Calendar effects Myth check

The Jewish calendar and the S&P 500: what 98 years show

A viral tip told traders to look up the Jewish calendar. We tested the three ways people read it on every S&P 500 session since 1928. The holidays show nothing, and the old Rosh Hashanah saying is mostly the time of year. The seven-year Shemitah cycle held up better than we expected, and the closer we looked, the less it looked like the calendar.

Atty Labs · 18 October 2026 · 12 min read

0.00% average S&P 500 move across 528 Jewish holiday stretches since 1928

46 of 99 years in which “sell Rosh Hashanah, buy Yom Kippur” was wrong

−3.4% average return in the 14 Shemitah years, against +9.8% in every other year

8 of 14 Shemitah “crash days” on which the S&P 500 closed higher

Where this came from

This started with a 14-second TikTok from @nikivanwyk: “Traders’ biggest piece of advice: look up the Jewish calendar.” It clearly struck a nerve, with more than 400,000 views within a month, and it is a fair question to put to the data. The video gives no rule, so we tested the three readings traders most often mean: the holidays themselves, the old Wall Street saying “sell Rosh Hashanah, buy Yom Kippur,” and the seven-year Shemitah cycle.

The three tests and the bar each had to clear were written down before anything ran. When the Shemitah test came back stronger than the other two, we added four more checks to find out why: the same test on five other stock markets, the Shemitah window moved a month at a time, the year after the Shemitah, and other cycle lengths. Those were added after we saw the first result, and we label them that way.

Three ways to read the tip

The Hebrew calendar is lunisolar: its months follow the moon, and a leap month every few years keeps it near the seasons. Its holidays therefore move against the civil calendar by up to a month from year to year. That movement is what makes it testable. If a date matters, it should matter wherever it lands, and the same stretch of the civil calendar in years when the holiday falls elsewhere becomes a natural comparison.

The holidays. US exchanges stay open on Rosh Hashanah, Yom Kippur, Sukkot, Passover and Shavuot. If the market behaves differently on those days, it should show up in the index.
“Sell Rosh Hashanah, buy Yom Kippur.” An old Wall Street saying that the ten days between the two holidays are a weak time to hold stocks. It has been studied before, with results that point both ways^1,2,3.
The Shemitah cycle. Every seventh year in the Hebrew calendar is a sabbatical year. A 2014 book⁶ argued that it lines up with market crashes, and pointed to its last day, Elul 29, which fell on the two largest one-day point drops the Dow had seen at the time, in 2001 and 2008.

Every test uses the S&P 500 index from January 1928 to October 2026: 24,815 sessions, price only. Each trade was run in AmiBroker, buying and selling at the close of the session the calendar picks, with no costs. Nothing here is a trading result; it is a test of direction.

The holidays: the market doesn’t notice

We bought the close before each run of holiday sessions and sold the close of its last one: 528 holiday stretches of 1.6 sessions on average. Across all of them the S&P 500 moved 0.00% (−0.003% before rounding), with a 95% interval from −0.14% to +0.13%. It closed higher 56% of the time.

To see what an ordinary stretch looks like, we moved every one of those stretches 7 to 60 sessions earlier or later, 108 shifts in all. The shifted stretches averaged +0.04%, and 31 of the 108 did worse than the real holidays. That is what a date with no effect looks like.

Six of the seven holidays have intervals that cross zero. The first days of Passover are the exception (+0.27%), and with seven holidays tested, one of them clearing the 5% level by chance happens about 30% of the time. All holidays together sit at zero.

This does not settle every holiday claim in the literature. A 2004 study of 1946 to 2000 found that trading volume drops by about a fifth on Rosh Hashanah and Yom Kippur, and that returns around the holidays leaned the way the holidays’ moods would suggest¹. We did not test volume. On returns, over a sample twice as long, the leans are inside the noise: Rosh Hashanah +0.00% (69 years), Yom Kippur −0.26% (68 years, interval −0.75% to +0.16%).

“Sell Rosh Hashanah, buy Yom Kippur”: mostly the time of year

Here the window is the close before Rosh Hashanah to the close before Yom Kippur, about six sessions. Over 99 years the S&P 500 fell 0.37% in an average window, with a median of −0.40%. So the saying leans the right way. It was also wrong in 46 of the 99 years, and the three worst windows (1932, −13.1%; 2008, −11.0%; 1931, −9.7%) carry much of the average.

The shift test is where it loses its edge. Moving the same six-session window earlier or later, 7 of 108 shifts did worse than the real one. That puts the real window in the weakest 7%, close to the 5% bar we set and short of it. The chart shows why it gets that close: the weeks before it are weak too. Windows one to four weeks earlier averaged −0.11%, against +0.30% for windows six to twelve weeks away. Rosh Hashanah falls between early September and early October, and the saying is picking up the soft part of the year it happens to sit in.

