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>
2026-06-20 19:50:39 +00:00
parent bf84bc91e2
commit 5ac4e16af8
111 changed files with 16924 additions and 0 deletions
@@ -0,0 +1,128 @@
+"""VOL08 — Realized-vol term structure overlay on long.
+
+HYPOTHESIS: Ratio short-window vol (5d) / long-window vol (30d).
+  >1 (vol rising, de-risk) -> reduce position
+  <1 (vol falling, risk-on) -> increase position
+
+Overlay on a long-only base position (TSMOM trend direction), vol-targeted.
+The vol-term-structure ratio modulates position size:
+  position = base_dir * vol_target * clamp(1 / ratio, 0.0, 1.0)
+
+Grid:
+  short_win: [5, 10] days
+  long_win: [21, 63] days
+  -> 4 configs x 2 TFs (1d, 12h) = 8 backtests total, but we pick best config first on 1d
+  then verify best config on 12h -> capped at 6 total.
+
+Plan:
+  - Run 4 configs on 1d to find best
+  - Run best config on 12h
+  - Report rep for best config
+
+Implementation:
+  1. Compute TSMOM direction (1m,3m,6m blend, long-flat)
+  2. Vol-target the direction (target_vol=0.20, cap=2x)
+  3. Multiply by vol-ratio scaling: scale = clip(long_vol / short_vol, 0, 1)
+     (when short_vol > long_vol -> ratio > 1 -> scale < 1: de-risk)
+     (when short_vol < long_vol -> ratio < 1 -> scale > 1, but clipped at 1: stay full)
+"""
+import sys
+sys.path.insert(0, "/opt/docker/PythagorasGoal/scripts/research/alt")
+import altlib as al
+import numpy as np
+
+
+def make_target(short_days: int, long_days: int):
+    """Return a target function for the given short/long vol windows."""
+    def target_fn(df):
+        c = df["close"].values.astype(float)
+        bpd = al.bars_per_day(df)
+        bpy = bpd * 365.25
+        r = al.simple_returns(c)
+
+        # --- TSMOM long-flat direction (1m, 3m, 6m) ---
+        horizons = [30 * bpd, 90 * bpd, 180 * bpd]
+        direction = np.zeros(len(c))
+        for h in horizons:
+            h = int(h)
+            sig = np.full(len(c), np.nan)
+            if h < len(c):
+                sig[h:] = np.sign(c[h:] / c[:-h] - 1.0)
+            direction += np.nan_to_num(sig)
+        # long-flat (0 or +1)
+        long_flat = np.clip(np.sign(direction), 0.0, 1.0)
+
+        # --- Vol-targeted base position ---
+        vol_win = max(2, 30 * bpd)
+        rv30 = al.realized_vol(r, int(vol_win), bpy)
+        base_scale = np.where((rv30 > 0) & np.isfinite(rv30), 0.20 / rv30, 0.0)
+        base_pos = np.clip(long_flat * base_scale, 0.0, 2.0)
+
+        # --- Vol term structure overlay ---
+        short_win = max(2, short_days * bpd)
+        long_win_b = max(2, long_days * bpd)
+        rv_short = al.realized_vol(r, int(short_win), bpy)
+        rv_long = al.realized_vol(r, int(long_win_b), bpy)
+
+        # scale = long_vol / short_vol, clipped to [0, 1]
+        # >1 vol rising (short > long): scale < 1 -> de-risk
+        # <1 vol falling (short < long): scale > 1, clipped at 1 -> stay full
+        with np.errstate(divide="ignore", invalid="ignore"):
+            ratio = np.where(
+                (rv_short > 0) & np.isfinite(rv_short) & np.isfinite(rv_long),
+                rv_long / rv_short,
+                1.0
+            )
+        scale = np.clip(ratio, 0.0, 1.0)
+
+        pos = base_pos * scale
+        pos = np.nan_to_num(pos, nan=0.0)
+        return pos
+
+    return target_fn
+
+
+if __name__ == "__main__":
+    print("VOL08 — Realized-vol term structure overlay")
+    print("=" * 60)
+
+    # Grid: 4 configs on 1d
+    grid = [
+        (5, 21),
+        (5, 63),
+        (10, 21),
+        (10, 63),
+    ]
+
+    best_rep = None
+    best_hold_sh = -999.0
+    best_label = ""
+
+    for short_d, long_d in grid:
+        label = f"VOL08-s{short_d}d-l{long_d}d"
+        print(f"\n--- Testing {label} on 1d ---")
+        rep = al.study_weights(
+            label,
+            make_target(short_d, long_d),
+            tfs=("1d",)
+        )
+        print(al.fmt(rep))
+        hold_sh = rep["verdict"].get("best_holdout_sharpe", -999.0)
+        if hold_sh > best_hold_sh:
+            best_hold_sh = hold_sh
+            best_rep = rep
+            best_label = label
+            best_short = short_d
+            best_long = long_d
+
+    print(f"\n*** Best config: {best_label} (hold_sh={best_hold_sh:.3f}) ***")
+    print("Now running best config on 1d + 12h for final report...")
+
+    final_rep = al.study_weights(
+        f"VOL08-s{best_short}d-l{best_long}d",
+        make_target(best_short, best_long),
+        tfs=("1d", "12h")
+    )
+    print("\n=== FINAL REPORT ===")
+    print(al.fmt(final_rep))
+    print("JSON:", al.as_json(final_rep))