ResearchMarketsGood Idea
From 1950 to 2024 the S&P 500 rose in the Santa Claus window in 58 of 75 years, by +1.26% on average. The window is the last 5 trading days of December plus the first 2 of January. An ordinary 7-trading-day window in the same years returned +0.23% and was up 57.4% of the time. That gap passes the platform's placebo gate. The Hirsch dates also beat every nearby start date, so this is more than a good December.
The edge has faded, though. From 2010 to 2024 the window averaged +0.32%, the same as an ordinary week. The sealed 2025 window, opened last, lost 0.11%. A failed rally predicts nothing about the rest of January or the next year. This year's window runs from the close on December 23, 2026 through January 5, 2027.
What is the Santa Claus rally?
"Stocks rise in the last days of December and the first days of January." The precise version comes from Yale Hirsch's Stock Trader's Almanac. It is the S&P 500's return from the close before the last 5 trading days of December through the close of the 2nd trading day of January. That is 7 trading days. The Almanac reports an average gain of 1.3% since 1950 (Almanac, January 2026).
Hirsch also wrote the companion rhyme, "If Santa Claus should fail to call, bears may come to Broad and Wall." Its earliest known citation is a Chicago Tribune column of December 20, 1972 (Barry Popik). We use 1972 as the year the rule became public.
We tested three questions, in this order.
- Is the window's return higher than an ordinary 7-day window?
- Did it survive after it became famous?
- Does a negative window predict a weak January or a weak year?
When is the Santa Claus rally in 2026?
By the NYSE calendar, the window starts from the close on December 23, 2026 and runs through the close on January 5, 2027. Its 7 sessions are December 24 (a 1 p.m. early close), 28, 29, 30 and 31, then January 4 and 5. Christmas Day and New Year's Day fall on Fridays, and the exchange is shut. The dates assume no unscheduled closures.
What others have found
On the platform, Lyra Data Engine collects this layer from news, filings and research. For this post we gathered each source by hand, read it, and checked every link.
- The Almanac's own numbers. The Almanac's January 2026 note measured the 2025 window from the S&P 500's close of 6909.79 on December 23, 2025. It found a 0.11% decline. Its January 2024 note counted 15 down windows since 1950. In 10 of them an up year followed, with an average gain of 5.0%. The note also said "the lack of a rally can be a preliminary indicator of tough times to come."
- The folk follow-through. LPL Financial, as reported by Business Insider, split years by the sign of the last 5 trading days of December alone. After a negative finish the S&P 500 averaged -0.1% in January and 6.1% over the next year. After a positive one it averaged 1.4% and 10.4%.
- The streak. The Motley Fool noted that the window was negative 3 years running for the first time since 1950. It put the hit rate at 78%.
- The explanations. Wikipedia's summary lists holiday-thin volume, the end of tax-loss selling, and investors buying ahead of January. It also says there is no generally accepted explanation.
- Why fame should hurt. McLean and Pontiff (Journal of Finance, 2016) studied 97 published return predictors. Returns were 26% lower out of sample and 58% lower after publication. That is the base rate for a published anomaly, and the reason we split the history at 1972.
The practitioner sources agree on the average and disagree on what a failed rally means. None of them compares the window with nearby dates or with random windows. Our test adds placebo windows, a split at publication, and a sealed year.
Data
| Series | Source | Range used | Use |
|---|---|---|---|
| S&P 500 price index, daily close, without dividends | Yahoo Finance | 1950 to 2026-11-17 | main series |
| US market total return, daily | Ken French data library | 1926-07-01 to 2026-09-30 | total-return check, years before 1950, T-bill rate |
| Nasdaq Composite and Russell 2000 | Yahoo Finance | from 1971 and 1987 | other indexes |
| SPY, adjusted close | Yahoo Finance, cross-checked on Alpaca daily bars, SIP feed, from 2016 | 1993 to 2025-09-30 | the tradable version |
| NYSE trading calendar | pandas market calendars | 1950 to 2027 | window dates, check |
Checks
- Calendar. Window dates come from each series' own trading days, never typed in. For the S&P 500 they match the NYSE calendar in 73 of 75 years. The 2 exceptions are 1950 and 1951. The exchange still traded on Saturdays then, and Yahoo's index history has no Saturday closes.
