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 +1,202 @@
|
||||
"""XAS04 — Lead-lag BTC->ETH
|
||||
|
||||
HYPOTHESIS: BTC returns lead ETH returns. ETH position = sign of BTC lagged return.
|
||||
We evaluate on ETH series with BTC-derived signals. BTC and ETH data must be at the SAME
|
||||
timeframe so that timestamp alignment is exact (no cross-TF look-ahead).
|
||||
|
||||
CAUSAL GUARANTEE:
|
||||
- BTC and ETH are loaded at the SAME timeframe (1d or 12h).
|
||||
- We align BTC to ETH by merging on common timestamps (inner/exact match).
|
||||
- target[i] uses BTC close[i] (same bar as ETH close[i]) -> held during bar i+1.
|
||||
- The altlib eval_weights shift handles the i->i+1 transition.
|
||||
- No cross-TF artifacts.
|
||||
|
||||
CRITICAL BUG AVOIDED: loading BTC at 1d and aligning to ETH at 12h creates look-ahead
|
||||
because the 1d bar timestamp (midnight) has the day's FINAL close, so the midnight 12h bar
|
||||
gets the full day's close which is the FUTURE relative to the midnight price.
|
||||
FIX: always use the same TF for both BTC and ETH.
|
||||
|
||||
CONFIGS TESTED:
|
||||
1. Sign of BTC 1-bar return (pure lag-1 momentum applied to ETH)
|
||||
2. BTC EMA5/20 cross -> ETH direction (BTC trend applied to ETH)
|
||||
3. BTC TSMOM multi-horizon (30/90/180d) -> ETH direction
|
||||
4. Blend: require BTC lag-1 AND BTC EMA trend to agree before entering ETH
|
||||
"""
|
||||
import sys
|
||||
sys.path.insert(0, "/opt/docker/PythagorasGoal/scripts/research/alt")
|
||||
import altlib as al
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
|
||||
def _align_same_tf(eth_df: pd.DataFrame, btc_df: pd.DataFrame) -> np.ndarray:
|
||||
"""Align BTC close to ETH timestamps using exact same-TF merge.
|
||||
Both DataFrames have the SAME timeframe, so merge on timestamp directly.
|
||||
Returns BTC close values aligned to ETH bar indices, NaN where missing."""
|
||||
eth_ts = eth_df["timestamp"].astype("int64")
|
||||
btc_df2 = btc_df[["timestamp", "close"]].copy()
|
||||
btc_df2["timestamp"] = btc_df2["timestamp"].astype("int64")
|
||||
btc_df2 = btc_df2.rename(columns={"close": "btc_close"})
|
||||
|
||||
merged = eth_ts.to_frame().merge(btc_df2, on="timestamp", how="left")
|
||||
return merged["btc_close"].values.astype(float)
|
||||
|
||||
|
||||
def make_target_lag1(tf: str):
|
||||
"""Config 1: Sign of BTC 1-bar return -> ETH vol-targeted position."""
|
||||
btc_df = al.get("BTC", tf)
|
||||
|
||||
def target_fn(df):
|
||||
# Align BTC close to ETH at same TF
|
||||
btc_c = _align_same_tf(df, btc_df)
|
||||
n = len(df)
|
||||
|
||||
# BTC 1-bar lagged return: r[i] = BTC_close[i] / BTC_close[i-1] - 1
|
||||
btc_ret = np.zeros(n)
|
||||
btc_ret[1:] = btc_c[1:] / btc_c[:-1] - 1.0
|
||||
|
||||
# Direction = sign of BTC return
|
||||
direction = np.sign(btc_ret)
|
||||
direction[~np.isfinite(direction)] = 0.0
|
||||
direction[:5] = 0.0 # warmup
|
||||
|
||||
# Vol-target position on ETH
|
||||
return al.vol_target(direction, df, target_vol=0.20, vol_win_days=30, leverage_cap=2.0)
|
||||
|
||||
return target_fn
|
||||
|
||||
|
||||
def make_target_ema(tf: str, ema_fast: int = 5, ema_slow: int = 20):
|
||||
"""Config 2: BTC EMA cross -> ETH direction (long-flat)."""
|
||||
btc_df = al.get("BTC", tf)
|
||||
|
||||
def target_fn(df):
|
||||
btc_c = _align_same_tf(df, btc_df)
|
||||
n = len(df)
|
||||
|
||||
fast = al.ema(btc_c, ema_fast)
|
||||
slow = al.ema(btc_c, ema_slow)
|
||||
|
||||
direction = np.where(fast > slow, 1.0, 0.0)
|
||||
direction[~np.isfinite(direction)] = 0.0
|
||||
direction[:ema_slow + 5] = 0.0 # warmup
|
||||
|
||||
return al.vol_target(direction, df, target_vol=0.20, vol_win_days=30, leverage_cap=2.0)
|
||||
|
||||
return target_fn
|
||||
|
||||
|
||||
def make_target_tsmom(tf: str):
|
||||
"""Config 3: BTC TSMOM multi-horizon (30/90/180 day) -> ETH direction (long-flat)."""
