PEN-TEST-004 — Phase 11b stock backtest (5 detection scenarios)¶

Field	Value
Test ID	PEN-TEST-004
Date	2026-05-11
Strategy	PEN-STOCK-001
Cohorts consumed	DET-BASELINE-2026-05-11, DET-V1-2026-05-11, DET-V2-2026-05-11, DET-V3-2026-05-11, DET-V4-2026-05-11
Status	complete

Purpose¶

Stock-only backtest of the PEN-STOCK-001 strategy (scaled exit: Leg 1 +15 % half-exit, 3 % trail, –7 % hard stop, 30-day time stop; $500 fixed sizing; regime gate; $0.50 + 5 bp friction) applied identically to each of the five detection cohorts from PEN-TEST-001 / -002 / -003. Isolates the detection effect — same strategy mechanics, different event populations.

Method¶

backtest.py walks the master calendar in chronological order per cohort. For each event, the exit fills are pre-resolved by walking forward-30 trading days at entry time, then the calendar loop applies cash + regime gating to take entries and book the pre-computed exits. Equity is marked daily.

Detection events come from the cached cohort parquets — no detection runs were re-executed for this test.

Headline¶

Cohort	Final equity	Total return	CAGR	Max DD	Sharpe	Trades	Skipped no-cash	Win rate
DET-BASELINE	$40,398	+304 %	7.50 %	−38.1 %	0.543	8,842	6,573	43.7 %
DET-V1	$30,858	+209 %	6.01 %	−20.4 %	0.609	4,533	591	45.1 %
DET-V2	$35,651	+257 %	6.80 %	−21.4 %	0.617	6,108	1,272	44.7 %
DET-V3	$31,240	+212 %	6.08 %	−32.3 %	0.547	5,200	1,070	45.0 %
DET-V4	$28,615	+186 %	5.59 %	−21.2 %	0.580	4,101	511	45.4 %

Baseline wins on absolute compounding (trade count × $500 per trade); V2 wins on risk-adjusted compounding (best Sharpe, half the drawdown).

Files in this directory¶

backtest.py — backtest driver (PEN-STOCK-001 spec embedded in module docstring + constants, lines 1–44)
plot_backtest.py — generates the equity-curve PNG + EOY CSV
backtest.log, backtest.stdout.log — run logs
backtest_summary.json — headline JSON for downstream tooling
backtest_eoy.csv — end-of-year equity table (5 scenarios)
backtest_equity_<cohort>.parquet × 5 — daily equity time series per cohort
backtest_trades_<cohort>.parquet × 5 — full per-trade ledger per cohort
report.md → ../../reports/Pennant/pennant_strategy_backtest_2026-05-11.md
Chart (referenced by report): ../../charts/Pennant/pennant_strategy_backtest_equity_curves_2026-05-11.png

F-004 — Baseline maximizes raw return; V2 maximizes risk-adjusted return; trade-count × average-profit moves the equity curve, not win rate.

Reproducer¶

cd /home/kungfujones/Projects/Uriel/build_v1 && source .venv/bin/activate
cd /home/kungfujones/Projects/Uriel/Pennant
python tests/2026-05-11_PEN-TEST-004/backtest.py