- The 2025 window. Our start close for the 2025 window is 6909.79 and the window return is -0.11%. Both match the Almanac's figures.
- SPY sources agree. On the 9 windows from 2016 to 2024, Yahoo and Alpaca give the same dates. Window returns differ by a median of 0.8 bp and at most 4.9 bp.
- Dividends. The S&P 500 series is a price index. Over the same 75 years the French total-return market averaged +1.42% in the window, against +1.26% for the S&P 500. The gap is 0.15 points. Part of it is dividends and part is the broader market.
Method
The rule. For each December from 1950 to 2024, we take the 7-day return from the close before the last 5 trading days of December to the close of the 2nd trading day of January.
Trials, declared before any result was computed. The bar for every test below counts all 8.
| # | Trial | Series |
|---|---|---|
| 1 | Hirsch window, 7 days (graded) | S&P 500 |
| 2 | Same window, total return | Ken French market |
| 3 | Same window | Nasdaq Composite |
| 4 | Same window | Russell 2000 |
| 5 | Last 5 December days only | S&P 500 |
| 6 | First 2 January days only | S&P 500 |
| 7 | Negative window → rest of January | S&P 500 |
| 8 | Negative window → next 12 months | S&P 500 |
Baselines.
- (a) Every other 7-trading-day window in the same years, 17,897 of them. Windows that overlap a Santa window are left out.
- (b) 2,000 placebo draws of one random 7-day window per year. Each draw has the same 75 windows as the real test.
- (c) The same 7-day window started k trading days before or after the Hirsch start, for k from -20 to +20. If the Hirsch dates only sit inside a strong season, nearby start days should do about as well.
Statistics. We report the mean, median and hit rate. We also report the difference from baseline (a) year by year. Its 95% interval comes from a block bootstrap over years, with blocks of 5 years and 2,000 resamples.
Following January. A full January return overlaps the window by 2 days, so a down window mechanically drags it. For trial 7 we measure January from the window's last close to the end of January. For trial 8 we measure the next 252 trading days from the same point.
Grading uses Lyra's own code. The pipeline imports the platform's grading modules and runs them unchanged.
- Event gate. Each window start is an event. The gate compares the average 7-day return after the 75 events with 2,000 draws of 75 random 7-day windows from the same history. The window must beat them at p ≤ 0.05 / 8 = 0.00625.
- Sealed holdout. The platform sets it to the 365 days that end at the latest quarter start, 2025-10-01 to 2026-09-30. That year holds exactly one window, December 2025 to January 2026. It stayed sealed until every other test was done.
- Grade. Good Idea needs more winning events than losing ones and a positive mean. Awesome Idea also needs the gate, an opened holdout that agrees, and a rule built from a screen. A calendar rule on one index is not a screen, so the best this idea can reach is Good Idea.
Results
The window against an ordinary week
| 1950–2024 | Santa window | Ordinary 7-day window |
|---|---|---|
| Mean | +1.26% | +0.23% |
| Median | +1.25% | +0.38% |
| Up | 58 of 75 (77.3%) | 57.4% |
| Excess over baseline, 95% interval | +1.04 points (0.52 to 1.58) |
The best window was 2008 (+7.45%) and the worst was 1999 (-4.04%). Per day, the market made 18.0 bp inside the windows against 3.1 bp outside, 5.8 times as much.
The whole distribution is shifted, not just the mean. 22.7% of Santa windows were at or below zero, against 42.6% of ordinary windows. In the per-year placebo, random windows were up a median 57.3% of the time. 95% of draws fell between 46.7% and 68.0%. The real 77.3% sits above all of them (p = 0.0005).
Is it the dates or the season?
December is a strong month. Since 1950 it has averaged +1.43% for the S&P 500, second only to November. If the Santa window only borrowed December's strength, windows a few days earlier or later should do about as well.
They do not. Of the 41 start days from -20 to +20, the Hirsch start has the highest average. Starting 1 day earlier gives +0.94%, and 1 day later gives +1.07%. Windows that start 7 to 20 trading days away average +0.30%, close to an ordinary week. The Hirsch window beats them by 0.97 points (0.49 to 1.53).
