(
[high - low, (high - prev_close).abs(), (low - prev_close).abs()],
axis=1,
).max(axis=1)
alpha = 1.0 / period
atr_w = tr.ewm(alpha=alpha, adjust=False).mean()
sdm_plus = dm_plus.ewm(alpha=alpha, adjust=False).mean()
sdm_minus = dm_minus.ewm(alpha=alpha, adjust=False).mean()
di_plus = 100 * sdm_plus / atr_w.replace(0, np.nan)
di_minus = 100 * sdm_minus / atr_w.replace(0, np.nan)
dx = 100 * (di_plus - di_minus).abs() / (di_plus + di_minus).replace(0, np.nan)
adx = dx.ewm(alpha=alpha, adjust=False).mean()
return pd.DataFrame({"adx": adx, "di_plus": di_plus, "di_minus": di_minus})
def compute_structure(df: pd.DataFrame, lookback: int = STRUCTURE_LOOKBACK) -> pd.Series:
roll_high = df["high"].rolling(lookback).max()
roll_low = df["low"].rolling(lookback).min()
half = max(1, lookback // 2)
prev_high = roll_high.shift(half)
prev_low = roll_low.shift(half)
hh = roll_high > prev_high
hl = roll_low > prev_low
lh = roll_high < prev_high
ll = roll_low < prev_low
structure = pd.Series(0, index=df.index)
structure[hh & hl] = 1
structure[lh & ll] = -1
return structure
def compute_volatility_compression(
atr_series: pd.Series,
lookback: int = VOL_COMPRESSION_LOOKBACK,
percentile: float = VOL_COMPRESSION_PERCENTILE,
) -> pd.Series:
"""
Returns True where current ATR is below the Nth percentile of its
recent history — i.e., volatility is compressed (coiled).
"""
rolling_pct = atr_series.rolling(lookback).quantile(percentile / 100.0)
return atr_series < rolling_pct
def compute_volatility_expanding_from_compression(
atr_series: pd.Series,
compressed_series: pd.Series,
mult: float = VOL_EXPANSION_CONFIRM_MULT,
lookback: int = 5,
) -> pd.Series:
"""
Returns True where ATR is now expanding (current > recent_min * mult)
AND was compressed within the last `lookback` bars.
Catches the precise moment of volatility breakout from a base.
"""
recent_min_atr = atr_series.rolling(lookback).min().shift(1)
expanding = atr_series > recent_min_atr * mult
was_compressed = compressed_series.shift(1).rolling(lookback).max().fillna(0) > 0
return expanding & was_compressed
def compute_distance_from_mean(
df: pd.DataFrame,
atr_series: pd.Series,
ma_period: int = DIST_FROM_MEAN_MA,
atr_max: float = DIST_FROM_MEAN_ATR_MAX,
) -> pd.Series:
"""
Returns ATR-normalised distance of close from its SMA.
Values > atr_max mean price is too extended for a fresh long entry.
"""
sma = df["close"].rolling(ma_period).mean()
distance_atr = (df["close"] - sma) / atr_series.replace(0, np.nan)
return distance_atr
def classify_trend(
structure_series: pd.Series,
adx_df: pd.DataFrame,
lookback: int = STRUCTURE_CONFIRM_BARS,
) -> str:
recent_struct = structure_series.iloc[-lookback:]
bullish = (recent_struct == 1).sum()
bearish = (recent_struct == -1).sum()
adx_val = float(adx_df["adx"].iloc[-1]) if not np.isnan(adx_df["adx"].iloc[-1]) else 0.0
di_plus = float(adx_df["di_plus"].iloc[-1]) if not np.isnan(adx_df["di_plus"].iloc[-1]) else 0.0
di_minus = float(adx_df["di_minus"].iloc[-1]) if not np.isnan(adx_df["di_minus"].iloc[-1]) else 0.0
adx_trending = adx_val >= ADX_TREND_THRESHOLD
if adx_trending and di_plus > di_minus and bullish >= max(1, lookback // 2):
return "bullish"
if adx_trending and di_minus > di_plus and bearish >= max(1, lookback // 2):
return "bearish"
return "ranging"
def compute_regime_confidence(
trend: str,
adx_val: float,
structure: int,
vol_expanding_from_base: bool,
vol_ratio: float,
dist_atr: float,
) -> float:
"""
Composite confidence [0, 1] requiring alignment across:
- ADX trend strength
- Price structure
- Volatility expanding from compression
- Price not extended
Low confidence = system holds off even if other scores look good.