The real window is low, and so are its neighbours. The holiday window sits at the bottom of a broader dip in September and early October, not on a spike of its own.

Earlier studies line up with this. One found about a 1% edge on the Dow from 1907 to 2008 with a wider window that also took in the days around both holidays². A 2013 look at 2001 to 2012 found a mixed picture, with Rosh Hashanah itself rising and the days around Yom Kippur falling³. Our own split says the same: −0.59% a window before 1970 (42 years) and −0.21% since (57 years).

The Shemitah cycle: the one that held up

Each Hebrew year here runs from the close before Rosh Hashanah to the close before the next one. Since 1930 there have been 14 Shemitah years. They averaged −3.4%, against +9.8% for the other 84 years, and 8 of the 14 fell, against 23 of the 84. Drawing 14 years at random from the 98, an average that low turns up about 8 times in 1,000 (p = 0.008). We expected this test to fail like the other two. It didn’t.

The Shemitah year is the weakest of the seven, and the year right after it is almost as weak. Every other year of the cycle averaged between +7.9% and +16.8%.

The result does not lean on the years the theory was built from. The six Shemitah years before 1970 averaged −4.3%. The two since the 2014 book both fell: −1.9% in 2014–15 and −18.6% in 2021–22. In hindsight, holding the S&P 500 every year except the Shemitah years, with the money in cash earning nothing, would have ended with 2.3 times as much as holding it straight through.

The average comes from six losses of more than 10%: 1930–31, 1937–38, 1965–66, 2000–01, 2007–08 and 2021–22. The other eight years include 1986–87, up 37%.

Why we still don’t read it as the calendar

A real calendar effect should start and stop with the year it belongs to. This one doesn’t. We moved the 12-month window a month at a time, from 18 months before the Shemitah year began to 18 months after. Every window starting between one month before and 14 months after the Shemitah year averaged a loss, and the deepest one starts four months in, at −7.4%, about twice as bad as the Shemitah year itself.

The weak stretch runs about two years and is deepest for windows that straddle the end of the Shemitah year and the start of the next. That is the shape of a few long bear markets that happen to fall in the same part of a seven-year grid, rather than of a date that does something.

Three more checks point the same way.

The year after is almost as weak. It averaged −2.7%, with 9 of 14 years down. Nothing in the Shemitah theory predicts a bad eighth year.
Other cycle lengths find slots as bad. Cut the same 98 years into 8-, 10- or 12-year cycles and the worst slot averages −3.3%, −4.7% and −5.5%. Skipping it in hindsight would have multiplied the ending balance by 1.8 to 2.0. Those slots were found by searching, which the Shemitah slot was not, so they don’t cancel its p-value. They do show how easily a 98-year history produces a slot like this. Testing every calendar rule people have proposed and correcting for the search is the standard way to handle this, and on 100 years of data it removes the calendar effects that look strongest on their own⁵.
Other markets agree, for the same reason. Shemitah years were the weakest or second weakest year of the cycle in Japan, Canada, the UK, Hong Kong and Germany as well. All five histories start between 1965 and 1988, so they share the dot-com bust, 2008 and 2022. A worldwide crash shows up in every index. That is the same few events counted again, not independent confirmation.

Every market shows lower Shemitah-year returns. The markets move together in a crash, so this chart repeats the US result more than it confirms it.

There is no “crash day”

The most specific claim is about one session: Elul 29, the last day of the Shemitah year. On the last trading session of each of the 14 Shemitah years, the S&P 500 closed higher 8 times. The two big drops behind the story were the first session after the September 11 attacks (−4.9%) and the day the House voted down the 2008 bank rescue (−8.8%). Both had a cause in that day’s news. Since the book was published, Elul 29 has fallen on a Sunday both times, in 2015 and 2022, when markets were closed.

Eleven of the fourteen sessions moved less than 2% either way. The average last session of every Hebrew year, Shemitah or not, was −0.21% across 99 years.

What to take from this

The holidays carry no edge. Across 528 holiday stretches the S&P 500 moved 0.00% on average. Holiday or not, it is an ordinary day.
“Sell Rosh Hashanah, buy Yom Kippur” is a September trade. It leans the right way, it was wrong 46 years out of 99, and much of what it catches is the weak stretch of the year the holidays fall in.
The Shemitah gap is real in the record and thin as evidence. It rests on 14 years and six large down years, the weak stretch is two years long and deepest across the end of the Shemitah year, and the famous crash day closed higher more often than not. It is worth watching, and the next one is the clean test.
None of this needs a hidden explanation. Every one of these dates is public and set decades in advance. Anyone can see them coming, which is the reason a lasting edge on a known date is rare.

