chore: untrack generated data; add momentum + fundamentals experiment
- Stop tracking data/sp500_history.json (rebuilt daily from Wikipedia by universe_history.load_sp500_history) and ignore data/books/ (per-machine live book state, like data/trader_*.json). - research/lab/momentum_fundamentals.py: financial-statement vetoes and confirmations on overnight_recovery_top10 (2011-16 design, 2017-26 holdout). No return gain; re-sorting the momentum top 30 by fundamental momentum lowers MaxDD at equal return. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>
This commit is contained in:
2
.gitignore
vendored
2
.gitignore
vendored
@@ -18,9 +18,11 @@ data/*.png
|
|||||||
|
|
||||||
# Universe caches — re-fetched daily
|
# Universe caches — re-fetched daily
|
||||||
data/universe_*.json
|
data/universe_*.json
|
||||||
|
data/sp500_history.json
|
||||||
|
|
||||||
# Trader state — per-machine, regenerated by auto/simulate
|
# Trader state — per-machine, regenerated by auto/simulate
|
||||||
data/trader_*.json
|
data/trader_*.json
|
||||||
|
data/books/
|
||||||
|
|
||||||
# Factor attribution output and cached factors
|
# Factor attribution output and cached factors
|
||||||
data/attribution_*/
|
data/attribution_*/
|
||||||
|
|||||||
@@ -61,6 +61,7 @@ PYTHONPATH=. uv run python -m research.lab.diversifiers # multi
|
|||||||
PYTHONPATH=. uv run python -m research.lab.us_target50 # the book 2006-2026 and the leverage 50%/yr needs
|
PYTHONPATH=. uv run python -m research.lab.us_target50 # the book 2006-2026 and the leverage 50%/yr needs
|
||||||
PYTHONPATH=. uv run --with numba python -m research.lab.qqq_scan # alpha vs QQQ of every sleeve / factor / mega-cap variant
|
PYTHONPATH=. uv run --with numba python -m research.lab.qqq_scan # alpha vs QQQ of every sleeve / factor / mega-cap variant
|
||||||
PYTHONPATH=. uv run --with numba python -m research.lab.beat_qqq # sleeves picked on 2006-16, judged 2017-26; odds of beating QQQ
|
PYTHONPATH=. uv run --with numba python -m research.lab.beat_qqq # sleeves picked on 2006-16, judged 2017-26; odds of beating QQQ
|
||||||
|
PYTHONPATH=. uv run --with numba python -m research.lab.momentum_fundamentals [--sensitivity] # financial-statement vetoes/confirmations on overnight_recovery_top10
|
||||||
uv run python -m research.lab.fetch_insider && PYTHONPATH=. uv run python -m research.lab.insider # SEC Form 4 insider factors
|
uv run python -m research.lab.fetch_insider && PYTHONPATH=. uv run python -m research.lab.insider # SEC Form 4 insider factors
|
||||||
PYTHONPATH=. uv run python -m research.lab.fetch_lazy_prices && PYTHONPATH=. uv run python -m research.lab.lazy_prices # 10-K/10-Q text similarity, 8-K earnings dates, SIC
|
PYTHONPATH=. uv run python -m research.lab.fetch_lazy_prices && PYTHONPATH=. uv run python -m research.lab.lazy_prices # 10-K/10-Q text similarity, 8-K earnings dates, SIC
|
||||||
PYTHONPATH=. uv run --with numba python -m research.lab.earnings # earnings-announcement return (needs sec_8k_earnings.csv)
|
PYTHONPATH=. uv run --with numba python -m research.lab.earnings # earnings-announcement return (needs sec_8k_earnings.csv)
|
||||||
@@ -118,7 +119,7 @@ uv run python -m unittest discover tests
|
|||||||
|
|
||||||
- Strategies output **weights**, not buy/sell signals. Weights are normalized so each row sums to 1 (or 0 during warm-up).
|
- Strategies output **weights**, not buy/sell signals. Weights are normalized so each row sums to 1 (or 0 during warm-up).
