Quad Witching vs Monthly OpEx — the Friday and the Week After
SPY daily bars 2002-01 → 2026-09 (tastk); the premarket section uses IB 5-minute bars
including the premarket session, 2018 → 2026. Expiration = third Friday of the month, rolled back one
session when that Friday is a holiday. Generated 2026-09-17.
In one line: on quad-witching Friday SPY falls
open→close 69% of the time (September: 71%), against 49% on a normal monthly
expiration — a coin flip. The week after a September quad falls 75% of the time, mean
−1.26%, while the week after a normal monthly expiration rises 63% of the time.
Thursday shows no pattern at all.
1. Daily bars (2002 onward)
The "Fri open→close" and "next 5 days" columns are mean / share down; every other column is
mean / share up.
| Group | n | Thursday | Fri gap (raw) | Fri gap (ex-div adj.) |
Fri open→close | Next 5 days |
|---|
| Quad witching (Mar/Jun/Sep/Dec) | 96 | +0.04% / 54% |
−0.24% / 26% | +0.22% / 66% |
−0.31% / 69% down | −0.10% / 60% down |
| September quad | 24 | +0.23% / 50% | −0.09% / 25% |
+0.36% / 71% | −0.40% / 71% down |
−1.26% / 75% down |
| Normal monthly OpEx | 192 | −0.10% / 51% | +0.10% / 56% |
+0.10% / 57% | −0.02% / 49% down | +0.37% / 37% down |
⚠ The "Friday gap" column is unreadable until you remove the dividend.
SPY's quarterly ex-dividend date is quad-witching Friday, worth 0.45% on average
that day (0.00% on a normal monthly). Before the adjustment the data says "gaps up 26% of the time";
after it, "gaps up 66% of the time" — the opposite conclusion. The same correction applies
to strikes: any expiration after the ex-date lives a dividend above spot, so Friday's 750 put wall equals
751.83 on Thursday's spot. The morning plan now prints an EX-DIV line for this.
2. Premarket (IB 5-minute bars incl. premarket, 2018 → 2026-06, 34 quads)
Daily bars cannot see the premarket high; by the opening bell the move has usually already faded. All figures are measured against the ex-dividend-adjusted prior close (dividend backed out of adjClose; checked against four actual SPY payouts, within half a cent).
| Mean | Median | Share |
|---|
| Premarket high (vs ex-div-adjusted prior close) | +0.49% | +0.36% | 85% trade up |
| Opening gap | −0.01% | −0.09% | 47% gap up |
| Open → close | −0.41% | −0.14% | 68% down |
| Premarket high → close | −0.90% | −0.74% | 82% down |
The premarket high clusters in two windows: 04:00–05:00 ET and 09:20–09:25 ET.
3. The executable version: rest a sell limit the night before, hold to Friday's close
Sell price = ex-div-adjusted prior close × (1 + threshold). Filled only if touched in the premarket (04:00–09:30 ET); an unfilled order is cancelled at the open.
| Threshold | Filled | Win rate | Mean | Worst | Excl. the 2018-12 / 2020-03 crashes | Trade by trade since 2024 |
|---|
| +0.20% | 25/34 | 72% | +0.69% | −0.87% | +0.45% | +0.91 / +0.44 / −0.28 / −0.70 / +1.67 / −0.81 |
| +0.30% | 19/34 | 74% | +0.75% | −0.71% | +0.41% | +0.55 / −0.18 / −0.60 / −0.71 |
| +0.40% | 15/34 | 73% | +0.92% | −0.61% | +0.50% | +0.65 / −0.50 / −0.61 |
| +0.50% | 8/34 | 88% | +1.34% | −0.51% | +0.52% | +0.75 / −0.51 |
⚠ It has clearly decayed. The +0.30% threshold has lost 3 of its last 4 fills. The long-run mean rests largely on two crashes (2018-12 and 2020-03); strip those out and roughly +0.4–0.5% per trade is left. With n=34 this does not carry size.
⚠ Cancel at the open. If the unfilled order is left working through the regular session it fills on the strong days instead: fills rise to 19/34 and 15/34 at the +0.40% / +0.50% thresholds, the win rate drops to 63% / 60%, and the mean excluding the two crashes falls to +0.31% / +0.15%.
