gahow/quant-trading
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

research: beta-hedged momentum (STOP) and 24-market futures TSMOM quick checks

Browse Source 
hedged_momentum: overnight_recovery_top10 hedged with QQQ futures at a
rolling 126d beta earns Sharpe 0.11; alpha vs QQQ on the Tiingo panel is
+5.3%/yr (t 1.10). futures_trend: Yahoo =F data is roughly usable (ES/NQ
corr 0.98, ZN/GC miss on settlement time), 24-market TSMOM Sharpe 0.44,
corr 0.22 with TrendRider: a crisis diversifier, not a core stream.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Gahow Wang
2026-10-08 22:45:21 -07:00
parent 8845f31b3b
commit f1c1210724
3 changed files with 333 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
CLAUDE.md
View File

@@ -82,6 +82,8 @@ uv run python -m research.lab.fetch_wiki_attention && PYTHONPATH=. uv run --with
PYTHONPATH=. uv run --with numba python -m research.lab.concentration # industry caps on the momentum sleeve
PYTHONPATH=. uv run python -m research.lab.universe_midcap # overnight_recovery_top10 on S&P 500 ∪ S&P 400 (preregistered; FAIL 2026-10-08)
PYTHONPATH=. uv run --with numba python -m research.lab.stream_portfolio # risk-budgeted book of momentum/lowrisk/TrendRider/ETF TSMOM/BTC streams, 2006-26 (preregistered; REJECT 2026-10-08)
PYTHONPATH=. uv run python -m research.lab.hedged_momentum # momentum with its beta hedged by index futures (quick check; STOP 2026-10-08)
PYTHONPATH=. uv run python -m research.lab.futures_trend [--no-proxy] # 24-market futures TSMOM on Yahoo =F + data checks (quick check; Sharpe 0.44, DATA 2026-10-08)
uv run python -m research.lab.fetch_us_fundamentals --universe mid # EDGAR PIT facts for S&P 400 ever-members (name-checked CIKs)
uv run python -m research.lab.fetch_us_fundamentals --universe small # the same for S&P 600 (IJR)
PYTHONPATH=. uv run python -m research.lab.midcap_sleeve [--only h1..h4] [--universe small] # S&P 400 sleeves vs IJH; S&P 600 replication vs IJR (all REJECT 2026-10-08)

186
research/lab/futures_trend.py Normal file
View File

@@ -0,0 +1,186 @@
"""
Quick check before building a futures pipeline: do free Yahoo continuous
futures (=F) support a backtest, and does a diversified futures trend book
(Moskowitz, Ooi & Pedersen 2012) add a better stream than the 8-ETF
long-only trend of research/lab/stream_portfolio.py (Sharpe 0.39)?
PYTHONPATH=. uv run python -m research.lab.futures_trend [--no-proxy]
Data: Yahoo front-month continuous contracts. They are not back-adjusted, so
each roll shows up as a jump between contracts. Data checks first:
tracking daily corr and annual drift of ES/NQ/ZN/GC vs SPY/QQQ/IEF/GLD
minus the T-bill (futures earn the excess return)
jumps days with |return| > 6x the 63-day vol, per market and year
Strategy (primary): 24 markets (4 equity, 4 rates, 6 FX, 3 metals,
4 energy, 3 grains). Position = sign(252-day return) x 40% / (EWMA-60 vol)
/ N, through t-1. Book return = T-bill + sum(w r), less 3 bp per unit of
notional traded. Jump days are kept as Yahoo prints them, and the
roll-cleaned variant (jumps zeroed) is reported next to it.
The decision rule is in PREREG and is written before the run; the script
refuses to overwrite results.json.
Result (2026-10-08): DATA. C1 and C2 fail; C3 passes.
- Tracking: ES 0.979 and NQ 0.978 pass. ZN vs IEF (0.914) and GC vs GLD
(0.888, drift +2.2%/yr) miss the bar. The likely cause is settlement
times (ZN 15:00, GC 13:30 ET vs the ETFs' 16:00 close), not bad prints.
- Roll jumps are rare: 0.0-0.7 days a year per market.
- tsmom24: 7.4% / Sharpe 0.44, vol 15%, MaxDD -34%, halves 0.34 / 0.56.
Correlation is 0.22 with TrendRider and 0.05 with momentum. Returns:
+24.5% in 2008, +30.9% in 2022, -24.8% in 2016.