Most of December's gain happens inside the window. December up to the window averages +0.63%. The last 5 days alone average +0.81%. The first 2 days of January add +0.45%.
One caveat matters. Hirsch picked these dates by looking at history, and we are now testing the dates he picked. The split below at 1972 is the out-of-sample test.
Does the Santa Claus rally still work?
| Era | Mean | Up | Excess over baseline | 95% interval |
|---|---|---|---|---|
| 1926–1949 (French) | +1.94% | 18 of 24 | +1.78 | 0.70 to 2.83 |
| 1950–1971, before the rhyme was public | +1.48% | 19 of 22 | +1.26 | 0.66 to 1.82 |
| 1972–2024, after | +1.17% | 39 of 53 | +0.95 | 0.34 to 1.64 |
| 1950–1999 | +1.48% | 39 of 50 | +1.23 | 0.67 to 1.98 |
| 2000–2024 | +0.84% | 19 of 25 | +0.66 | -0.14 to 1.44 |
| 2010–2024 (not declared) | +0.32% | 11 of 15 | 0.00 | -0.67 to 0.60 |
Three things stand out.
- It held up out of sample, twice. The window worked in the 24 years before 1950, which Hirsch's statistics do not cover. It also worked in the 53 years after publication. The post-publication excess (+0.95) is about a quarter smaller than before (+1.26). That fits the decay McLean and Pontiff find for published anomalies.
- It is weaker since 2000. The 2000–2024 interval includes zero.
- Since 2010 it has been an ordinary week. The window still rose in 11 of 15 years, but by an average of +0.32%. Any 7-day window in those years averaged +0.33%. The 2010s averaged +0.30% and the 2020s +0.36%. Then came 3 down windows in a row, -0.88% in 2023, -0.53% in 2024 and -0.11% in the sealed 2025 window. We did not declare the 2010 split in advance, so treat it as a description, not a test. 15 years is also a small sample, and the interval runs from -0.67 to 0.60.
Does it work on the Nasdaq and small caps?
The window beats an ordinary week on every series. The total-return market made +1.42% (p = 0.001). The Nasdaq Composite made +1.64% since 1971 (p = 0.002). The Russell 2000 made +1.37% since 1987, but it does not clear the gate (p = 0.0175 against a bar of 0.00625). It has only 38 years.
What happens when Santa doesn't come?
The folk version looks right only because of the overlap. Over calendar January, a negative window was followed by -0.05% and a positive one by +1.40%. But 2 of January's days are part of the window, so a down window drags January down by construction.
Measured from the window's close, the claim reverses.
| 1950–2024 | After a negative window | After a positive window | All years |
|---|---|---|---|
| Rest of January, mean | +1.31% (17 years) | +0.42% (58 years) | +0.62% |
| Rest of January, share down | 8 of 17 (47.1%) | 39.7% | 41.3% |
| Next 12 months, mean | +8.17% (16 years) | +9.25% (58 years) | +9.01% |
| Next 12 months, share down | 5 of 16 (31.2%) | 29.3% | 29.7% |
After a negative window, the rest of January was better. The next year was about 1 point weaker, with almost the same chance of a down year. Neither difference survives the placebo. Random December windows separate the outcomes just as well (p = 0.7746 for January, p = 0.3298 for the next year). We count 16 negative windows from 1950 to 2023, where the Almanac counts 15. We did not reconcile the difference.
"Bears may come to Broad and Wall" has no support in this data.
Can I trade it?
We ran the tradable version on SPY from December 1993 to September 2025. The rule holds SPY for the 7 days, pays 5 bp to enter and 5 bp to exit, and sits in T-bills the rest of the year.
- The 32 windows averaged +0.83% before costs and +0.73% after. After costs, 22 of 32 were positive.
- The rule compounded at 3.1% a year, against 2.4% for T-bills alone and 10.7% for holding SPY throughout.
The window adds a little to cash. It is no substitute for being invested, because 7 good days a year cannot outrun the rest of the year's equity return. This is a description of history, not a recommendation.