4. SPY vs QQQ — which one fades more?
QQQ pops higher, fades by the same amount, and does it less reliably. Trading the fade through NQ instead of ES buys no extra edge per unit of risk — only more noise.
Premarket layer (same 34 quads, paired by date). QQQ does not go ex-dividend on quad-witching Friday (0 of 34), so no adjustment applies to it.
| Premarket high | Opening gap | Open → close | Premarket high → close |
|---|
| SPY | +0.49% / 85% up | −0.01% / 47% up | −0.41% / 68% down | −0.90% (median −0.74) / 82% down |
| QQQ | +0.60% / 94% up | +0.09% / 50% up | −0.40% / 59% down | −0.90% (median −0.63) / 79% down |
| Paired QQQ − SPY | +0.107pp, t=+2.43 (QQQ higher on 71% of days) | +0.097pp, t=+2.19 | +0.009pp, t=+0.10 | 0.000pp, t=0.00 |
Daily layer (common history 2006-05 → 2026-09, 81 quads, ex-dividend adjusted):
| Gap | Intraday high vs prior close | Open → close | High → close |
|---|
| SPY | +0.20% / 63% up | +0.62% | −0.34% / 68% down | −0.76% |
| QQQ | +0.26% / 67% up | +0.78% | −0.33% / 64% down | −0.85% |
- The pop is bigger in QQQ, and significantly so — +0.11pp more premarket high (t=2.43).
- The fade is the same size. Premarket high → close is −0.90% in both; the paired difference is exactly zero. QQQ simply starts from a higher point.
- SPY is the more reliable short: it closes below the open 68% of the time against QQQ's 59%.
- On daily bars QQQ's high → close is 0.09pp worse in raw terms (t=−2.34), but QQQ's ordinary open→close volatility is 1.19× SPY's; normalised by that, the gap is +0.05σ (t=+1.17) — not significant.
- The resting-sell-limit test confirms it. At +0.30% QQQ wins 60% (SPY 74%), worst trade −2.07% (SPY −0.71%); excluding the two crashes QQQ's edge falls to +0.04% / +0.08% at the two higher thresholds (SPY +0.50% / +0.52%).
| QQQ threshold | Filled | Win rate | Mean | Worst | Excl. the two crashes |
|---|
| +0.20% | 24/34 | 67% | +0.65% | −2.17% | +0.36% |
| +0.30% | 20/34 | 60% | +0.65% | −2.07% | +0.29% |
| +0.40% | 14/34 | 50% | +0.61% | −1.98% | +0.04% |
| +0.50% | 12/34 | 67% | +0.76% | −1.88% | +0.08% |
Side finding, unrelated to quad witching: on normal monthly expirations (n=163) QQQ is measurably weaker intraday than SPY — open → close −0.10% vs −0.02%, paired t=−2.69, and still t=−2.62 after volatility normalisation. No such gap exists on quad-witching days.
5. How to use it
- Horizon: the Friday itself (open→close) plus the following week. This is a calendar
effect and belongs to the same 1–5 day layer as walls and OI — it must not drive medium- or
long-term position sizing.
- Instrument: ES / MES futures, or EW3 options, which settle on Friday's
close. Quarterly
ES options settle on the Friday opening SOQ —
they settle at the very moment of the premarket high and capture none of the fade, so they
cannot be used for this.
- Dividend: SPY goes ex-dividend that day, so any trigger tied to a post-ex strike
needs the dividend added back; ES/MES are unaffected.
- The September week-after is the strongest bearish tilt (18 of 24 years down), but in
2026-09 it contradicted Dr. Sai's multi-week bullish view and was not taken.
- ES, not NQ. Section 4: the Nasdaq leg pops harder but fades by the same amount with a lower hit rate and a three-times-larger worst case — the same reasoning that keeps the Connors trades on SPY.
Source notes: ~/research/notes/quad_witching_seasonality.md; ex-dividend conversion: ~/research/notes/spy_exdiv_wall_offset.md; reproducible from ~/research/scripts/quad_witch_premarket_spy_qqq.py (cached bars in ~/research/data/quad_premarket.csv) and quad_witch_spy_vs_qqq.py. Sections 2–4 re-run 2026-09-17 on a corrected dividend base.