|
||||||
- Warm-up periods: each strategy zeros out its first N rows where indicators are undefined.
|
- Warm-up periods: each strategy zeros out its first N rows where indicators are undefined.
|
||||||
- `data/` contains persistent price CSVs, universe JSON caches, and trader state files, plus `data/precomputed/` (precomputed backtests), `data/web/` (watchlist, per-stock cache, backtest result cache) and `data/research/` (research panels, ~2 GB). Not checked into git (only `data/.gitkeep`, `data/sp500_history.json` and `data/csi300_history.json` are tracked).
|
- `data/` contains persistent price CSVs, universe JSON caches, and trader state files, plus `data/precomputed/` (precomputed backtests), `data/web/` (watchlist, per-stock cache, backtest result cache) and `data/research/` (research panels, ~2 GB). Not checked into git (only `data/.gitkeep` and `data/csi300_history.json` are tracked; `data/sp500_history.json` is rebuilt from Wikipedia on first use).
|
||||||
- Python 3.12+, managed with `uv`.
|
- Python 3.12+, managed with `uv`.
|
||||||
- Do not show matplotlib figures when running backtests; use `--no-plot`.
|
- Do not show matplotlib figures when running backtests; use `--no-plot`.
|
||||||
|
|
||||||
|
|||||||
File diff suppressed because one or more lines are too long
140
research/lab/momentum_fundamentals.py
Normal file
140
research/lab/momentum_fundamentals.py
Normal file
@@ -0,0 +1,140 @@
|
|||||||
|
"""
|
||||||
|
Does financial-statement analysis improve overnight_recovery_top10?
|
||||||
|
|
||||||
|
PYTHONPATH=. uv run --with numba python -m research.lab.momentum_fundamentals [--sensitivity]
|
||||||
|
|
||||||
|
Averaging fundamental ranks into the momentum score was already tried and
|
||||||
|
hurts (fund_eval.py: Sharpe 0.80 -> 0.63; momentum's alpha lives in its most
|
||||||
|
extreme names). So every variant here keeps momentum's ranking and uses the
|
||||||
|
statements only to veto or to confirm, the way an analyst would screen a
|
||||||
|
momentum list. Fundamentals are point-in-time EDGAR values (first filing,
|
||||||
|
usable the day after it; research/lab/fundamentals.py); they cover ~90% of
|
||||||
|
S&P 500 members from 2011, so the test runs 2011-2026 (design 2011-2016,
|
||||||
|
holdout 2017-2026). A name without data is never vetoed.
|
||||||
|
|
||||||
|
Variants, fixed before looking at results (7 trials):
|
||||||
|
profitable veto trailing losses: ROE < 0 or operating cash flow < 0
|
||||||
|
quality_veto veto the bottom 20% of a sector-relative quality score
|
||||||
|
(mean rank of ROE, CFO/assets, accruals, gross profit/assets)
|
||||||
|
earnings_confirm keep only a positive latest earnings surprise (SUE > 0):
|
||||||
|
price momentum backed by earnings momentum (Chan, Jegadeesh &
|
||||||
|
Lakonishok 1996; Novy-Marx 2015)
|
||||||
|
revenue_confirm keep only positive year-on-year revenue growth
|
||||||
|
fund_momentum top 30 by momentum, then the 10 with the best fundamental
|
||||||
|
momentum (mean rank of SUE, revenue SUE, gross-margin change,
|
||||||
|
ROA change)
|
||||||
|
red_flags veto the worst decile of accruals (Sloan 1996) or of share
|
||||||
|
issuance (dilution)
|
||||||
|
fscore veto a Piotroski-style F-score below 4
|
||||||
|
Then a neighbourhood check of the two that looked best (``--sensitivity``):
|
||||||
|
fund_momentum with the top 20 / 40 instead of 30, quality_veto at 10% / 30%.
