MA-Stretch Query · Recency-Weighted Event Study

SMH/SPY at the 21-day line

SMH/SPY  ·  > 4% below its 21-day MA  ·  forward 21 sessions  ·  recency window from 2021-01-01  ·  half-life 9mo  ·  de-clustered ≥21d
Now: -5.8% vs the 21-day MA  · 4th pctile (2021+) · 2th all-time
Close 0.72 · MA 0.77 · IN BREACH now · all-time range -11.4% … +13.3%

Forward returns by regime

Median forward return (% positive) at 2 / 5 / 8 / 11 / 15 / 21 sessions, conditioned on each market regime — and the recency-weighted read (the headline). Watch how the shape shifts as the window tightens toward now.
-0.8%+0.7%+2.1%+3.6%+5.0%2d5d8d11d15d21d
Full historyPost-CovidAI eraRallyRecency-wtdbaseline
HorizonFull historyn=58Post-Covidn=24AI eran=13Rallyn=10Recency-wtdn=24
2 sessions+0.8% 66%+0.7% 75%+1.0% 77%+0.6% 70%+1.0% 65%
5 sessions+1.1% 57%+0.1% 50%+1.8% 69%+0.9% 60%+1.8% 74%
8 sessions+0.5% 53%+1.7% 58%+3.3% 62%+1.1% 50%+2.4% 51%
11 sessions+1.0% 57%+1.8% 62%+2.9% 77%+1.9% 70%+2.9% 63%
15 sessions+1.1% 60%+1.9% 67%+4.1% 69%+3.3% 60%+4.0% 59%
21 sessions+1.7% 68%+2.2% 74%+3.8% 83%+4.1% 78%+4.2% 84%
Regime delta: the recency-weighted 21-day median is +2.5 pts vs the full-history read — the recent regime is markedly stronger here.
Headline is recency-weighted (faster 9-month half-life) over 24 post-2021 events. Regime windows: AI era from Apr 1 2023 · Rally from the Oct 30 2023 bottom. Tight regimes carry few events (see the n per column) — read them with the recency-weighted line, which keeps sample by fading rather than cutting.
Forward 21d · median
+4.2%
mean +3.7%
% positive
84%
24 events
vs baseline
+3.6 pts
base +0.7% / 56% pos
Range
-3% … +13%
25th +0.0% · 75th +5.2%
Path · max-gain
+5.9%
median MFE in window
Path · max-draw
-2.5%
median MAE in window

Every event · outcome over time

Each dot is one de-clustered event at its date; height is the forward 21-session return. Gold-ringed = post-2021 — if there are few gold rings, the signal is a pre-Covid-regime artifact.
-9%-3%+3%+9%+15%median +1.7%2008201120142017202020232026each dot = one event · y = forward return
By decade: 2000s 15 2010s 18 2020s 24

The event log

Event dateStretchFwd 21dMax gainMax draw
2005-04-15-4.1%+7.7%+8.3%+0.0%
2005-10-28-4.7%+5.8%+6.1%-0.6%
2006-03-09-3.9%+0.6%+1.7%-3.9%
2006-05-23-4.5%-2.1%+0.4%-4.2%
2006-07-10-5.7%-0.6%+2.5%-3.9%
2006-10-24-4.7%+4.1%+4.8%-0.7%
2006-12-13-3.9%+1.0%+3.2%-0.8%
2007-05-30-4.1%+7.7%+8.4%+0.0%
2007-10-12-6.5%-2.3%+2.9%-5.4%
2008-01-03-4.3%+0.6%+1.9%-5.4%
2008-07-09-4.2%+3.9%+7.0%-2.4%
2008-09-03-7.1%-2.9%+1.9%-3.4%
2008-10-21-4.8%-7.1%+7.9%-7.1%
2008-11-19-7.9%+5.5%+9.1%-2.1%
2009-05-07-4.7%+1.6%+5.5%-4.3%
2010-01-22-4.0%+1.2%+3.3%-2.0%
2010-08-10-4.2%-5.3%+0.2%-5.4%
2010-09-10-4.5%+7.9%+7.9%+0.0%
2011-03-09-3.8%+0.7%+1.2%-2.7%
2011-10-04-4.1%+2.7%+5.2%+0.0%
2012-07-12-4.2%+6.9%+6.9%-2.1%
2012-09-20-3.9%-3.8%+0.1%-3.8%
2014-10-10-6.8%+6.5%+7.5%+0.0%
2015-07-07-3.9%-4.6%+0.0%-5.7%
2015-08-20-4.2%+7.8%+9.4%+0.0%
2016-01-15-4.1%+2.7%+4.1%-0.2%
2017-07-03-4.6%+3.7%+6.5%+0.0%
2017-11-29-5.0%-2.5%+0.0%-3.5%
2018-04-05-4.0%-1.7%+2.3%-5.0%
2018-06-25-4.4%-1.7%+1.1%-1.7%
2018-08-17-4.1%+1.0%+5.0%-0.3%
2018-10-10-4.7%+0.2%+3.5%-3.3%
2019-05-13-5.3%-3.1%+2.0%-6.3%
2020-03-18-4.0%+8.8%+12.3%+0.0%
2021-03-04-6.2%+7.2%+7.2%-3.7%
2021-04-30-4.2%+1.8%+1.9%-5.3%
2021-07-16-3.8%+2.2%+7.6%+0.0%
2021-10-04-4.0%+4.0%+4.0%-0.8%
2022-01-20-4.7%-1.2%+1.8%-4.3%
2022-03-07-5.4%-1.4%+5.9%-1.4%
2022-04-05-4.1%-0.7%+0.4%-4.9%
2022-06-16-4.1%+1.7%+1.7%-8.0%
2022-08-22-4.1%-3.5%+2.4%-5.2%
2022-10-11-5.7%+5.1%+5.8%-3.2%
2022-12-27-4.3%+12.6%+13.6%-0.0%
2023-04-25-4.9%+6.8%+8.6%+0.0%
2023-08-11-4.7%+2.1%+6.2%+0.0%
2023-09-21-3.9%+3.4%+7.6%+0.0%
2024-04-19-6.5%+10.2%+10.2%+0.0%
2024-07-17-6.0%-2.4%+1.8%-10.7%
2024-09-03-5.4%+4.1%+7.9%-2.5%
2024-11-15-4.5%+1.0%+1.9%-2.9%
2025-01-27-7.3%+4.2%+7.4%+0.0%
2025-02-27-4.4%-2.7%+3.2%-2.7%
2025-03-31-3.9%+0.8%+1.8%-5.3%
2025-11-20-4.3%+5.4%+9.2%-0.7%
2026-03-06-4.3%+6.9%+7.0%+0.0%
Read it honestly. The recency-weighted read leans on 24 post-2021 events (half-life 9mo); the full-history column carries 58. The regime delta shows how far recent behaviour has drifted from the long base rate. De-clustered by a 21-session gap; overlapping windows remain. A context gauge, not a mechanical signal.
Computed live from adjusted-close history (yfinance). Forward return = close-to-close over 21 sessions from each event. MFE/MAE = max favorable / adverse excursion within the window. Educational research, not investment advice.