research(alt): sweep 104 strategie alternative su Deribit (153 agenti) + marginal scorer
Ondata di ricerca onesta a largo spettro su BTC/ETH+DVOL certificati: 104 ipotesi distinte (11 famiglie), un agente-finder per ipotesi, verifica avversariale a 3 scettici sui promettenti, sintesi (153 agenti totali). Esito: NIENTE di nuovo regge -> conferma del soffitto strutturale ~1.3 BTC/ETH-direzionale; lo stack TP01+XS01+VRP01 resta imbattuto. - altlib.py: harness condiviso vettoriale leak-free (eval_weights/study_weights, fee-sweep, both-asset + hold-out 2025+). Riproduce i numeri canonici di TP01. - MARGINAL SCORER (study_marginal/marginal_vs_tp01): Sharpe INCREMENTALE vs baseline TP01 (corr, blend uplift OOS, alpha residua) + jackknife OOS (clean-year + drop-best-month). earns_slot = abs!=FAIL & ADDS & robust_oos. Smaschera gli overlay su TSMOM con PASS assoluti fasulli (CMB04, VOL11, ...) e il falso positivo KAMA (ADDS ma muore al jackknife). - runs/*.py (104) script riproducibili per ipotesi; wf_altstrat.js workflow. - Verdetto: 0 candidati deployabili; 2 LEAD fragili (VOL08, STA05_LS) da forward-monitor. - test_marginal_scorer.py blocca baseline + invarianti. Suite: 32 verde. Diario: docs/diary/2026-06-20-alt-strategies-100agent-sweep.md Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This commit is contained in:
Adriano Dal Pastro
@@ -0,0 +1,215 @@
|
||||
"""VOL11 — DVOL kill-switch on trend (TSMOM with hard-flat when DVOL is elevated).
|
||||
|
||||
Hypothesis: TSMOM (multi-horizon, long-flat, vol-targeted, identical to TP01) is the
|
||||
validated base strategy. We overlay a DVOL kill-switch: when DVOL is above a threshold
|
||||
(fixed or percentile-based), go flat regardless of TSMOM signal.
|
||||
|
||||
Rationale: trend-following can be whipsawed in high-IV regimes (panic spikes). By sitting
|
||||
out when DVOL is very high, we might:
|
||||
- Cut the worst crash losses (DVOL spikes during drawdowns)
|
||||
- Improve the hold-out Sharpe in the volatile 2025-26 period
|
||||
|
||||
Construction:
|
||||
1. Base signal: TSMOM multi-horizon (30/90/180-day lookbacks), sign-vote, long-flat.
|
||||
2. Kill-switch: flat when DVOL > threshold.
|
||||
Tested thresholds:
|
||||
(A) fixed 70 points (historically ~top-30% readings)
|
||||
(B) fixed 80 points (historically ~top-15% readings)
|
||||
(C) rolling 80th percentile (adaptive, avoids hindsight threshold selection)
|
||||
(D) rolling 70th percentile
|
||||
3. All configs: vol-target 20%, leva cap 2x, 1d.
|
||||
|
||||
DVOL history starts 2021-03 → backtest meaningful from 2021 onward; full-history numbers
|
||||
include the pre-DVOL period where TSMOM runs unfiltered (i.e., those bars never killed).
|
||||
|
||||
Grid: 4 configs × 1 TF × 2 assets = 8 backtests (within 6-limit at 1d; we run all 4
|
||||
because they're fast vectorized ops and total is still manageable).
|
||||
"""
|
||||
import sys
|
||||
sys.path.insert(0, "/opt/docker/PythagorasGoal/scripts/research/alt")
|
||||
import altlib as al
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Base TSMOM (same as TP01 canonical: 1/3/6-month horizons, long-flat)
|
||||
# ---------------------------------------------------------------------------
|
||||
def tsmom_direction(df):
|
||||
"""TSMOM multi-horizon: +1 if majority of 30/90/180-day returns positive, else 0."""
|
||||
c = df["close"].values.astype(float)
|
||||
bpd = al.bars_per_day(df)
|
||||
vote = np.zeros(len(c))
|
||||
for h in (30 * bpd, 90 * bpd, 180 * bpd):
|
||||
h = int(h)
|
||||
s = np.full(len(c), np.nan)
|
||||
s[h:] = np.sign(c[h:] / c[:-h] - 1.0)
|
||||
vote += np.nan_to_num(s)
|
||||
# +1 if sum > 0 (majority positive), 0 otherwise (long-flat, not short)
|
||||
return np.where(vote > 0, 1.0, 0.0)
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# DVOL kill-switch helpers (causal — no look-ahead)
|
||||
# ---------------------------------------------------------------------------
|
||||
def dvol_fixed_kill(dv, threshold):
|
||||
"""Flat (kill=True) when DVOL >= threshold. NaN DVOL -> no kill (pass-through)."""