- Roll-cleaned: 0.48 (0.34 / 0.65). Cleaning does not lift 2006-16, so
better data is unlikely to reach 0.5 in both halves.
- By sector: equity 0.44, metals 0.45, energy 0.27, grains 0.15, rates 0.11,
FX -0.01.
- Exploration afterwards (not preregistered, stream_portfolio blend):
- tsmom24 in place of the ETF xtrend: invvol5 1.07 -> 1.21, equal5 1.23 -> 1.30.
- Added to the three sleeves: 0.96-1.02 -> 1.05-1.07, MaxDD -32/-39% -> -20/-28%.
It is a diversifier (crisis years) with a low stand-alone Sharpe, not a core stream.
"""
from __future__ import annotations
import argparse
import json
import os
import pickle
import numpy as np
import pandas as pd
from research.lab.factor_scan import RESULTS
from research.lab.fetch_long_history import fetch_yahoo
from research.lab.levered_book import stats, tbill
OUT = os.path.join(RESULTS, "futures_trend")
STREAMS = os.path.join(RESULTS, "stream_portfolio", "streams.pkl")
MARKETS = {
"equity": ["ES=F", "NQ=F", "YM=F", "RTY=F"],
"rates": ["ZT=F", "ZF=F", "ZN=F", "ZB=F"],
"fx": ["6E=F", "6J=F", "6B=F", "6A=F", "6C=F", "6S=F"],
"metals": ["GC=F", "SI=F", "HG=F"],
"energy": ["CL=F", "NG=F", "HO=F", "RB=F"],
"grains": ["ZC=F", "ZW=F", "ZS=F"],
}
ALL = [t for v in MARKETS.values() for t in v]
TRACK = {"ES=F": "SPY", "NQ=F": "QQQ", "ZN=F": "IEF", "GC=F": "GLD"}
HALVES = [("2006-10-02", "2016-12-30"), ("2017-01-03", "2026-09-24")]
SPAN = (HALVES[0][0], HALVES[1][1])
COST = 0.0003
PREREG = {
"question": "Is free Yahoo =F data good enough, and does 24-market futures TSMOM beat the ETF trend stream?",
"primary": "tsmom24 (as printed by Yahoo)",
"data_ok_if": "daily corr >= 0.95 for ES/NQ/ZN/GC vs their ETF and |annual drift| <= 2%",
"continue_if_all": [
"C1 data_ok",
"C2 tsmom24 excess Sharpe >= 0.50 in both halves (2006-16, 2017-26)",
"C3 monthly corr of tsmom24 with TrendRider V10 <= 0.40",
],
"decision": {"CONTINUE": "buy roll-adjusted data (e.g. Norgate / CSI) and build a futures trend sleeve",
"STOP": "drop TODO 3 as a strategy source",
"DATA": "if C1 fails the strategy numbers are not trusted: decide on buying data alone"},
"halves": HALVES,
}
def jumps(r: pd.Series) -> pd.Series:
vol = r.rolling(63, min_periods=20).std().shift(1)
return r.abs() > 6 * vol
def tsmom(R: pd.DataFrame, rf: pd.Series, look: int = 252) -> tuple[pd.Series, pd.Series]:
lvl = (1 + R.fillna(0.0)).cumprod().where(R.notna())
sig = np.sign(lvl / lvl.shift(look) - 1)
vol = R.ewm(span=60, min_periods=60).std() * np.sqrt(252)
n = sig.notna().sum(axis=1).clip(lower=1)
W = (sig * (0.40 / vol)).div(n, axis=0).shift(1).fillna(0.0)
ret = rf.reindex(R.index) / 252 + (W * R.fillna(0.0)).sum(axis=1) - W.diff().abs().sum(axis=1) * COST
return ret, W.abs().sum(axis=1)
def main() -> None:
ap = argparse.ArgumentParser()
ap.add_argument("--no-proxy", action="store_true")
if not ap.parse_args().no_proxy:
from research.lab.fetch_us import _proxy_env
_proxy_env()
os.makedirs(OUT, exist_ok=True)
res_path = os.path.join(OUT, "results.json")
if os.path.exists(res_path):
raise SystemExit(f"{res_path} exists; this study does not overwrite results")
with open(os.path.join(OUT, "prereg.json"), "w") as fh:
json.dump(PREREG, fh, indent=2)
close, _ = fetch_yahoo(ALL, start="2000-01-01")
missing = [t for t in ALL if t not in close or close[t].dropna().empty]
print("missing:", missing)
with open(os.path.join("data", "research", "long_history.pkl"), "rb") as fh:
etf = pickle.load(fh)["close"]
days = etf["SPY"].dropna().loc["2000-01-01":].index
