Archive / Equities & volatility / E889
Options-expiry weeks and Fridays no longer beat a normal S&P 500 day
Options-expiry weeks and Fridays earned +6.2 and +6.9 bp per day in 1975-1999 but +0.3 and +3.1 bp in 2010-2026, no more than the same days shifted five trading days (+5.5 and +4.1 bp) or an ordinary day (about +4 bp).
Traders have long said the S&P 500 does better in the week of monthly options expiry and on Fridays. We held the index through every day in those windows, close to close, after a 1.2 bp cost. In 1975-1999, the historic period, the average day earned +6.2 bp in expiry weeks and +6.9 bp on Fridays. In 2010-2026 it earned +0.3 bp and +3.1 bp, while the same days shifted five trading days earned +5.5 bp and +4.1 bp. An ordinary day earned about +4 bp in those years, so neither window beat it.
The stock market tends to rise in the week of monthly options expiry and on Fridays.
Why people believe it
People argue that options dealers hedging around the monthly expiry push the index up during that week, and that traders avoid holding risk into the weekend or reduce it on Fridays, which leaves Fridays with a different pattern from other days. These are among the oldest and most repeated calendar rules of thumb.
Why it would persist The argument, as stated before testing
The windows are fixed and public, but the argument is that the flows behind them are mechanical and not a mispricing anyone is paid to remove: a buyer holds ordinary index risk for one day and is compensated like any other day.
We used daily S&P 500 closes from 1975 to 2026. The expiry-week set is every trading day from Monday to Friday of the week that contains the third Friday of the month. The Friday set is every Friday. For each day in a set we took the close-to-close return, one trading day, and subtracted a 1.2 bp cost. The years 1975-1999 were descriptive, 2000-2026 decided the verdict, and 2010-2026 was the holdout. Two comparisons guard against reading ordinary drift as an effect: the same day sets shifted forward by five trading days (a placebo), and the average of all days in the same years.
2010-2026 net at least +4 bp per day, week-block 95% interval lower bound above 0, same sign in 2000-2009, placebo mean below half the effect, and Holm-adjusted p below 0.05 on 2000-2026.
2010-2026 net at or below 0, or placebo mean at or above the effect.
1.2 bp per day held.
Placebo: each day set shifted forward five trading days, 2010-2026. Shifting an expiry week by five days places it on the following week; shifting Fridays places most of them on the next Friday, except where a market holiday falls in between. Alpha against the average of all days in the same years, net of the same 1.2 bp. Week-block bootstrap intervals and sign tests; Holm correction across all tests of the calendar family run together.
In 2010-2026 the placebo days earned at least as much as the effect for both windows (expiry week +5.5 vs +0.3 bp, Friday +4.1 vs +3.1 bp), which is the kill criterion that fired; net was positive for both, so the first kill condition did not fire.
- Expiry-week days in 2010-2026 (n 968) earned +0.3 bp per day net (95% interval -6.4 to +6.6 bp). Friday days (n 848) earned +3.1 bp (interval -4.1 to +10.3 bp). The result statistics above describe the expiry-week set.
- In 1975-1999 both windows were positive with intervals above zero: expiry weeks +6.2 bp (n 1462, interval +1.5 to +11.0 bp) and Fridays +6.9 bp (n 1259, interval +1.8 to +11.8 bp). These are historic windows, from before the decay seen in later years.
- In 2000-2009 both fell to or below zero: expiry weeks -0.6 bp (interval -11.8 to +10.0 bp), Fridays -7.1 bp (interval -17.9 to +3.4 bp). Over 2000-2026 they were -0.0 bp (n 1545, interval -6.0 to +5.7 bp) and -0.7 bp (n 1352, interval -6.7 to +5.2 bp).
- The placebo days, shifted five trading days, earned +5.5 bp (interval -0.9 to +11.4 bp) for expiry weeks and +4.1 bp (interval -2.8 to +10.6 bp) for Fridays in 2010-2026. Both are at or above the effect, which is the pre-registered kill condition. Five trading days is one week, so the Friday placebo is mostly other Fridays and adds little information beyond the Friday result itself.