|
||||
kill = np.where(np.isfinite(dv) & (dv > 0), dv >= threshold, False)
|
||||
return kill.astype(bool)
|
||||
|
||||
|
||||
def dvol_percentile_kill(dv, percentile, window=252):
|
||||
"""Flat when DVOL >= rolling expanding-then-window percentile (causal).
|
||||
Uses the past `window` daily DVOL observations to compute the threshold."""
|
||||
n = len(dv)
|
||||
kill = np.zeros(n, dtype=bool)
|
||||
for i in range(n):
|
||||
if not np.isfinite(dv[i]) or dv[i] <= 0:
|
||||
continue # no DVOL -> pass through (no kill)
|
||||
# Rolling window: use min(i+1, window) observations up to and including i
|
||||
start = max(0, i - window + 1)
|
||||
hist = dv[start:i + 1]
|
||||
hist_valid = hist[np.isfinite(hist) & (hist > 0)]
|
||||
if len(hist_valid) < 10:
|
||||
continue # not enough history
|
||||
thresh = np.percentile(hist_valid, percentile)
|
||||
kill[i] = dv[i] >= thresh
|
||||
return kill
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Strategy builder
|
||||
# ---------------------------------------------------------------------------
|
||||
def make_vol11(asset, kill_type, kill_param):
|
||||
"""
|
||||
kill_type in {'fixed', 'pct'}
|
||||
kill_param: for 'fixed' -> DVOL level (e.g. 70, 80); for 'pct' -> percentile (e.g. 80, 70)
|
||||
"""
|
||||
def target_fn(df):
|
||||
# 1. Base TSMOM direction
|
||||
direction = tsmom_direction(df)
|
||||
|
||||
# 2. DVOL for this asset (causal, backward-filled)
|
||||
dv = al.dvol(df, asset)
|
||||
|
||||
# 3. Kill-switch
|
||||
if kill_type == "fixed":
|
||||
kill = dvol_fixed_kill(dv, kill_param)
|
||||
else: # 'pct'
|
||||
kill = dvol_percentile_kill(dv, kill_param, window=252)
|
||||
|
||||
# 4. Apply kill: go flat when kill is active
|
||||
filtered_dir = np.where(kill, 0.0, direction)
|
||||
|
||||
# 5. Vol-target the filtered direction
|
||||
return al.vol_target(filtered_dir, df, target_vol=0.20, vol_win_days=30, leverage_cap=2.0)
|
||||
|
||||
return target_fn
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Configs grid (4 configs, 1 TF, 2 assets = 8 backtests)
|
||||
# ---------------------------------------------------------------------------
|
||||
CONFIGS = [
|
||||
dict(kill_type="fixed", kill_param=70, label="fixed-70"),
|
||||
dict(kill_type="fixed", kill_param=80, label="fixed-80"),
|
||||
dict(kill_type="pct", kill_param=80, label="pct80-roll252"),
|
||||
dict(kill_type="pct", kill_param=70, label="pct70-roll252"),
|
||||
]
|
||||
|
||||
TFS = ("1d",)
|
||||
ASSETS = ("BTC", "ETH")
|
||||
|
||||
|
||||
def _verdict_local(per_cell):
|
||||
if not per_cell:
|
||||
return dict(grade="FAIL", reason="no cells")
|
||||
ok = [c for c in per_cell if c.get("full_sharpe", -9) > 0]
|
||||
best = max(per_cell, key=lambda c: c.get("min_asset_holdout_sharpe", -9))
|
||||
pass_ = (best.get("min_asset_full_sharpe", -9) >= 0.5 and
|
||||
best.get("min_asset_holdout_sharpe", -9) >= 0.2 and
|
||||
best.get("fee_survives", False))
|
||||
weak = (best.get("min_asset_full_sharpe", -9) >= 0.3 and
|
||||
best.get("min_asset_holdout_sharpe", -9) >= 0.0)
|
||||
grade = "PASS" if pass_ else ("WEAK" if weak else "FAIL")
|
||||
