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 +1,174 @@
|
||||
"""XAS01 — ETH/BTC ratio z-score reversion strategy.
|
||||
|
||||
IDEA: The ETH/BTC ratio (price ratio) exhibits mean-reversion. When the z-score of the
|
||||
ratio falls below a threshold (ETH is cheap relative to BTC), go long the ratio
|
||||
(long ETH, short BTC). When z-score rises above threshold, go short the ratio.
|
||||
|
||||
IMPLEMENTATION:
|
||||
- Build ratio = ETH_close / BTC_close on aligned timestamps (inner join).
|
||||
- Compute rolling z-score of the log-ratio over a lookback window.
|
||||
- Position: +1 when z < -threshold (long ratio), -1 when z > +threshold (short ratio), 0 otherwise.
|
||||
- The SPREAD P&L is: pos * (ETH_return - BTC_return) per bar.
|
||||
- We use eval_weights on a synthetic series where close = ratio, so that simple_returns(ratio)
|
||||
gives the ratio return which equals ETH_return - BTC_return (approximately for log returns).
|
||||
- Actually: ratio_return = ETH/BTC new / ETH/BTC old - 1 ≈ r_ETH - r_BTC (log approximation)
|
||||
But for precise spread return: r_spread = r_ETH - r_BTC exactly in log space.
|
||||
We construct a synthetic df with close=ratio so eval_weights gives us ratio simple returns.
|
||||
|
||||
GRID (4 configs, 2 TFs = 8 backtests — within limit):
|
||||
lookback_days: [20, 60]
|
||||
threshold: [1.5, 2.0]
|
||||
|
||||
We pick the best config (highest min holdout sharpe) and report that.
|
||||
"""
|
||||
|
||||
import sys
|
||||
import json
|
||||
import numpy as np
|
||||
import pandas as pd
|
||||
|
||||
sys.path.insert(0, "/opt/docker/PythagorasGoal/scripts/research/alt")
|
||||
import altlib as al
|
||||
|
||||
# ===========================================================================
|
||||
# Helper: build synthetic ratio df aligned between BTC and ETH
|
||||
# ===========================================================================
|
||||
|
||||
def build_ratio_df(tf: str) -> pd.DataFrame:
|
||||
"""Merge BTC and ETH on timestamp (inner join), build close=ETH/BTC ratio."""
|
||||
btc = al.get("BTC", tf)[["timestamp", "datetime", "close"]].rename(columns={"close": "btc"})
|
||||
eth = al.get("ETH", tf)[["timestamp", "close"]].rename(columns={"close": "eth"})
|
||||
merged = pd.merge(btc, eth, on="timestamp", how="inner").reset_index(drop=True)
|
||||
merged["close"] = merged["eth"] / merged["btc"]
|
||||
# Add stub OHLCV columns so eval_weights works (it only needs close)
|
||||
merged["open"] = merged["close"]
|
||||
merged["high"] = merged["close"]
|
||||
merged["low"] = merged["close"]
|
||||
merged["volume"] = 1.0
|
||||
return merged[["timestamp", "datetime", "open", "high", "low", "close", "volume"]]
|
||||
|
||||
|
||||
# ===========================================================================
|
||||
# Strategy: z-reversion on log(ratio)
|
||||
# ===========================================================================
|
||||
|
||||
def make_target(lookback_days: int, threshold: float, vol_tgt: bool = True):
|
||||
"""Return a target_fn(df) for eval_weights on the ratio df."""
