feat: log Phase A decisions and add execution-recovery matrix

Document Phase A (max-hold/vol/corr closed; next-open as decision baseline).
Add stale_close and next_open gap-cap fill modes plus a small matrix to test
whether near-close scheduling recovers overnight momentum drift.
This commit is contained in:
2026-07-18 16:27:10 +02:00
parent 723d47338e
commit 3eb6192a1e
5 changed files with 768 additions and 26 deletions
+494
View File
@@ -0,0 +1,494 @@
"""Execution-recovery matrix: is the close→next_open gap recoverable by scheduling?
Hypothesis (from Phase A A4)
----------------------------
The 1.77 → 1.20 full-period Sharpe gap under next_open fill is mostly overnight
momentum drift that a 07:00-Berlin scanner cannot earn. Near-close / MOC-style
execution (scan ~15:45 ET, fill at/near that close) should recover it.
Pre-registered arms (N for DSR = 4)
----------------------------------
1. ``close_control`` — historical optimistic control (signal = fill at same close).
2. ``next_open`` — honest overnight scanner (decision baseline for future promotion).
3. ``stale_close`` — signal at t1 close, fill at t close (one-session-stale MOC proxy).
Expectation: ≈ close_control; if so, the gap is scheduling, not physics.
4. ``next_open_gap2`` — next_open but skip entries that open > +2% above signal close.
Measure whether large gap-ups are toxic or the best continuations.
Promotion / read rule (pre-registered)
--------------------------------------
- Decision baseline for *future* strategy work: ``next_open``.
- Recovery success for ``stale_close``: validation Sharpe within 0.5×SE of
``close_control`` **and** validation Sharpe ≥ ``next_open``; train Sharpe not
worse than close_control by more than 0.5×SE. State 1-SE distinguishability.
- ``next_open_gap2`` is measurement-only vs ``next_open`` (no auto-promote to live).
Reuses the same daily candidate cache as the Phase A matrix when the cache key
matches.
Usage
-----
python scripts/run_execution_recovery_matrix.py backtest_snapshots/prod.sqlite \\
--workers 7 --allow-spawn \\
--candidate-cache reports/.cache/research-cands.pkl \\
--out reports/execution-recovery-matrix.json
"""
from __future__ import annotations
import argparse
import asyncio
import json
import multiprocessing
import os
import pickle
import sys
from concurrent.futures import ProcessPoolExecutor, as_completed
from datetime import date, datetime
from pathlib import Path
from typing import Any
from sqlalchemy import select
from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker, create_async_engine
ROOT = Path(__file__).resolve().parents[1]
if str(ROOT) not in sys.path:
sys.path.insert(0, str(ROOT))
# Must match Phase A cache when reusing research-cands.pkl
CACHE_VERSION = "research-matrix-v1-daily-prod"
PRE_REGISTERED_ARMS: tuple[dict[str, Any], ...] = (
{
"id": "close_control",
"label": "Close fill (historical control)",
"fill_mode": "close",
},
{
"id": "next_open",
"label": "Next-open fill (decision baseline)",
"fill_mode": "next_open",
},
{
"id": "stale_close",
"label": "Stale-signal close fill (MOC proxy: signal t-1, fill t close)",
"fill_mode": "stale_close",
},
{
"id": "next_open_gap2",
"label": "Next-open + skip gap-up > 2%",
"fill_mode": "next_open",
"max_entry_gap_pct": 0.02,
},
)
N_TRIALS = len(PRE_REGISTERED_ARMS)
def _sqlite_url(path: Path) -> str:
return f"sqlite+aiosqlite:///{path.resolve().as_posix()}"
def _parse_args() -> argparse.Namespace:
p = argparse.ArgumentParser(
description=__doc__,
formatter_class=argparse.RawDescriptionHelpFormatter,
)
p.add_argument("snapshot")
p.add_argument("--workers", type=int, default=6)
p.add_argument("--allow-spawn", action="store_true")
p.add_argument("--out", default=None)
p.add_argument("--candidate-cache", default=None)
p.add_argument("--validation-split", default="2024-07-01")
p.add_argument("--cadence", choices=("daily", "weekly"), default="daily")
p.add_argument("--quiet", action="store_true")
return p.parse_args()
def _period_percentiles(
observations: list[dict], value_key: str
) -> dict[tuple[str, str], float]:
by_period: dict[tuple, list[dict]] = {}