px = close.reindex(columns=[t for t in ALL if t not in missing]).reindex(days)
R = px.pct_change(fill_method=None)
rf = tbill(days)
res = {"data": {}, "jumps_per_year": {}}
print("\n=== tracking vs ETF (futures earn ETF return - T-bill)")
for f, e in TRACK.items():
if f in missing:
continue
a = R[f]
b = etf[e].reindex(days).pct_change(fill_method=None) - rf / 252
ok = a.notna() & b.notna()
corr = float(a[ok].corr(b[ok]))
drift = float((a[ok].mean() - b[ok].mean()) * 252)
res["data"][f] = {"etf": e, "corr": corr, "annual_drift": drift, "start": str(a.first_valid_index().date())}
print(f" {f:6s} vs {e:4s} from {a.first_valid_index().date()} corr {corr:.3f} drift {drift*100:5.1f}%/yr")
data_ok = all(v["corr"] >= 0.95 and abs(v["annual_drift"]) <= 0.02 for v in res["data"].values())
print("\n=== jump days (|r| > 6x 63d vol) per year, and first date")
J = R.apply(jumps)
for t in px.columns:
n = int(J[t].sum())
yrs = max(R[t].notna().sum() / 252, 1)
res["jumps_per_year"][t] = n / yrs
print(f" {t:6s} from {px[t].first_valid_index().date()} jumps {n:3d} ({n/yrs:4.1f}/yr)")
book, gross = tsmom(R, rf)
clean, _ = tsmom(R.mask(J, 0.0), rf)
groups = {g: tsmom(R[[t for t in v if t in R]], rf)[0] for g, v in MARKETS.items()}
S = pd.read_pickle(STREAMS)
out = {"tsmom24": book, "tsmom24_roll_cleaned": clean, **{f"tsmom_{g}": r for g, r in groups.items()}}
res["streams"] = {}
print("\n=== strategies (excess Sharpe; halves 2006-16 / 2017-26)")
m_tr = (1 + S["trend"].dropna()).resample("ME").prod() - 1
m_mo = (1 + S["momentum"].dropna()).resample("ME").prod() - 1
for k, r in {**out, "etf_xtrend": S["xtrend"].dropna(), "trend_v10": S["trend"].dropna()}.items():
r = r.loc[SPAN[0]:SPAN[1]]
s = stats(r, rf)
h = [float(stats(r.loc[a:b], rf)["sharpe"]) for a, b in HALVES]
m = (1 + r).resample("ME").prod() - 1
c_tr, c_mo = float(m.corr(m_tr)), float(m.corr(m_mo))
res["streams"][k] = {"cagr": float(s["cagr"]), "sharpe": float(s["sharpe"]), "vol": float(s["vol"]),
"mdd": float(s["mdd"]), "halves": h, "corr_trend": c_tr, "corr_momentum": c_mo}
print(f" {k:22s} cagr {s['cagr']*100:5.1f}% xsh {s['sharpe']:5.2f} vol {s['vol']*100:4.1f}% "
f"mdd {s['mdd']*100:6.1f}% halves {h[0]:5.2f} {h[1]:5.2f} corr trend {c_tr:5.2f} mom {c_mo:5.2f}")
print(f" avg gross leverage tsmom24 {gross.loc[SPAN[0]:].mean():.2f}")
yr = (1 + book.loc[SPAN[0]:SPAN[1]]).groupby(book.loc[SPAN[0]:SPAN[1]].index.year).prod() - 1
print("tsmom24 yearly %:", (yr * 100).round(1).to_dict())
res["yearly_tsmom24"] = {int(k): float(v) for k, v in yr.items()}
p = res["streams"]["tsmom24"]
checks = {"C1": data_ok, "C2": all(x >= 0.50 for x in p["halves"]), "C3": p["corr_trend"] <= 0.40}
decision = "CONTINUE" if all(checks.values()) else ("DATA" if not data_ok else "STOP")
res["verdict"] = {"checks": checks, "decision": decision}
print(f"\nchecks {checks} -> {decision}")
with open(res_path, "w") as fh:
json.dump(res, fh, indent=2)
pd.DataFrame(out).to_pickle(os.path.join(OUT, "returns.pkl"))
from research.lab.trials import append_trials
append_trials(os.path.join(RESULTS, "trials_us.jsonl"),
[{"name": f"futures_trend:{k}", **v} for k, v in res["streams"].items()],
{"study": "futures_trend", "role": "preregistered quick check; both halves previously inspected"})
if __name__ == "__main__":
main()

145
research/lab/hedged_momentum.py Normal file
View File

@@ -0,0 +1,145 @@
"""
Quick check before building a futures-hedge pipeline: is the momentum
sleeve's alpha, with its beta hedged by short index futures (MNQ / MES), a
stream worth holding next to TrendRider?
