research: gate PORT06 index_comp_disp — promosso marginale, documentato e rimandato
Decorrela bene (corr 0.06 col MASTER, smentisce il timore ridondanza) ma beneficio OOS nullo (Sharpe 8.58->8.56, DD 1.36->1.40); migliora solo FULL DD 3.96->3.73. Non deployato (wiring + simulato per guadagno OOS nel rumore). Gate riusabile committato. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Adriano Dal Pastro
@@ -0,0 +1,156 @@
|
||||
"""GATE PORT06 del candidato index_comp_disp W=168 (ricerca dispersion 2026-06-08).
|
||||
|
||||
Edge confermato avversarialmente: fade della componente idiosincratica di BTC verso
|
||||
l'indice EW, gated da alta dispersione. Config: rel_len=12, z_win=336, z_thr=1.5,
|
||||
disp_168 >= quantile rolling 0.7 (win 720), TP=1.0*ATR14, SL=1.5*ATR14, max_bars=24.
|
||||
|
||||
Domanda del gate (lezione FR01: robusto != migliora-il-portafoglio):
|
||||
1) correlazione daily col MASTER e con le fade BTC esistenti (e' un diversificatore?)
|
||||
2) PORT06 BASE (17 sleeve) vs +DISP (18 sleeve) con pesi cap: DeltaSharpe/DeltaDD FULL e OOS.
|
||||
PROMOSSO solo se decorrela E migliora (o non degrada) l'OOS.
|
||||
|
||||
uv run python scripts/analysis/dispersion_edges/gate_index_comp_disp.py
|
||||
"""
|
||||
from __future__ import annotations
|
||||
|
||||
import sys
|
||||
from pathlib import Path
|
||||
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
PROJECT_ROOT = Path(__file__).resolve().parents[3]
|
||||
sys.path.insert(0, str(PROJECT_ROOT))
|
||||
|
||||
from scripts.analysis.dispersion_lab import features, align_to
|
||||
from scripts.analysis.explore_lab import get_df, atr
|
||||
from scripts.analysis.combine_portfolio import _norm, IDX, port_returns, metrics, SPLIT, OOS_DATE
|
||||
from scripts.analysis.honest_improve2 import _daily_equity
|
||||
from scripts.portfolios._defs import PORTFOLIOS
|
||||
from src.portfolio import weighting as W
|
||||
|
||||
FEE_RT, LEV, POS, INIT = 0.001, 3.0, 0.15, 1000.0
|
||||
CFG = dict(rel_len=12, z_win=336, z_thr=1.5, disp_q=0.7, disp_q_win=720,
|
||||
tp_atr=1.0, sl_atr=1.5, max_bars=24)
|
||||
|
||||
|
||||
def _last_rank(x):
|
||||
if x.shape[0] < 2:
|
||||
return np.nan
|
||||
return float((x[:-1] < x[-1]).mean())
|
||||
|
||||
|
||||
def build_trades(asset="BTC"):
|
||||
"""Entries CAUSALI + exit intrabar (TP/SL/max_bars) -> [(i, j, ret_netto)]."""
|
||||
df = get_df(asset, "1h")
|
||||
F = features()
|
||||
fa = align_to(F, df)
|
||||
c, h, l = df["close"].values, df["high"].values, df["low"].values
|
||||
n = len(c)
|
||||
a14 = atr(df, 14)
|
||||
rel = fa[f"rel_{asset}"].values.astype(float)
|
||||
disp = fa["disp_168"].values.astype(float)
|
||||
# somma rolling rel su rel_len, z-score causale (mean/std rolling z_win shift 1)
|
||||
rs = pd.Series(rel).rolling(CFG["rel_len"]).sum()
|
||||
rmean = rs.rolling(CFG["z_win"]).mean().shift(1)
|
||||
rstd = rs.rolling(CFG["z_win"]).std().shift(1)
|
||||
z = ((rs - rmean) / rstd.replace(0, np.nan)).values
|
||||
dpct = pd.Series(disp).rolling(CFG["disp_q_win"]).apply(_last_rank, raw=True).values
|
||||
fee = FEE_RT * LEV
|
||||
out = []
|
||||
last = -1
|
||||
for i in range(n - 1):
|
||||
if i <= last or not np.isfinite(z[i]) or not np.isfinite(dpct[i]):
|
||||
continue
|
||||
if dpct[i] < CFG["disp_q"] or abs(z[i]) < CFG["z_thr"]:
|
||||
continue
|
||||
ai = a14[i]
|
||||
if not np.isfinite(ai) or ai <= 0:
|
||||
continue
|
||||
d = -1 if z[i] > 0 else 1
|
||||
tp = c[i] + d * CFG["tp_atr"] * ai
|
||||
sl = c[i] - d * CFG["sl_atr"] * ai
|
||||
mb = CFG["max_bars"]
|
||||
j = min(i + mb, n - 1)
|
||||
exit_p = c[j]
|
||||
for k in range(1, mb + 1):
|
||||
j = i + k
|
||||
if j >= n:
