Backtest · 复盘

Universe Backtest

Every 1-XH buy signal across the scanned universe, simulated: enter the bar after the signal, manage adds / adaptive sells / −25% hard stop, record the realised return. This is the aggregate "if you took every signal" view.

As of 2026-08-13 · uniform 92-day window (2026-05-13 → 2026-08-13) · 266 closed trades across 172/244 symbols · exits 2026-05-15 → 2026-08-13.

Per-symbol archives span very different lengths (30-min data is capped at 60 days by the source; hourly reaches further back), so all statistics are restricted to the trailing window above to stay comparable — 186 older trades excluded. Prices are split-adjusted; trades beyond ±100% are quarantined as data artifacts. Since 2026-08-10 the scan universe excludes leveraged / inverse / volatility ETPs (structural decay breaks the rebound thesis; they ran a 22% win rate here) — the live track record keeps their past published trades unchanged.

📊 Aggregate clean closed trades in window

Closed trades

266

Win rate

23%

Avg / trade

-2.9%

Median / trade

-2.0%

Book NAV (cashflow proxy)

-2.3%

best / worst trade

+32.4% / -27.8%

open positions

⚔️ Managed tracks · same entries, three managements 197 trades closed on every track

Management	Win rate	Avg / trade	Median	Edge vs native	NAV proxy
Native adds · adaptive sells · −25% stop, intraday	27%	-2.7%	-1.9%	—	-23.4%
Overlay (A) -25% stop / +15% target / 20d, daily closes	49%	-1.9%	-0.2%	+0.77pp	+0.2%
V3 confirm (B+C) single leg · buy signal extends hold (≤60d) · same stop/target	53%	-1.6%	+1.0%	+1.07pp	+0.7%

All tracks start from the same entry events (same first-leg entry price). Overlay (A) exits: 35 target · 26 stop · 136 horizon. V3 confirm (B+C) exits: 42 target · 34 stop · 121 horizon. V3 = actions B+C from the management-rules audit: a later native buy signal confirms the position — it resets the 20-session clock (total hold ≤ 60) instead of adding capital. Pre-registered comparisons — whichever edge persists as the sample grows becomes the default management.

⚖️ vs S&P 500 & Nasdaq-100 same window · same trade dates

	Period return	Avg over trade windows	Our excess / trade	Trades beating it
This backtest (equal-weight, cashflow-anchored proxy)	-2.3%	-2.9%	—	—
S&P 500 (SPY)	+5.6%	+0.6%	-3.5%	21% of 262
Nasdaq-100 (QQQ)	+3.6%	-0.2%	-2.7%	31% of 262

"Avg over trade windows" holds the benchmark over exactly the same entry→exit dates as each closed trade — the fair like-for-like comparison. "Excess / trade" = our trade return minus the benchmark over its own window.

📈 Book NAV proxy vs benchmarks 2026-05-12 → 2026-08-13, all normalised to 1.00

Equal weight across every simulated position live that day. Entry and terminal marks are anchored to recorded strategy cashflows; intervening days use underlying closes as a proxy because per-leg add/reduce cashflows are unavailable. This is not a true daily MTM reconstruction, but unlike sequential trade compounding its terminal values reconcile to the strategy ledger.

🏆 Top performers by compounded return per symbol, in window

Symbol	Variant	Trades	Win	Total
MRNA	reborn_v3	3	33%	+22.9%
IBM	gogo_v2	1	100%	+19.8%
SCHW	gogo_v2	1	100%	+13.5%
INSM	gogo_v2	2	100%	+11.3%
FIG	gogo_v2	1	100%	+10.9%
MRVL	reborn_v3	1	100%	+10.9%
UPS	gogo_v2	1	100%	+10.7%
DJT	gogo_v2	3	67%	+7.9%
PDD	gogo_v2	1	100%	+7.8%
NVST	gogo_v2	1	100%	+7.4%
XYZ	reborn_v3	1	100%	+6.4%
JNJ	reborn_v3	2	50%	+4.1%
VRSK	gogo_v2	1	100%	+4.0%
MRK	reborn_v3	3	67%	+3.9%
DHR	gogo_v2	1	100%	+3.4%

📉 Worst performers

Symbol	Variant	Trades	Win	Total
HTZ	gogo_v2	3	0%	-59.8%
BYND	gogo_v2	2	0%	-37.3%
OKLO	gogo_v2	1	0%	-26.1%
QUBT	gogo_v2	1	0%	-25.9%
JOBY	gogo_v2	1	0%	-25.6%
USAR	gogo_v2	1	0%	-25.4%
LAC	gogo_v2	1	0%	-25.2%
SERV	gogo_v2	2	50%	-24.8%
CLSK	reborn_v3	3	0%	-24.3%
QUBT	reborn_v3	3	0%	-23.9%
MP	gogo_v2	1	0%	-22.9%
RDDT	reborn_v3	2	50%	-20.1%
RGTI	reborn_v3	2	0%	-19.6%
IBM	reborn_v3	1	0%	-18.4%
CRML	reborn_v3	2	0%	-18.1%

Equal-weight, no transaction costs or slippage. Demonstration of signal behaviour only — not investment advice. The forward-tracked version of these signals (which can't be hindsight-fit) lives on the Live Track Record page.