PYTHONPATH=. uv run --with numba python -m research.lab.stream_portfolio # once: streams.pkl
PYTHONPATH=. uv run python -m research.lab.hedged_momentum
Inputs: the cached daily streams of research/lab/stream_portfolio.py
(momentum = overnight_recovery_top10 on the PIT S&P 500, TrendRider V10 on
synthetic 3x ETFs). Index returns are QQQ / SPY total return from
long_history.pkl.
Hedge: the account holds the stocks with its own cash and is short beta_t
units of index futures; a short future earns -(index - T-bill). So
hedged_t = mom_t - beta_{t-1} (idx_t - rf_t) - |d beta| * 2 bp - 0.5%/yr / 252
beta_{t-1} is the 126-day OLS beta of momentum on the index, from returns
through t-1. The 0.5%/yr covers roll and basis slippage. Variants: QQQ (MNQ,
primary) or SPY (MES), rolling or fixed beta 1.
Books (monthly reset, research/lab/us_final.monthly_blend):
trend_raw 1/2 TrendRider + 1/2 momentum
trend_hedged 1/2 TrendRider + 1/2 hedged momentum
trend TrendRider alone
The decision rule is in PREREG and is written before the run; the script
refuses to overwrite results.json.
Result (2026-10-08): STOP. C1 and C2 both fail.
Excess Sharpe, 2006-10..2026-09, halves 2006-16 / 2017-26:
- momentum unhedged: 20.3% / 0.69 (0.59 / 0.81). Alpha vs QQQ is +5.3%/yr,
t 1.10, beta 1.04. On the Tiingo-overlaid PIT panel this alpha is about half
the +9.4% (t 1.86) the pre-backfill panel gave.
- hedged, QQQ, rolling beta: 1.8% / 0.11 (0.02 / 0.23), MaxDD -66%.
Beta drifts from 0.5 to 1.8 by year, and the 126-day estimate trails it.
- hedged, SPY, beta 1 (best variant): 10.3% / 0.48 (0.49 / 0.46).
- 1/2 TrendRider + 1/2 momentum: 1.01 (0.63 / 1.47).
With hedged momentum instead: 0.85 (0.37 / 1.42).
Interpretation: about 21% of the hedged stream's vol is idiosyncratic risk
from 10 names, against an alpha of a few percent. Hedging removes the beta
that paid for the momentum sleeve.
"""
from __future__ import annotations
import json
import os
import numpy as np
import pandas as pd
from research.lab.alpha import alpha_stats, bench_returns
from research.lab.factor_scan import RESULTS
from research.lab.levered_book import stats, tbill
from research.lab.us_final import monthly_blend
OUT = os.path.join(RESULTS, "hedged_momentum")
STREAMS = os.path.join(RESULTS, "stream_portfolio", "streams.pkl")
HALVES = [("2006-10-02", "2016-12-30"), ("2017-01-03", "2026-09-24")]
SPAN = (HALVES[0][0], HALVES[1][1])
TRADE, FRICTION = 0.0002, 0.005
PREREG = {
"question": "Is beta-hedged momentum (short QQQ futures, rolling 126d beta) a stream worth a futures pipeline?",
"primary": "hedged_qqq_roll126",
"continue_if_all": [
"C1 hedged_qqq_roll126 excess Sharpe >= 0.50 in both halves",
"C2 trend_hedged Sharpe >= trend_raw Sharpe + 0.10 in both halves",
],
"decision": {"CONTINUE": "build the futures hedge (MNQ) into the sleeve framework and preregister a book",
"STOP": "drop TODO 2; momentum stays an unhedged long sleeve"},
"halves": HALVES, "costs": {"hedge_trade": TRADE, "hedge_friction_per_year": FRICTION},
}
def rolling_beta(r: pd.Series, b: pd.Series, n: int = 126) -> pd.Series:
cov = r.rolling(n, min_periods=63).cov(b)
return (cov / b.rolling(n, min_periods=63).var()).shift(1)
