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 +1,103 @@
+"""RSK09 — Target-vol + floor/cap + trend gate.
+
+HYPOTHESIS: Long-flat TSMOM multi-horizon (like TP01), but with a hard exposure
+floor=0.2 and cap=1.5 (instead of raw [0, leverage_cap]) when trend is UP,
+and flat when trend is DOWN (same as TP01). The idea: smoother, more persistent
+exposure when in-trend avoids whipsaw from momentary vol spikes reducing position
+to near-zero, potentially improving risk-adjusted returns vs raw vol-target.
+
+Grid:
+  - vol_win_days: 20 or 30
+  - floor when long: 0.2 (fixed — the core of the hypothesis)
+  - cap when long: 1.5 (fixed — slightly higher than TP01's 2.0 but with floor)
+  TFs tested: 1d, 12h (total 4 backtests, within 6-cell limit)
+"""
+import sys
+sys.path.insert(0, "/opt/docker/PythagorasGoal/scripts/research/alt")
+import altlib as al
+import numpy as np
+
+
+def tsmom_direction(df, horizons_days=(21, 63, 126)):
+    """Multi-horizon TSMOM direction: sign of blend of returns over multiple horizons.
+    Returns +1 (trend up) or 0 (trend down/flat). Causal: uses close[i] vs close[i-k]."""
+    c = df["close"].values.astype(float)
+    bpd = al.bars_per_day(df)
+    scores = []
+    for h_days in horizons_days:
+        win = max(2, int(h_days * bpd))
+        ret = np.zeros(len(c))
+        ret[win:] = c[win:] / c[:-win] - 1.0
+        scores.append(np.sign(ret))
+    blend = np.mean(scores, axis=0)
+    # Long when majority of horizons agree (blend > 0), else flat
+    direction = np.where(blend > 0, 1.0, 0.0)
+    return direction
+
+
+def rsk09_target(df, vol_win_days=30, exposure_floor=0.2, exposure_cap=1.5,
+                 target_vol=0.20):
+    """RSK09: vol-targeted TSMOM with floor/cap clamp on long exposure.
+
+    When trend is UP:
+      - compute raw vol-target scalar (target_vol / realized_vol)
+      - clamp to [floor, cap] instead of [0, leverage_cap]
+      -> ensures we're never near-zero even in high-vol regimes,
+         but also never overleveraged
+    When trend is DOWN (or mixed): flat (0.0)
+    """
+    direction = tsmom_direction(df)  # 0 or 1
+
+    c = df["close"].values.astype(float)
+    bpd = al.bars_per_day(df)
+    bpy = bpd * 365.25
+    r = al.simple_returns(c)
+    vol = al.realized_vol(r, max(2, int(vol_win_days * bpd)), bpy)
+
+    # Raw vol-scalar (avoid div-by-zero)
+    scal = np.where((vol > 0) & np.isfinite(vol), target_vol / vol, 0.0)
+
+    # When in trend: clamp to [floor, cap]
+    # floor ensures we hold minimum exposure even in high-vol periods
+    # cap ensures we don't over-lever in low-vol periods
+    raw_exposure = np.clip(scal, exposure_floor, exposure_cap)
+
+    # Apply trend gate: long-flat
+    target = direction * raw_exposure
+    target = np.nan_to_num(target, nan=0.0)
+    return target
+
+
+# Small grid: vol_win_days x TF (2 params x 2 TFs = 4 total backtests)
+configs = [
+    {"vol_win_days": 20, "label": "vw20"},
+    {"vol_win_days": 30, "label": "vw30"},
+]
+
+best_rep = None
+best_score = -9999.0
+
+for cfg in configs:
+    name = f"RSK09-floor02-cap15-{cfg['label']}"
+    rep = al.study_weights(
+        name,
+        lambda df, c=cfg: rsk09_target(df, vol_win_days=c["vol_win_days"]),
+        tfs=("1d", "12h"),
+    )
+    # Score by min hold-out Sharpe across cells
+    cells = rep.get("cells", [])
+    if cells:
+        score = max((c.get("min_asset_holdout_sharpe", -9) for c in cells), default=-9)
+    else:
+        score = -9
+
+    print(f"\n=== Config: {cfg['label']} | score={score:.3f} ===")
+    print(al.fmt(rep))
+
+    if score > best_score:
+        best_score = score
+        best_rep = rep
+
+print("\n\n=== BEST CONFIG ===")
+print(al.fmt(best_rep))
+print("JSON:", al.as_json(best_rep))