for row in observations:
if row.get(value_key) is None:
continue
period = tuple(row["ranking_period"])
by_period.setdefault(period, []).append(row)
result: dict[tuple[str, str], float] = {}
for group in by_period.values():
ordered = sorted(
group, key=lambda row: (float(row[value_key]), str(row["symbol"]))
)
denominator = len(ordered) - 1
for rank, row in enumerate(ordered):
result[(str(row["symbol"]), str(row["date"]))] = round(
rank / denominator * 100.0 if denominator > 0 else 100.0,
2,
)
return result
def _live_universe_rank_map(
observations: list[dict],
benchmark_closes: dict[date, float],
momentum_weight: float,
) -> dict[tuple[str, str], dict[str, float | None]]:
raw_pct = _period_percentiles(observations, "momentum")
residual_pct = _period_percentiles(observations, "residual_momentum")
vol_pct = _period_percentiles(observations, "vol_6m")
benchmark_ords = sorted(value.toordinal() for value in benchmark_closes)
residual_start_ord = benchmark_ords[251] if len(benchmark_ords) >= 252 else None
ranks: dict[tuple[str, str], dict[str, float | None]] = {}
for row in observations:
identity = (str(row["symbol"]), str(row["date"]))
asof_ord = date.fromisoformat(identity[1]).toordinal()
momentum_pct = (
residual_pct.get(identity)
if residual_start_ord is not None and asof_ord >= residual_start_ord
else raw_pct.get(identity)
)
volatility_pct = vol_pct.get(identity)
strategy_rank = (
round(
momentum_pct * momentum_weight
+ volatility_pct * (1.0 - momentum_weight),
2,
)
if momentum_pct is not None and volatility_pct is not None
else momentum_pct
)
ranks[identity] = {
"momentum_percentile": momentum_pct,
"volatility_percentile": volatility_pct,
"strategy_rank": strategy_rank,
}
return ranks
def _window(arm: dict, name: str) -> dict | None:
for row in arm.get("windows") or []:
if row.get("window") == name:
return row
return None
def _grade_stale_close(close_arm: dict, next_arm: dict, stale_arm: dict) -> dict:
c_val = _window(close_arm, "validation") or {}
n_val = _window(next_arm, "validation") or {}
s_val = _window(stale_arm, "validation") or {}
c_tr = _window(close_arm, "train") or {}
s_tr = _window(stale_arm, "train") or {}
keys = ("sharpe", "sharpe_se")
if any(c_val.get(k) is None for k in keys) or s_val.get("sharpe") is None:
return {"recover": False, "reason": "missing Sharpe rows"}
se = float(c_val.get("sharpe_se") or s_val.get("sharpe_se") or 0.0)
half_se = 0.5 * se if se > 0 else 0.0
cs, ss, ns = float(c_val["sharpe"]), float(s_val["sharpe"]), n_val.get("sharpe")
cts, sts = c_tr.get("sharpe"), s_tr.get("sharpe")
near_close = abs(ss - cs) <= half_se if half_se > 0 else abs(ss - cs) < 0.05
beats_next = ns is None or ss >= float(ns)
train_ok = (
cts is None
or sts is None
or float(sts) >= float(cts) - half_se
)
recover = near_close and beats_next and train_ok
return {
"recover": recover,
"near_close_control": near_close,
"beats_next_open": beats_next,
"train_ok": train_ok,
"validation_delta_vs_close": round(ss - cs, 4),
"validation_delta_vs_next_open": (
round(ss - float(ns), 4) if ns is not None else None
),
"half_se": half_se,
"reason": (
"stale_close recovers close-fill economics (within 0.5 SE) and beats next_open"
if recover
else "stale_close does not meet recovery criteria — see flags"
),
}
def _markdown(report: dict) -> str:
lines = [
f"# Execution recovery matrix — {report.get('generated_at', '')}",
"",
f"Validation split: **{report.get('validation_split')}**. N for DSR: **{report.get('n_trials')}**.",
"",
"| arm | window | Sharpe | SE | CAGR | MaxDD | trades | gap/drift |",
"|---|---|---:|---:|---:|---:|---:|---|",
]
for arm in report.get("arms") or []:
for w in arm.get("windows") or []:
slip = w.get("overnight_slippage") or w.get("signal_to_fill_drift") or {}
slip_s = (
f"mean {slip.get('mean_pct')}% n={slip.get('n')}"
if slip
else ""
)
if w.get("skipped_gap_cap") is not None:
slip_s += f"; gap_skips={w.get('skipped_gap_cap')}"
lines.append(
f"| {arm.get('id')} | {w.get('window')} | {w.get('sharpe')} | "
f"{w.get('sharpe_se')} | {w.get('cagr_pct')} | {w.get('max_drawdown_pct')} | "