def hedge(r: pd.Series, idx: pd.Series, rf: pd.Series, beta: pd.Series) -> pd.Series:
beta = beta.fillna(1.0)
return r - beta * (idx - rf / 252) - beta.diff().abs().fillna(0.0) * TRADE - beta * FRICTION / 252
def describe(r: pd.Series, rf: pd.Series, q: pd.Series) -> dict:
s = stats(r.loc[SPAN[0]:SPAN[1]], rf)
a = alpha_stats(r.loc[SPAN[0]:SPAN[1]], q, rf)
return {"cagr": float(s["cagr"]), "sharpe": float(s["sharpe"]), "vol": float(s["vol"]),
"mdd": float(s["mdd"]), "worst_yr": float(s["worst_yr"]),
"halves": [float(stats(r.loc[a_:b_], rf)["sharpe"]) for a_, b_ in HALVES],
"beta_qqq": float(a["beta"]), "alpha_qqq": float(a["alpha"]), "alpha_t": float(a["t"])}
def main() -> None:
os.makedirs(OUT, exist_ok=True)
res_path = os.path.join(OUT, "results.json")
if os.path.exists(res_path):
raise SystemExit(f"{res_path} exists; this study does not overwrite results")
with open(os.path.join(OUT, "prereg.json"), "w") as fh:
json.dump(PREREG, fh, indent=2)
R = pd.read_pickle(STREAMS).loc[SPAN[0]:SPAN[1]]
rf = tbill(R.index)
q, s = bench_returns(R.index, "QQQ"), bench_returns(R.index, "SPY")
mom = R["momentum"]
S = {"momentum": mom, "trend": R["trend"], "qqq": q}
for name, idx in (("qqq", q), ("spy", s)):
S[f"hedged_{name}_roll126"] = hedge(mom, idx, rf, rolling_beta(mom, idx))
S[f"hedged_{name}_beta1"] = hedge(mom, idx, rf, pd.Series(1.0, index=mom.index))
D = pd.DataFrame(S)
for k, h in (("trend_raw", "momentum"), ("trend_hedged", "hedged_qqq_roll126"),
("trend_hedged_spy", "hedged_spy_roll126")):
D[k] = monthly_blend(D[["trend", h]].fillna(0.0), {"trend": 0.5, h: 0.5})
res = {k: describe(D[k], rf, q) for k in D.columns}
beta = rolling_beta(mom, q)
res["beta_path"] = {str(y): float(v) for y, v in beta.groupby(beta.index.year).mean().items()}
for k, d in res.items():
if k == "beta_path":
continue
print(f"{k:22s} cagr {d['cagr']*100:5.1f}% xsh {d['sharpe']:5.2f} vol {d['vol']*100:4.1f}% "
f"mdd {d['mdd']*100:6.1f}% halves {d['halves'][0]:5.2f} {d['halves'][1]:5.2f} | "
f"vs QQQ beta {d['beta_qqq']:4.2f} alpha {d['alpha_qqq']*100:5.1f}% t {d['alpha_t']:4.2f}")
print("mean rolling beta to QQQ by year:", {k: round(v, 2) for k, v in res["beta_path"].items()})
m = (1 + D[["hedged_qqq_roll126", "trend", "momentum", "qqq"]]).resample("ME").prod() - 1
print("monthly corr:\n", m.corr().round(2).to_string())
yr = (1 + D[["momentum", "hedged_qqq_roll126", "trend_raw", "trend_hedged", "qqq"]]).groupby(D.index.year).prod() - 1
print((yr * 100).round(1).to_string())
h, tr, th = res["hedged_qqq_roll126"], res["trend_raw"], res["trend_hedged"]
checks = {"C1": all(x >= 0.50 for x in h["halves"]),
"C2": all(th["halves"][i] >= tr["halves"][i] + 0.10 for i in range(2))}
res["verdict"] = {"checks": checks, "decision": "CONTINUE" if all(checks.values()) else "STOP"}
print(f"checks {checks} -> {res['verdict']['decision']}")
with open(res_path, "w") as fh:
json.dump(res, fh, indent=2)
from research.lab.trials import append_trials
append_trials(os.path.join(RESULTS, "trials_us.jsonl"),
[{"name": f"hedged_momentum:{k}", **v} for k, v in res.items() if k not in ("beta_path", "verdict")],
{"study": "hedged_momentum", "role": "preregistered quick check; both halves previously inspected"})
if __name__ == "__main__":
main()