"""
score = 0.0
# ADX contribution (0 to 0.35)
if adx_val >= ADX_STRONG_THRESHOLD:
score += 0.35
elif adx_val >= ADX_TREND_THRESHOLD:
score += 0.20
else:
score += 0.05
# Structure alignment (0 to 0.25)
if trend == "bullish" and structure == 1:
score += 0.25
elif trend == "bearish" and structure == -1:
score += 0.25
elif structure == 0:
score += 0.10
else:
score += 0.0
# Volatility expanding from base (0 to 0.25)
if vol_expanding_from_base:
score += 0.25
elif 1.0 < vol_ratio < VOLATILITY_EXPANSION_MULT:
score += 0.10
else:
score += 0.0
# Price not extended (0 to 0.15)
abs_dist = abs(dist_atr) if not np.isnan(dist_atr) else 0.0
if abs_dist < 1.0:
score += 0.15
elif abs_dist < DIST_FROM_MEAN_ATR_MAX:
score += 0.07
else:
score += 0.0
return float(np.clip(score, 0.0, 1.0))
def detect_regime(df: pd.DataFrame) -> Dict[str, Any]:
atr_series = compute_atr(df, ATR_PERIOD)
adx_df = compute_adx(df, ADX_PERIOD)
structure_series = compute_structure(df, STRUCTURE_LOOKBACK)
compressed_series = compute_volatility_compression(atr_series)
expanding_from_base = compute_volatility_expanding_from_compression(
atr_series, compressed_series
)
dist_atr_series = compute_distance_from_mean(df, atr_series)
last_atr = float(atr_series.iloc[-1])
last_close = float(df["close"].iloc[-1])
last_structure = int(structure_series.iloc[-1])
last_adx = float(adx_df["adx"].iloc[-1]) if not np.isnan(adx_df["adx"].iloc[-1]) else 0.0
last_di_plus = float(adx_df["di_plus"].iloc[-1]) if not np.isnan(adx_df["di_plus"].iloc[-1]) else 0.0
last_di_minus = float(adx_df["di_minus"].iloc[-1]) if not np.isnan(adx_df["di_minus"].iloc[-1]) else 0.0
last_compressed = bool(compressed_series.iloc[-1])
last_expanding_from_base = bool(expanding_from_base.iloc[-1])
last_dist_atr = float(dist_atr_series.iloc[-1]) if not np.isnan(dist_atr_series.iloc[-1]) else 0.0
atr_ma = atr_series.rolling(ATR_PERIOD * 2).mean()
last_atr_ma = float(atr_ma.iloc[-1]) if not np.isnan(atr_ma.iloc[-1]) else last_atr
vol_ratio = last_atr / last_atr_ma if last_atr_ma > 0 else 1.0
vol_expanding = vol_ratio > VOLATILITY_EXPANSION_MULT
vol_contracting = vol_ratio < VOLATILITY_CONTRACTION_MULT
atr_pct = last_atr / last_close if last_close > 0 else 0.0
trend = classify_trend(structure_series, adx_df, STRUCTURE_CONFIRM_BARS)
price_too_extended_long = last_dist_atr > DIST_FROM_MEAN_ATR_MAX
price_too_extended_short = last_dist_atr < -DIST_FROM_MEAN_ATR_MAX
regime_confidence = compute_regime_confidence(
trend=trend,
adx_val=last_adx,
structure=last_structure,
vol_expanding_from_base=last_expanding_from_base,
vol_ratio=vol_ratio,
dist_atr=last_dist_atr,
)
# Regime score: raw directional quality
if trend == "bullish" and not vol_expanding:
regime_score = 1.0
elif trend == "bullish" and vol_expanding:
regime_score = 0.55
elif trend == "ranging":
regime_score = 0.25
elif trend == "bearish" and not vol_expanding:
regime_score = 0.15
else:
regime_score = 0.05
if last_adx >= ADX_STRONG_THRESHOLD:
regime_score = min(1.0, regime_score + 0.1)
elif last_adx < ADX_TREND_THRESHOLD:
regime_score = max(0.0, regime_score - 0.15)
if last_structure == 1:
regime_score = min(1.0, regime_score + 0.1)
elif last_structure == -1:
regime_score = max(0.0, regime_score - 0.1)
atr_ma_20 = atr_series.rolling(20).mean().iloc[-1]
atr_ma_50 = atr_series.rolling(50).mean().iloc[-1] if len(df) >= 50 else atr_ma_20
atr_trend_dir = "rising" if atr_ma_20 > atr_ma_50 else "falling"
return {
"atr": last_atr,
"atr_pct": atr_pct,
"atr_pct_pct": round(atr_pct * 100, 3),
"structure": last_structure,
"trend": trend,
"vol_ratio": round(vol_ratio, 3),
"vol_expanding": vol_expanding,
"vol_contracting": vol_contracting,
"vol_compressed": last_compressed,
"vol_expanding_from_base": last_expanding_from_base,
"adx": round(last_adx, 2),
"di_plus": round(last_di_plus, 2),
"di_minus": round(last_di_minus, 2),
"dist_atr": round(last_dist_atr, 3),
"price_extended_long": price_too_extended_long,
"price_extended_short": price_too_extended_short,
"regime_confidence": round(regime_confidence, 4),
"regime_score": round(float(np.clip(regime_score, 0.0, 1.0)), 4),
"atr_trend": atr_trend_dir,
"atr_series": atr_series,
"structure_series": structure_series,
"adx_series": adx_df,
"compressed_series": compressed_series,
"dist_atr_series": dist_atr_series,
}
|