What we’d test next

The next Shemitah year, which runs from September 2028 to September 2029. It is the first one we can call in advance, and we will report it either way.
US stocks before 1928. Monthly index data reaches back to 1871 and holds eight more Shemitah years that no one used to build the theory. That is the cheapest independent test available today.
Recession and bear-market start dates rather than prices. If the seven-year slot catches the start of downturns, the dates will show it without any choice of window.

Method

Data: S&P 500 index (Yahoo ^GSPC) daily closes, 3 January 1928 to 16 October 2026, 24,815 sessions, price only with no dividends. Yahoo leaves out the Saturday sessions the NYSE held before 1952, so a few early windows end one session sooner. Other markets: Nikkei 225 (from 1965), S&P/TSX (1979), FTSE 100 (1984), Hang Seng (1986) and DAX (1987), daily closes from the same source, each to October 2026.
Calendar: Hebrew dates from the pyluach library, diaspora holiday rules. A session takes the Hebrew date of its daytime: 17 September 2001 is 29 Elul 5761, and Rosh Hashanah 5762 began that evening. Holidays counted: Rosh Hashanah (two days), Yom Kippur, the first two days of Sukkot, Shemini Atzeret and Simchat Torah, the first two and last two days of Passover, and Shavuot (two days). Shemitah years are Hebrew years divisible by seven (5691, 5698 … 5782).
Trades: Every test was run in AmiBroker on daily bars, long only, buying and selling at the close of the session the calendar picks, with $100,000 per trade, fractional shares, one position at a time and no costs. Every calendar date is known years ahead, so a same-session close is not look-ahead. Returns are AmiBroker’s exit price over entry price. Holiday stretches run from the close before a holiday to the close of its last holiday session. The Rosh Hashanah window runs from the close before Rosh Hashanah to the close before Yom Kippur. Hebrew years run from the close before Rosh Hashanah to the close before the next one.
Comparisons: Shift test: the same holiday stretches or Rosh Hashanah windows moved 1 to 60 sessions earlier or later; shifts under 7 sessions overlap the real window and are left out of the comparison, leaving 108. Rolling test: the Shemitah-year window moved in steps of 21 sessions from 378 sessions earlier to 378 later. Cycle-length check: the 98 Hebrew-year returns sliced into cycles of 4 to 12 years.
Statistics: Intervals are 95% bootstrap intervals on the mean (20,000 resamples). The Shemitah p-value is the share of 20,000 random draws of 14 years from all 98 whose average is at or below the Shemitah average. Yearly returns are treated as independent. No result is adjusted for running several tests.
Pre-registration: Written on 18 October 2026 before any result: the three tests, the shift test, the phase comparison for the Shemitah cycle and the crash-day test, with the rule that a reading counts only if the real dates beat at least 95% of their placebos in the claimed direction. The first run covered 1970 to 2026; the extension back to 1928 was added before any 1928 to 1969 result was seen. Added after the first results, as exploratory checks: the other markets, the rolling window, the year after the Shemitah, other cycle lengths and the by-holiday split.
Reconciliation: The Shemitah years were run two ways in AmiBroker: once per year of the cycle, and once with every Hebrew year in two alternating sets. The two give the same return for all 98 years. A sample of trades was checked by hand against the daily bars and the Hebrew dates.
Limits: One index for the main tests, price only, no costs, and close-to-close fills. The Shemitah result rests on 14 observations. The other markets overlap the US sample in time and share its crashes. Nothing here was tested out of sample, because no rule here is a strategy until a future Shemitah year has been called in advance.

References

Frieder, Laura, and Avanidhar Subrahmanyam. “Nonsecular Regularities in Returns and Volume.” Financial Analysts Journal 60, no. 4 (2004): 29–34. Link
Yatrakis, Pan, and Albert Williams. “The Jewish Holiday Effect: Sell Rosh Hashanah, Buy Yom Kippur.” Advances in Business Research 1, no. 1 (2010). Link
Zweig, Jason. “Is It Sell Rosh, Buy Yom? Or Vice Versa?” 7 September 2013. Link
Lakonishok, Josef, and Seymour Smidt. “Are Seasonal Anomalies Real? A Ninety-Year Perspective.” The Review of Financial Studies 1, no. 4 (1988): 403–425. Link
Sullivan, Ryan, Allan Timmermann, and Halbert White. “Dangers of Data Mining: The Case of Calendar Effects in Stock Returns.” Journal of Econometrics 105, no. 1 (2001): 249–286. Link
Cahn, Jonathan. The Mystery of the Shemitah. Lake Mary, FL: FrontLine, 2014.
@nikivanwyk. TikTok video, 22 September 2026. The prompt for this study; every figure here is computed independently. Link

Historical behaviour of stock indexes around calendar dates, not a strategy or a recommendation. Past results do not predict future ones. Full disclaimer.