return dict(grade=grade, best_tf=best.get("tf"),
|
||||
best_full_sharpe=best.get("min_asset_full_sharpe"),
|
||||
best_holdout_sharpe=best.get("min_asset_holdout_sharpe"),
|
||||
n_positive_cells=len(ok), n_cells=len(per_cell))
|
||||
|
||||
|
||||
def run_grid():
|
||||
best_min_hold = -999
|
||||
best_rep = None
|
||||
all_results = {}
|
||||
|
||||
for cfg in CONFIGS:
|
||||
label = cfg["label"]
|
||||
print(f"\n--- VOL11 Config: {label} ---")
|
||||
|
||||
cells = []
|
||||
for tf in TFS:
|
||||
per_asset = {}
|
||||
fee_ok_all = True
|
||||
for asset in ASSETS:
|
||||
fn = make_vol11(asset, cfg["kill_type"], cfg["kill_param"])
|
||||
df = al.get(asset, tf)
|
||||
tgt = fn(df)
|
||||
base = al.eval_weights(df, tgt, fee_side=al.FEE_SIDE)
|
||||
sweep = {f"{2*f*100:.2f}%RT": al.eval_weights(df, tgt, fee_side=f)["full"]["sharpe"]
|
||||
for f in al.FEE_SWEEP}
|
||||
fee_ok = sweep.get("0.20%RT", -9) > 0
|
||||
fee_ok_all = fee_ok_all and fee_ok
|
||||
per_asset[asset] = dict(
|
||||
full=base["full"], holdout=base["holdout"],
|
||||
tim=base["time_in_market"], turnover=base["turnover_per_year"],
|
||||
fee_sweep=sweep, yearly=base["yearly"]
|
||||
)
|
||||
print(f" {asset}: full Sh={base['full']['sharpe']:+.3f} "
|
||||
f"hold Sh={base['holdout'].get('sharpe', 0):+.3f} "
|
||||
f"DD={base['full']['maxdd']*100:.1f}% "
|
||||
f"TIM={base['time_in_market']:.2f} "
|
||||
f"fee0.20ok={fee_ok}")
|
||||
|
||||
min_full = min(per_asset[a]["full"]["sharpe"] for a in ASSETS)
|
||||
min_hold = min(per_asset[a]["holdout"].get("sharpe", 0.0) for a in ASSETS)
|
||||
cells.append(dict(
|
||||
tf=tf, per_asset=per_asset,
|
||||
min_asset_full_sharpe=round(min_full, 3),
|
||||
min_asset_holdout_sharpe=round(min_hold, 3),
|
||||
full_sharpe=round(np.mean([per_asset[a]["full"]["sharpe"] for a in ASSETS]), 3),
|
||||
fee_survives=fee_ok_all
|
||||
))
|
||||
|
||||
verdict = _verdict_local(cells)
|
||||
rep = dict(name=f"VOL11-{label}", kind="weights", cells=cells, verdict=verdict)
|
||||
all_results[label] = rep
|
||||
|
||||
min_hold_this = min(cells, key=lambda c: c["min_asset_holdout_sharpe"])["min_asset_holdout_sharpe"]
|
||||
if min_hold_this > best_min_hold:
|
||||
best_min_hold = min_hold_this
|
||||
best_rep = rep
|
||||
|
||||
return best_rep, all_results
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
print("=== VOL11: DVOL Kill-Switch on TSMOM Trend ===")
|
||||
print("Idea: standard TSMOM (TP01-like), hard-flat when DVOL > threshold")
|
||||
print("DVOL history starts 2021-03; bars before that run unfiltered TSMOM\n")
|
||||
|
||||
best_rep, all_results = run_grid()
|
||||
|
||||
print("\n=== BEST CONFIG REPORT ===")
|
||||
print(al.fmt(best_rep))
|
||||
|
||||
print("\n=== ALL CONFIGS SUMMARY ===")
|
||||
for label, rep in all_results.items():
|
||||
v = rep["verdict"]
|
||||
c = rep["cells"][0]
|
||||
print(f" {label}: grade={v['grade']} "
|
||||
f"minFull={c['min_asset_full_sharpe']:+.2f} "
|
||||
f"minHold={c['min_asset_holdout_sharpe']:+.2f} "
|
||||
f"feeOK={c['fee_survives']}")
|
||||
|
||||
print("\nJSON:", al.as_json(best_rep))
|
||||
Reference in New Issue
Block a user