|
||||
def target_fn(df: pd.DataFrame) -> np.ndarray:
|
||||
c = df["close"].values.astype(float)
|
||||
log_ratio = np.log(c) # log(ETH/BTC)
|
||||
|
||||
bpd = al.bars_per_day(df)
|
||||
win = max(2, lookback_days * bpd)
|
||||
|
||||
z = al.zscore(log_ratio, win)
|
||||
|
||||
# Mean-reversion: short when z > threshold (ratio overbought), long when z < -threshold
|
||||
direction = np.where(z < -threshold, 1.0,
|
||||
np.where(z > threshold, -1.0, 0.0))
|
||||
|
||||
if vol_tgt:
|
||||
# Vol-target the spread position
|
||||
pos = al.vol_target(direction, df, target_vol=0.20, vol_win_days=30, leverage_cap=2.0)
|
||||
else:
|
||||
pos = direction.astype(float)
|
||||
|
||||
pos = np.nan_to_num(pos, nan=0.0)
|
||||
return pos
|
||||
|
||||
return target_fn
|
||||
|
||||
|
||||
# ===========================================================================
|
||||
# Run grid: 2 lookbacks x 2 thresholds = 4 configs; 2 TFs = 8 backtests
|
||||
# ===========================================================================
|
||||
|
||||
GRID = [
|
||||
{"lookback_days": 20, "threshold": 1.5},
|
||||
{"lookback_days": 20, "threshold": 2.0},
|
||||
{"lookback_days": 60, "threshold": 1.5},
|
||||
{"lookback_days": 60, "threshold": 2.0},
|
||||
]
|
||||
|
||||
TFS = ("1d", "12h")
|
||||
|
||||
best_rep = None
|
||||
best_score = -999.0
|
||||
best_label = ""
|
||||
|
||||
print("=== XAS01: ETH/BTC ratio z-reversion ===")
|
||||
print(f"Grid: {len(GRID)} configs x {len(TFS)} TFs = {len(GRID)*len(TFS)} backtests")
|
||||
print()
|
||||
|
||||
for params in GRID:
|
||||
lb = params["lookback_days"]
|
||||
thr = params["threshold"]
|
||||
name = f"XAS01-lb{lb}-thr{thr}"
|
||||
print(f"--- {name} ---")
|
||||
|
||||
# We need a custom study_weights that uses ratio df instead of per-asset dfs
|
||||
# Build ratio df for each TF, run eval_weights on it
|
||||
cells = []
|
||||
for tf in TFS:
|
||||
try:
|
||||
ratio_df = build_ratio_df(tf)
|
||||
tgt_fn = make_target(lb, thr, vol_tgt=True)
|
||||
tgt = tgt_fn(ratio_df)
|
||||
|
||||
# Eval with fee sweep
|
||||
base = al.eval_weights(ratio_df, tgt, fee_side=al.FEE_SIDE)
|
||||
sweep = {f"{2*f*100:.2f}%RT": al.eval_weights(ratio_df, tgt, fee_side=f)["full"]["sharpe"]
|
||||
for f in al.FEE_SWEEP}
|
||||
fee_ok = sweep.get("0.20%RT", -9) > 0
|
||||
|
||||
full = base["full"]
|
||||
hold = base["holdout"]
|
||||
yearly = base["yearly"]
|
||||
|
||||
print(f" TF={tf}: full Sh={full['sharpe']:+.3f}, DD={full['maxdd']*100:.1f}%,"
|
||||
f" hold Sh={hold.get('sharpe', 0):+.3f}, feeOK={fee_ok}")
|
||||
print(f" fee sweep: {sweep}")
|
||||
|
||||
cells.append(dict(
|
||||
tf=tf,
|
||||
per_asset={"RATIO": dict(full=full, holdout=hold, tim=base["time_in_market"],
|
||||
turnover=base["turnover_per_year"], fee_sweep=sweep,
|
||||
yearly=yearly)},
|
||||
min_asset_full_sharpe=round(full["sharpe"], 3),
|
||||
min_asset_holdout_sharpe=round(hold.get("sharpe", 0.0), 3),
|
||||
full_sharpe=round(full["sharpe"], 3),
|
||||
fee_survives=fee_ok,
|
||||
))
|
||||
except Exception as e:
|
||||
print(f" TF={tf}: ERROR {e}")
|
||||
|
||||
# Score = best holdout sharpe across TFs
|
||||
score = max((c["min_asset_holdout_sharpe"] for c in cells), default=-999.0)
|
||||
if score > best_score:
|
||||
best_score = score
|
||||
best_label = name
|
||||
# Build a "rep" compatible with al.fmt
|
||||
# We adapt it to show ratio as both BTC and ETH (same series)
|
||||
adapted_cells = []
|
||||
for c in cells:
|
||||
ratio_res = c["per_asset"]["RATIO"]
|
||||
adapted_cells.append(dict(
|
||||
tf=c["tf"],
|
||||
per_asset={
|
||||
"BTC": ratio_res, # spread result attributed to both
|
||||
"ETH": ratio_res,
|
||||
},
|
||||
min_asset_full_sharpe=c["min_asset_full_sharpe"],
|
||||
min_asset_holdout_sharpe=c["min_asset_holdout_sharpe"],
|
||||
full_sharpe=c["full_sharpe"],
|
||||
fee_survives=c["fee_survives"],
|
||||
))
|
||||
best_rep = dict(name=name, kind="weights", cells=adapted_cells,
|
||||
verdict=al._verdict(adapted_cells))
|
||||
print()
|
||||
|
||||
print()
|
||||
print("=== BEST CONFIG:", best_label, "===")
|
||||
print(al.fmt(best_rep))
|
||||
print()
|
||||
print("JSON:", al.as_json(best_rep))
|
||||
Reference in New Issue
Block a user