f"{w.get('trades')} | {slip_s} |"
)
rec = report.get("recovery") or {}
lines.extend(
[
"",
"## Recovery decision (stale_close)",
"",
f"- **recover: {rec.get('recover')}** — {rec.get('reason')}",
f"- flags: { {k: rec.get(k) for k in ('near_close_control', 'beats_next_open', 'train_ok', 'validation_delta_vs_close', 'validation_delta_vs_next_open', 'half_se')} }",
"",
]
)
return "\n".join(lines)
def _checkpoint(path: Path, report: dict) -> None:
path.parent.mkdir(parents=True, exist_ok=True)
tmp = path.with_suffix(path.suffix + ".tmp")
tmp.write_text(json.dumps(report, indent=2, default=str), encoding="utf-8")
tmp.replace(path)
path.with_suffix(".md").write_text(_markdown(report), encoding="utf-8")
async def _main() -> None:
args = _parse_args()
snapshot = Path(args.snapshot)
if not snapshot.exists():
raise SystemExit(f"Snapshot not found: {snapshot}")
validation_split = date.fromisoformat(args.validation_split)
out_path = (
Path(args.out)
if args.out
else Path("reports")
/ f"execution-recovery-matrix-{datetime.now().strftime('%Y%m%d-%H%M%S')}.json"
)
os.environ["BACKTEST_SNAPSHOT_OFFLINE"] = "1"
if args.allow_spawn:
os.environ["BACKTEST_ALLOW_SPAWN"] = "1"
from app.models.ticker import Ticker
from app.services import backtest_service as bt
from app.services.admin_service import get_activation_config
from app.services.paper_trade_service import get_exit_policy
from app.services.recommendation_service import get_recommendation_config
db_engine = create_async_engine(_sqlite_url(snapshot), pool_pre_ping=True)
Session = async_sessionmaker(db_engine, class_=AsyncSession, expire_on_commit=False)
try:
async with Session() as db:
recommendation_config = await get_recommendation_config(db)
activation = await get_activation_config(db)
exit_config = await get_exit_policy(db)
benchmark_closes = await bt._load_benchmark_closes_for_backtest(
db, days=None, refresh=False
)
ticker_result = await db.execute(select(Ticker).order_by(Ticker.symbol))
symbols = [t.symbol for t in ticker_result.scalars().all()]
prices: dict[str, tuple] = {}
for index, symbol in enumerate(symbols, 1):
columns = await bt._fetch_columns(db, symbol)
if columns is not None:
prices[symbol] = columns
if not args.quiet and index % 50 == 0:
print(f"loaded prices: {index}/{len(symbols)}", flush=True)
finally:
await db_engine.dispose()
snapshot_stat = snapshot.stat()
cache_key = {
"version": CACHE_VERSION,
"snapshot": str(snapshot.resolve()),
"snapshot_size": snapshot_stat.st_size,
"snapshot_mtime_ns": snapshot_stat.st_mtime_ns,
"cadence": args.cadence,
"target_model": "production_gtl",
}
cache_path = Path(args.candidate_cache) if args.candidate_cache else None
qualified: list[dict] | None = None
entry_candidate_count = 0
if cache_path is not None and cache_path.exists():
with cache_path.open("rb") as handle:
cached = pickle.load(handle) # noqa: S301
if cached.get("key") == cache_key:
qualified = list(cached["qualified_candidates"])
entry_candidate_count = int(cached.get("entry_candidate_count") or 0)
if not args.quiet:
print(f"loaded candidate cache: {cache_path}", flush=True)
if qualified is None:
workers = max(1, min(int(args.workers), max(1, multiprocessing.cpu_count() - 1)))
context = bt._mp_context() or multiprocessing.get_context("spawn")
replay_rows: list[dict] = []
with ProcessPoolExecutor(max_workers=workers, mp_context=context) as pool:
futures = {
pool.submit(
bt._replay_candidates_for_period,
symbol,
columns,
recommendation_config,
activation,
benchmark_closes,
date(1900, 1, 1),
args.cadence,
True,
True,
): symbol
for symbol, columns in prices.items()
}
for index, future in enumerate(as_completed(futures), 1):
replay_rows.extend(future.result())
if not args.quiet and index % 25 == 0:
print(f"replay: {index}/{len(futures)}", flush=True)
setup_candidates = [row for row in replay_rows if not row.get("_rank_only")]
rank_observations = [
row for row in replay_rows if row.get("_universe_rank_observation")
]
entry_candidate_count = len(setup_candidates)
live_ranks = _live_universe_rank_map(
rank_observations,
benchmark_closes,
bt.STRATEGY_RANK_MOMENTUM_WEIGHT,
)
