Setup Score Backtest

What does setup score actually predict?

Forward T+10/T+30/T+60 returns for every historical Stage 2 row, bucketed by setup_score band and entry regime. Every number is computed from real OHLCV — no model assumptions.

399,423 samples2024-01-01 → 2026-08-21Window: T+10 / T+30 / T+60 trading days

By setup score band

Mean / median forward return + win rate (% positive). Higher bands should produce higher means + higher win rates.

Band	Samples	Mean return %			Median return %			Win rate %			Risk / recency (T+30)
Band	Samples	T+10	T+30	T+60	T+10	T+30	T+60	T+10	T+30	T+60	Payoff	Downside	Recent
under-4	43,122	-0.36%	-0.71%	-1.44%	-1.04%	-2.49%	-4.93%	43.6%	41.0%	38.3%	1.26×	-9.48%	35.5%
4-5	60,908	-0.15%	-0.50%	+0.21%	-0.93%	-2.30%	-3.71%	44.0%	41.7%	40.8%	1.27×	-9.28%	39.4%
5-6	87,784	+0.33%	+0.65%	+1.34%	-0.52%	-1.33%	-2.27%	46.6%	45.4%	44.3%	1.34×	-8.79%	43.1%
6-7	92,465	+0.58%	+1.25%	+2.26%	-0.26%	-0.68%	-1.30%	48.1%	47.7%	47.1%	1.35×	-8.61%	46.5%
7-8	81,013	+0.80%	+1.99%	+3.30%	-0.13%	+0.09%	-0.14%	49.0%	50.3%	49.7%	1.34×	-8.76%	50.2%
8-9	31,309	+1.18%	+2.38%	+4.67%	+0.01%	+0.37%	+0.78%	50.0%	51.0%	51.5%	1.34×	-9.29%	51.8%
9-10	2,822	+1.81%	+2.45%	+3.80%	+0.41%	+0.83%	-0.64%	51.3%	52.1%	49.1%	1.27×	-9.88%	55.2%

Color: green = better, red = worse. Win rate >50% means more setups in the band closed up than down at that horizon. Payoff = avg win ÷ |avg loss| (>1× = winners ran bigger). Downside = 25th-percentile T+30 (typical bad case). Recent = trailing-365d win rate (is the edge still holding?).

Edge by sector

Where the Stage 2 edge has been strongest. T+30 win rate, median return, payoff asymmetry, downside, and the trailing-365d win rate for every sector — ordered best-first.

Sector	Samples	Win% (T+30)	Median %	Payoff	Downside (p25)	Recent Win%
Health Care	33,872	52.0%	+0.61%	1.32×	-6.68%	53.3%
Energy	5,946	50.2%	+0.08%	1.47×	-7.53%	53.3%
Financials	45,676	49.0%	-0.27%	1.33×	-7.16%	49.1%
Utilities	9,609	48.7%	-0.31%	1.24×	-8.59%	32.6%
Real Estate	12,624	48.6%	-0.38%	1.25×	-8.20%	43.3%
Industrials	81,163	46.8%	-1.02%	1.45×	-9.45%	45.8%
Consumer Staples	26,043	46.6%	-0.99%	1.32×	-8.49%	47.0%
Consumer Discretionary	70,628	45.3%	-1.34%	1.33×	-9.36%	44.0%
Communication Services	11,170	45.1%	-1.61%	1.27×	-10.30%	37.9%
Materials	76,063	43.4%	-2.03%	1.31×	-9.43%	42.4%
Information Technology	26,629	42.6%	-2.81%	1.35×	-11.52%	33.8%

Ordered by T+30 win rate. Payoff = avg win ÷ |avg loss| (>1× means winners ran bigger than losers). Downside = 25th-percentile T+30 (a typical bad case). Recent = trailing-365d win rate.

Regime × Score win-rate matrix

T+30 win rate (% positive) for each combination of market regime and setup score band. Higher cells = better historical odds. Hover for sample size.

Regime ↓ / Score →	under-4	4-5	5-6	6-7	7-8	8-9	9-10
BULL	38% n=8,466	41% n=16,827	45% n=31,291	48% n=31,770	51% n=24,975	51% n=10,115	52% n=608
NEUTRAL_UP	41% n=6,988	37% n=15,027	39% n=16,475	41% n=18,525	44% n=19,625	45% n=11,275	47% n=1,241
TURNING	62% n=355	58% n=1,127	60% n=1,919	59% n=3,030	65% n=4,142	64% n=3,167	64% n=438
NEUTRAL_DOWN	35% n=15,382	38% n=15,256	44% n=21,932	47% n=23,150	52% n=19,209	56% n=4,354	55% n=274
CAUTION	53% n=9,014	51% n=8,794	53% n=10,142	51% n=7,105	54% n=4,096	56% n=399	—
BEAR	38% n=2,917	44% n=3,877	52% n=6,025	53% n=8,885	50% n=8,966	51% n=1,999	51% n=261

Win rate<40%40-5050-6060-70>70%Hover any cell for sample size + avg return.

Method: for every (stock, date) where setup_score was computed since 2024-01-01, we took the close price on that date and the close at +10/+30/+60 trading days. The forward return is `(close_t / close_0 - 1) × 100`. The regime is `state_5` from `market_analytics` on the entry date. Past performance does not guarantee future results — use these numbers as a calibration of the score's signal quality, not as a forecast.