|
||||
btc_df = al.get("BTC", tf)
|
||||
|
||||
def target_fn(df):
|
||||
btc_c = _align_same_tf(df, btc_df)
|
||||
n = len(df)
|
||||
bpd = al.bars_per_day(df)
|
||||
|
||||
signal = np.zeros(n)
|
||||
for days in (30, 90, 180):
|
||||
h = int(days * bpd)
|
||||
if h < n:
|
||||
s = np.full(n, np.nan)
|
||||
valid = ~np.isnan(btc_c) & (np.roll(btc_c, h) != 0)
|
||||
s[h:] = np.sign(btc_c[h:] / btc_c[:-h] - 1.0)
|
||||
signal = signal + np.nan_to_num(s)
|
||||
|
||||
direction = np.clip(np.sign(signal), 0, None) # long-flat
|
||||
direction[:int(180 * bpd) + 5] = 0.0
|
||||
|
||||
return al.vol_target(direction, df, target_vol=0.20, vol_win_days=30, leverage_cap=2.0)
|
||||
|
||||
return target_fn
|
||||
|
||||
|
||||
def make_target_blend(tf: str, lag: int = 1, ema_span: int = 10):
|
||||
"""Config 4: Blend BTC lag-1 sign + BTC EMA trend; enter ETH only when both agree."""
|
||||
btc_df = al.get("BTC", tf)
|
||||
|
||||
def target_fn(df):
|
||||
btc_c = _align_same_tf(df, btc_df)
|
||||
n = len(df)
|
||||
|
||||
# Signal 1: BTC 1-bar lag sign
|
||||
btc_ret = np.zeros(n)
|
||||
btc_ret[lag:] = btc_c[lag:] / btc_c[:-lag] - 1.0
|
||||
sig1 = np.sign(btc_ret)
|
||||
sig1[:lag + 3] = 0.0
|
||||
|
||||
# Signal 2: BTC EMA momentum
|
||||
fast = al.ema(btc_c, ema_span)
|
||||
slow = al.ema(btc_c, ema_span * 4)
|
||||
sig2 = np.where(fast > slow, 1.0, 0.0)
|
||||
sig2[:ema_span * 4 + 5] = 0.0
|
||||
|
||||
# Both must agree (long ETH only when BTC shows positive momentum AND positive lag)
|
||||
direction = np.where((sig1 > 0) & (sig2 > 0), 1.0, 0.0)
|
||||
direction[~np.isfinite(direction)] = 0.0
|
||||
|
||||
return al.vol_target(direction, df, target_vol=0.20, vol_win_days=30, leverage_cap=2.0)
|
||||
|
||||
return target_fn
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
print("XAS04 — Lead-lag BTC->ETH (same-TF, causal alignment)")
|
||||
print("=" * 70)
|
||||
|
||||
# We test on both 1d and 12h but use SAME TF for BTC signal source.
|
||||
# This avoids the cross-TF look-ahead artifact.
|
||||
# The study_weights will apply the function to BOTH BTC and ETH at each TF.
|
||||
# For BTC asset: BTC signal applied to BTC (degenerate = BTC follows BTC lag)
|
||||
# For ETH asset: BTC signal applied to ETH (the actual lead-lag hypothesis)
|
||||
|
||||
# Config 1: pure lag-1 BTC signal -> run on 1d and 12h
|
||||
print("\n--- C1: Lag-1 BTC return sign -> ETH (1d) ---")
|
||||
rep_c1_1d = al.study_weights(
|
||||
"XAS04-C1-lag1-1d",
|
||||
make_target_lag1("1d"),
|
||||
tfs=("1d",)
|
||||
)
|
||||
print(al.fmt(rep_c1_1d))
|
||||
|
||||
print("\n--- C1: Lag-1 BTC return sign -> ETH (12h) ---")
|
||||
rep_c1_12h = al.study_weights(
|
||||
"XAS04-C1-lag1-12h",
|
||||
make_target_lag1("12h"),
|
||||
tfs=("12h",)
|
||||
)
|
||||
print(al.fmt(rep_c1_12h))
|
||||
|
||||
print("\n--- C2: BTC EMA5/20 cross -> ETH (1d) ---")
|
||||
rep_c2 = al.study_weights(
|
||||
"XAS04-C2-ema5x20-1d",
|
||||
make_target_ema("1d", 5, 20),
|
||||
tfs=("1d",)
|
||||
)
|
||||
print(al.fmt(rep_c2))
|
||||
|
||||
print("\n--- C3: BTC TSMOM -> ETH (1d) ---")
|
||||
rep_c3 = al.study_weights(
|
||||
"XAS04-C3-tsmom-1d",
|
||||
make_target_tsmom("1d"),
|
||||
tfs=("1d",)
|
||||
)
|
||||
print(al.fmt(rep_c3))
|
||||
|
||||
print("\n--- C4: BTC blend lag+ema -> ETH (1d) ---")
|
||||
rep_c4 = al.study_weights(
|
||||
"XAS04-C4-blend-1d",
|
||||
make_target_blend("1d", lag=1, ema_span=10),
|
||||
tfs=("1d",)
|
||||
)
|
||||
print(al.fmt(rep_c4))
|
||||
|
||||
# Identify best config
|
||||
all_reps = [rep_c1_1d, rep_c1_12h, rep_c2, rep_c3, rep_c4]
|
||||
best_rep = max(all_reps, key=lambda r: r["verdict"].get("best_holdout_sharpe") or -999)
|
||||
|
||||
print("\n" + "=" * 70)
|
||||
print(f"BEST CONFIG: {best_rep['name']} -> {best_rep['verdict']['grade']}")
|
||||
print(al.fmt(best_rep))
|
||||
print("\nJSON:", al.as_json(best_rep))
|
||||
Reference in New Issue
Block a user