threshold = float(activation.get("min_momentum_percentile", 80.0))
qualified = []
for setup in setup_candidates:
if setup.get("direction") != "long":
continue
candidate = {
k: v
for k, v in setup.items()
if not k.startswith("_universe_")
}
rank = live_ranks.get((str(setup["symbol"]), str(setup["date"])))
if rank is None:
continue
candidate[bt.PRODUCTION_PERCENTILE_KEY] = rank["momentum_percentile"]
candidate[bt.VOL_PERCENTILE_KEY] = rank["volatility_percentile"]
candidate[bt.RESIDUAL_HIGH_VOL_BLEND_80_20_KEY] = rank["strategy_rank"]
candidate["qualified"] = bt._momentum_qualifies(candidate, threshold)
if candidate["qualified"]:
qualified.append(candidate)
if cache_path is not None:
cache_path.parent.mkdir(parents=True, exist_ok=True)
with cache_path.open("wb") as handle:
pickle.dump(
{
"key": cache_key,
"entry_candidate_count": entry_candidate_count,
"qualified_candidates": qualified,
},
handle,
protocol=pickle.HIGHEST_PROTOCOL,
)
if not qualified:
raise SystemExit("No qualified candidates")
strategy = next(s for s in bt.PORTFOLIO_MONITOR_STRATEGIES if s.get("is_production"))
entry_config = bt._entry_variant_config(str(strategy["entry_variant"]))
assert entry_config is not None
ranking_key = str(entry_config.get("ranking_key") or entry_config["percentile_key"])
exit_policy = bt.LIVE_EXIT_MODE_TO_SIM.get(
str(exit_config.get("mode", "atr_trailing")), "atr_trail3"
)
hold_days = int(exit_config.get("hold_days", 30))
trail_multiplier = float(exit_config.get("atr_multiplier", bt.ATR_TRAIL_MULTIPLIER))
risk_per_trade = float(entry_config["risk_per_trade"])
max_positions = int(entry_config["max_positions"])
post_stop = bt._make_gate_reset_reentry_fn(
qualified, prices, cadence=args.cadence, ranking_key=ranking_key
)
report: dict[str, Any] = {
"generated_at": datetime.now().isoformat(),
"snapshot": str(snapshot.resolve()),
"validation_split": validation_split.isoformat(),
"n_trials": N_TRIALS,
"hypothesis": (
"stale_close (signal t-1, fill t close) recovers close-fill economics; "
"the next_open haircut is scheduling, not lost edge"
),
"decision_baseline": "next_open",
"qualified_longs": len(qualified),
"arms": [],
"recovery": {},
}
_checkpoint(out_path, report)
by_id: dict[str, dict] = {}
for arm in PRE_REGISTERED_ARMS:
if not args.quiet:
print(f"running {arm['id']} ...", flush=True)
windows = []
for window_name, start, end in (
("train", None, validation_split),
("validation", validation_split, None),
("full", None, None),
):
sim = bt._simulate_portfolio(
qualified,
prices,
benchmark_closes,
exit_policy,
hold_days,
ranking_key=ranking_key,
max_positions=max_positions,
risk_per_trade=risk_per_trade,
atr_trail_multiplier=trail_multiplier,
post_stop_reentry_fn=post_stop,
start_date=start,
end_date=end,
fill_mode=str(arm["fill_mode"]),
max_entry_gap_pct=arm.get("max_entry_gap_pct"),
include_trades=True,
)
if sim is None:
windows.append({"window": window_name, "error": "no_trades"})
continue
dsr = bt.deflated_sharpe_ratio(
sim.get("sharpe"),
sim.get("sharpe_se"),
N_TRIALS,
n_returns=sim.get("n_returns"),
return_skew=sim.get("return_skew"),
return_kurtosis=sim.get("return_kurtosis"),
)
sim.pop("trade_details", None)
sim.pop("equity_curve", None)
sim.pop("benchmark_curve", None)
sim.pop("reentry_events", None)
windows.append({"window": window_name, "dsr": dsr, **sim})
row = {"id": arm["id"], "label": arm["label"], "config": {
k: arm[k] for k in arm if k not in {"id", "label"}
}, "windows": windows}
by_id[arm["id"]] = row
report["arms"].append(row)
_checkpoint(out_path, report)
if not args.quiet:
val = _window(row, "validation") or {}
print(
f" {arm['id']}: val Sharpe={val.get('sharpe')} "
f"DD={val.get('max_drawdown_pct')} trades={val.get('trades')}",
flush=True,
)
if all(k in by_id for k in ("close_control", "next_open", "stale_close")):
report["recovery"] = _grade_stale_close(
by_id["close_control"], by_id["next_open"], by_id["stale_close"]
)
_checkpoint(out_path, report)
if not args.quiet:
print(f"wrote {out_path}", flush=True)
print(f"recovery: {report.get('recovery')}", flush=True)
if __name__ == "__main__":
asyncio.run(_main())