|
||||||
|
Base: overnight_mom + rec_126, top 10, every 10 days, buffer 2 (the live
|
||||||
|
overnight_recovery_top10). Each variant is compared with the base on the
|
||||||
|
same days (paired difference, Newey-West t) and against QQQ.
|
||||||
|
"""
|
||||||
|
|
||||||
|
from __future__ import annotations
|
||||||
|
|
||||||
|
import dataclasses
|
||||||
|
import json
|
||||||
|
import os
|
||||||
|
import warnings
|
||||||
|
|
||||||
|
import numpy as np
|
||||||
|
import pandas as pd
|
||||||
|
|
||||||
|
from research.lab import fundamentals as fu
|
||||||
|
from research.lab.alpha import _nw_se, alpha_stats, bench_returns
|
||||||
|
from research.lab.factor_scan import RESULTS, load
|
||||||
|
from research.lab.report import stats
|
||||||
|
from research.lab.rotation import factor_ranks
|
||||||
|
from research.lab.us_sweep import _blend, log_trials, run, sector_rank
|
||||||
|
|
||||||
|
warnings.filterwarnings("ignore")
|
||||||
|
SPAN = ("2011-01-03", "2026-09-24")
|
||||||
|
DESIGN = ("2011-01-03", "2016-12-31")
|
||||||
|
HOLDOUT = ("2017-01-01", "2026-09-24")
|
||||||
|
|
||||||
|
|
||||||
|
def variants(p, sensitivity: bool = False) -> dict[str, tuple[pd.DataFrame, pd.DataFrame, float]]:
|
||||||
|
"""name -> (score, eligible mask, rank buffer)."""
|
||||||
|
pr = factor_ranks(p, "us")
|
||||||
|
F = fu.load(p)
|
||||||
|
mom = _blend(pr["overnight_mom"], pr["rec_126"]).where(p.member)
|
||||||
|
m = p.member
|
||||||
|
ok = lambda veto: m & ~veto.fillna(False) # noqa: E731
|
||||||
|
pct = lambda x: x.where(m).rank(axis=1, pct=True) # noqa: E731
|
||||||
|
quality = _blend(*(sector_rank(F[k], p) for k in ("roe", "cfo_a", "accruals", "gp_a")))
|
||||||
|
fmom = _blend(*(pct(F[k]) for k in ("sue", "sue_rev", "gm_change", "roa_change")))
|
||||||
|
mom_rank = mom.rank(axis=1, ascending=False)
|
||||||
|
top30 = mom_rank <= 30
|
||||||
|
if sensitivity:
|
||||||
|
return {
|
||||||
|
"base": (mom, m, 2.0),
|
||||||
|
"fund_momentum top20": (fmom.fillna(0.5).where(mom_rank <= 20), m, 1.0),
|
||||||
|
"fund_momentum top30": (fmom.fillna(0.5).where(top30), m, 1.0),
|
||||||
|
"fund_momentum top40": (fmom.fillna(0.5).where(mom_rank <= 40), m, 1.0),
|
||||||
|
"quality_veto 10%": (mom, ok(quality < 0.1), 2.0),
|
||||||
|
"quality_veto 30%": (mom, ok(quality < 0.3), 2.0),
|
||||||
|
}
|
||||||
|
return {
|
||||||
|
"base": (mom, m, 2.0),
|
||||||
|
"profitable": (mom, ok((F["roe"] < 0) | (F["cfo_a"] < 0)), 2.0),
|
||||||
|
"quality_veto": (mom, ok(quality < 0.2), 2.0),
|
||||||
|
"earnings_confirm": (mom, ok(F["sue"] <= 0), 2.0),
|
||||||
|
"revenue_confirm": (mom, ok(F["rev_growth"] <= 0), 2.0),
|
||||||
|
"fund_momentum": (fmom.fillna(0.5).where(top30), m, 1.0), # no data: neutral
|