- Against the average of all days in the same years (alpha), expiry weeks were -3.6 bp and Fridays -0.9 bp in 2010-2026; in 1975-1999 they were +2.2 bp and +2.9 bp; in 2000-2026 -2.0 bp and -2.7 bp. The average ordinary day net of cost was therefore about +4 bp in 2010-2026 (+0.3 plus 3.6 for expiry weeks, +3.1 plus 0.9 for Fridays; derived from the alpha figures).
- In 2010-2026 the median day earned +4.8 bp in expiry weeks and +7.9 bp on Fridays, above the means. The worst expiry-week day lost 1199 bp (about 12%) and the worst Friday lost 599 bp.
- After Holm correction across all calendar tests run together, the adjusted p-value for 2000-2026 is 1 for both windows (unadjusted week-block bootstrap p: 0.993 for expiry weeks, 0.800 for Fridays).
Positive before 2000, not distinguishable from zero since
Mean net return per day in each window and era, with week-block bootstrap 95% intervals. Both windows have intervals above zero in 1975-1999, but their excess over an ordinary day was only +2.2 and +2.9 bp (alpha chart); in 2000-2009 and 2010-2026 every interval includes zero.
Chart data (bp)
| Mean | 95% low | 95% high | |
|---|---|---|---|
| Expiry week 1975-1999 (n 1462) | 6.2 | 1.5 | 11 |
| Expiry week 2000-2009 (n 577) | -0.6 | -11.8 | 10 |
| Expiry week 2010-2026 (n 968) | 0.3 | -6.4 | 6.6 |
| Friday 1975-1999 (n 1259) | 6.9 | 1.8 | 11.8 |
| Friday 2000-2009 (n 504) | -7.1 | -17.9 | 3.4 |
| Friday 2010-2026 (n 848) | 3.1 | -4.1 | 10.3 |
The placebo days earned as much or more
Window daysSame days shifted 5 trading days
Mean net return per day in 2010-2026 for each window and for the same days shifted five trading days. The pre-registered rule kills the idea when the placebo is at or above the effect, which holds for both windows.
Chart data (bp)
| Window days | Same days shifted 5 trading days | |
|---|---|---|
| Expiry week | 0.3 | 5.5 |
| Friday | 3.1 | 4.1 |
Little left over the ordinary day
Expiry weekFriday
Net return per day minus the average net return of all days in the same years. The excess over an ordinary day was at most +2.9 bp in 1975-1999 and is zero or negative in 2000-2009 and 2010-2026, except +0.7 bp for expiry weeks in 2000-2009.
Chart data (bp)
| Expiry week | Friday | |
|---|---|---|
| 1975-1999 | 2.2 | 2.9 |
| 2000-2009 | 0.7 | -5.7 |
| 2010-2026 | -3.6 | -0.9 |
Why it failed Interpretation · not established
Interpretation, not established: the numbers fit a calendar pattern that was modest even in 1975-1999 (no interval was computed for alpha), where the excess over an ordinary day was +2.2 bp and +2.9 bp, and that has been absent since 2000, when the windows earned about the same as other days or less. The sample does not say why. A common explanation is that rules of thumb lose their edge once they are widely known; this test cannot separate that from chance, because the 2010-2026 intervals are wide.
Lesson
A window can show a positive net return simply because the market drifts up on most days; an average day here earned about +4 bp. A calendar rule should be judged against the ordinary day and against neighbouring days, not against zero. A placebo shifted by whole weeks keeps the weekday, so for a weekday rule it is a weak control and the comparison with the ordinary day carries more weight.
Scope of this result
Covers the S&P 500 cash index, close to close, a one-day hold on each day of an options-expiry week (Monday to Friday of the third-Friday week) and on Fridays, 1975-2026, at a 1.2 bp cost.
A rejected experiment applies to the tested conditions and does not prove an idea can never work. bp = basis point, 0.01%.
1. Test design
Pre-registeredDataUniverseSignalExecutionCostsControlsSample splitPass criterionKill criterionPower2. Result
ObservationsInstrumentsNet mean95% intervalMedianWorstFinding 1Finding 2Finding 3Finding 4Finding 5Finding 6Finding 73. Interpretation
Why it failedLessonScope4. Reproducibility
StepsGrade