|
||||
j = n - 1; exit_p = c[j]; break
|
||||
if d == 1:
|
||||
if l[j] <= sl: exit_p = sl; break
|
||||
if h[j] >= tp: exit_p = tp; break
|
||||
else:
|
||||
if h[j] >= sl: exit_p = sl; break
|
||||
if l[j] <= tp: exit_p = tp; break
|
||||
if k == mb: exit_p = c[j]
|
||||
out.append((i, j, (exit_p - c[i]) / c[i] * d * LEV - fee))
|
||||
last = j
|
||||
return df, out
|
||||
|
||||
|
||||
def daily_equity(df, trades):
|
||||
ts = pd.to_datetime(df["timestamp"], unit="ms", utc=True)
|
||||
cap = INIT; eq_ts, eq_v = [], []
|
||||
for i, j, ret in sorted(trades, key=lambda t: t[1]):
|
||||
cap = max(cap + cap * POS * ret, 10.0)
|
||||
eq_ts.append(ts.iloc[j]); eq_v.append(cap)
|
||||
return _norm(_daily_equity(eq_ts, eq_v, IDX))
|
||||
|
||||
|
||||
def pmetrics(members, p, extra=None):
|
||||
ids = list(p.sleeve_ids) + ([extra] if extra else [])
|
||||
dr = pd.DataFrame({i: members[i].pct_change().fillna(0.0) for i in ids})
|
||||
if extra:
|
||||
caps = dict(p.caps); caps["DISP"] = caps.get("DISP", None)
|
||||
w = W.weight_vector(p.weighting, ids, dr, weights=p.weights,
|
||||
caps=p.caps, clusters={**{i:(p.clusters or {}).get(i,i) for i in p.sleeve_ids},
|
||||
**({extra:"disp"} if extra else {})},
|
||||
lookback=p.vol_lookback)
|
||||
drp = port_returns({i: members[i] for i in ids}, w)
|
||||
return metrics(drp), metrics(drp, lo=SPLIT)
|
||||
|
||||
|
||||
def main():
|
||||
p = PORTFOLIOS["PORT06"]
|
||||
print("=" * 100)
|
||||
print(" GATE PORT06 — candidato index_comp_disp W=168 (BTC) | famiglia DISP nuova")
|
||||
print(f" config {CFG} | OOS da {OOS_DATE}")
|
||||
print("=" * 100)
|
||||
|
||||
from src.portfolio.sleeves import all_sleeve_equities
|
||||
eq_base = dict(all_sleeve_equities())
|
||||
|
||||
df, trades = build_trades("BTC")
|
||||
disp_eq = daily_equity(df, trades)
|
||||
fr = (disp_eq.iloc[-1] / disp_eq.iloc[0] - 1) * 100
|
||||
o = disp_eq.iloc[SPLIT:]; ofr = (o.iloc[-1] / o.iloc[0] - 1) * 100
|
||||
print(f"\n[1] candidato standalone: {len(trades)} trade | FULL {fr:+.0f}% | OOS {ofr:+.0f}%")
|
||||
|
||||
# correlazione daily col MASTER e con le fade BTC
|
||||
dr_cand = disp_eq.pct_change().fillna(0.0)
|
||||
print("\n[2] correlazione daily col candidato (decorrela?):")
|
||||
for sid in ["MR01_BTC", "MR02_BTC", "MR07_BTC", "DIP01_BTC"]:
|
||||
corr = dr_cand.corr(eq_base[sid].pct_change().fillna(0.0))
|
||||
print(f" {sid:<12} corr {corr:+.3f}")
|
||||
master_dr = pd.DataFrame({i: eq_base[i].pct_change().fillna(0.0) for i in p.sleeve_ids}).mean(axis=1)
|
||||
print(f" {'MASTER(EW)':<12} corr {dr_cand.corr(master_dr):+.3f}")
|
||||
|
||||
# PORT06 base vs +DISP
|
||||
f_b, o_b = pmetrics(eq_base, p)
|
||||
members = dict(eq_base); members["DISP_BTC"] = disp_eq
|
||||
f_e, o_e = pmetrics(members, p, extra="DISP_BTC")
|
||||
print("\n[3] PORT06 BASE (17) vs +DISP (18):")
|
||||
print(f" {'':<10}{'FULL Sh':>9}{'FULL DD%':>10}{'OOS Sh':>9}{'OOS DD%':>9}")
|
||||
print(f" {'BASE':<10}{f_b['sharpe']:>9.2f}{f_b['dd']:>10.2f}{o_b['sharpe']:>9.2f}{o_b['dd']:>9.2f}")
|
||||
print(f" {'+DISP':<10}{f_e['sharpe']:>9.2f}{f_e['dd']:>10.2f}{o_e['sharpe']:>9.2f}{o_e['dd']:>9.2f}")
|
||||
print(f" {'DELTA':<10}{f_e['sharpe']-f_b['sharpe']:>+9.2f}{f_e['dd']-f_b['dd']:>+10.2f}"
|
||||
f"{o_e['sharpe']-o_b['sharpe']:>+9.2f}{o_e['dd']-o_b['dd']:>+9.2f}")
|
||||
|
||||
promoted = (o_e['sharpe'] >= o_b['sharpe'] - 0.02 and o_e['dd'] <= o_b['dd'] + 0.20
|
||||
and f_e['sharpe'] >= f_b['sharpe'] - 0.02)
|
||||
print("\n VERDETTO: " + (">>> PROMOSSO <<<" if promoted else ">>> BOCCIATO (diluisce, come FR01) <<<"))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
Reference in New Issue
Block a user