||||||
|
"red_flags": (mom, ok((pct(F["accruals"]) < 0.1) | (pct(F["net_issuance"]) < 0.1)), 2.0),
|
||||||
|
"fscore": (mom, ok(F["fscore"] < 4), 2.0),
|
||||||
|
}
|
||||||
|
|
||||||
|
|
||||||
|
def paired_t(d: pd.Series) -> float:
|
||||||
|
d = d.dropna()
|
||||||
|
X = np.ones((len(d), 1))
|
||||||
|
se = _nw_se(X, (d - d.mean()).to_numpy())[0]
|
||||||
|
return float(d.mean() / se)
|
||||||
|
|
||||||
|
|
||||||
|
def main() -> None:
|
||||||
|
import sys
|
||||||
|
sens = "--sensitivity" in sys.argv
|
||||||
|
p, _ = load("us")
|
||||||
|
b = bench_returns(p.dates)
|
||||||
|
R, rows = {}, []
|
||||||
|
for name, (score, elig, buf) in variants(p, sens).items():
|
||||||
|
pv = dataclasses.replace(p, member=elig)
|
||||||
|
st = run(pv, score.where(elig), [SPAN], top_n=10, buffer=buf, every=10)
|
||||||
|
R[name] = st["_rets"]
|
||||||
|
rows.append({"score": f"momfund:{name}", **{k: v for k, v in st.items() if not k.startswith("_")}})
|
||||||
|
log_trials(rows[1:], "momentum_fundamentals" + ("_sensitivity" if sens else ""))
|
||||||
|
R = pd.DataFrame(R)
|
||||||
|
out = {}
|
||||||
|
print(f"{'':18s}{'2011-2026':>34s} | {'2011-2016 (design)':>34s} | {'2017-2026 (holdout)':>34s} | vs base 17-26")
|
||||||
|
for name in R.columns:
|
||||||
|
r = R[name]
|
||||||
|
cells, rec = [], {}
|
||||||
|
for label, (a, z) in (("full", SPAN), ("design", DESIGN), ("holdout", HOLDOUT)):
|
||||||
|
x = r.loc[a:z]
|
||||||
|
s = stats((1 + x).cumprod())
|
||||||
|
al = alpha_stats(x, b)
|
||||||
|
rec[label] = {"cagr": s["cagr"], "sharpe": s["sharpe"], "mdd": s["mdd"], "alpha": al["alpha"], "t": al["t"]}
|
||||||
|
cells.append(f"{s['cagr']*100:5.1f}% sh {s['sharpe']:.2f} mdd {s['mdd']*100:4.0f}% a {al['alpha']*100:5.1f}%")
|
||||||
|
d = (r - R["base"]).loc[HOLDOUT[0]:HOLDOUT[1]]
|
||||||
|
dd = (r - R["base"]).loc[DESIGN[0]:DESIGN[1]]
|
||||||
|
extra = "" if name == "base" else (f"{d.mean()*252*100:+5.1f}%/yr t {paired_t(d):+.2f} "
|
||||||
|
f"(design {dd.mean()*252*100:+5.1f}% t {paired_t(dd):+.2f})")
|
||||||
|
rec["diff_holdout"] = float(d.mean() * 252)
|
||||||
|
out[name] = rec
|
||||||
|
print(f"{name:18s}" + " | ".join(cells) + f" | {extra}", flush=True)
|
||||||
|
yr = (1 + R).groupby(R.index.year).prod() - 1
|
||||||
|
print("\nyearly returns (%):")
|
||||||
|
print((yr * 100).round(1).to_string())
|
||||||
|
tag = "_sensitivity" if sens else ""
|
||||||
|
with open(os.path.join(RESULTS, f"momentum_fundamentals{tag}.json"), "w") as fh:
|
||||||
|
json.dump(out, fh, indent=1, default=float)
|
||||||
|
R.to_pickle(os.path.join(RESULTS, f"momentum_fundamentals{tag}.pkl"))
|
||||||
|
|
||||||
|
|
||||||
|
if __name__ == "__main__":
|
||||||
|
main()
|
||||||
Reference in New Issue
Block a user