Validation
All of this comes from the platform's own check code.
| Test | Result | Bar | Passed |
|---|---|---|---|
| Event gate, trial 1, S&P 500 | mean +1.26%, 0 of 2,000 draws as high, p = 0.0005 | p ≤ 0.00625 | yes |
| Placebo, one random window per year | p = 0.0005 for the hit rate, p = 0.001 for the mean | yes | |
| Trial 2, total-return market | p = 0.001 | p ≤ 0.00625 | yes |
| Trial 3, Nasdaq Composite | p = 0.002 | p ≤ 0.00625 | yes |
| Trial 4, Russell 2000 | p = 0.0175 | p ≤ 0.00625 | no |
| Trial 5, last 5 December days | +0.81%, 55 of 75 up, p = 0.005 | p ≤ 0.00625 | yes |
| Trial 6, first 2 January days | +0.45%, 51 of 75 up, p = 0.0085 | p ≤ 0.00625 | no |
| Trial 7, negative window → rest of January | p = 0.7746 | p ≤ 0.00625 | no |
| Trial 8, negative window → next 12 months | p = 0.3298 | p ≤ 0.00625 | no |
| Past cases | 58 of 75 up, mean +1.26% | more than half, mean above 0 | yes |
| Sealed holdout, 2025 window | -0.11%, 0 of 1 | agrees | no |
The random draws averaged +0.25%. The 95th percentile was +0.73%, and the highest of the 2,000 was +1.2%. No draw reached the real +1.26%.
Leave one decade out. With one event a year, combinatorial cross-validation on daily returns says little. Instead we dropped each decade in turn and recomputed the excess over ordinary windows. It ranged from +0.8 points without the 1970s to +1.2 without the 2010s, and stayed positive every time.
Big years. Without 2008 (+7.45%) and 2018 (+1.29%), the mean is +1.18% and the excess +0.93 points (0.36 to 1.57). Our plan listed 2018 as a large December move, but its window was not. The Christmas Eve drop and the December 26 rebound both fall inside it and largely cancel.
Holdout. The gate passed, so the platform opened the sealed year. The 2025 window lost 0.11%. The last 5 December days fell 0.93%, and the first 2 January days gained 0.83%. One event is a very small holdout, and this one does not agree.
Verdict and limits
Grade. Good Idea. The window beat an ordinary week in 58 of 75 years. It passed the platform's placebo gate with all 8 trials counted. It stops short of Awesome Idea for two reasons. A calendar rule on one index is not a screened universe, and the one sealed window lost money.
- The effect is real and tied to the dates. For 75 years the window returned about 1 point more than an ordinary 7-day window. It beat every nearby start date, and it existed before 1950 and after publication.
- It is not just December. Windows a week or more away from the Hirsch dates averaged +0.30%, even inside the November to January season.
- It has faded. The excess is smaller after 1972, not significant after 2000, and zero from 2010 to 2024. Then came 3 down windows in a row.
- A failed rally is not a bear signal. After the window closes, January and the next 12 months look the same whatever the window did.
Limits
- Hirsch chose the window by looking at history. The tests before 1950 and after 1972 are the only clean out-of-sample evidence. We test the dates he chose, not dates found blind.
- 75 events, one a year. A few large years carry weight. Dropping 2008 and 2018, or any one decade, keeps the result, but intervals are wide.
- The 2010 split was not declared in advance. It describes the recent record. It is not a test.
- Price index. The S&P 500 price index leaves out dividends, worth a few basis points over 7 days. The total-return series gives the same answer.
- Costs are estimates. We assumed 5 bp each way, not actual fills.
- One sealed window cannot confirm or reject a 7-day seasonal.
To test your own version of this idea, Lyra runs the same placebo and holdout checks. For another idea that is true as a description but weak as a signal, see gold and real rates. Every idea we have tested this way is in Research.
Questions
What are the Santa Claus rally dates? The last 5 NYSE trading days of December and the first 2 of January, measured from the close before them. In 2026 that is December 24, 2026 through January 5, 2027, measured from the close on December 23, 2026.
Does the Santa Claus rally still work? Less than it did. From 2010 to 2024 the window averaged +0.32%, the same as an ordinary week. The sealed 2025 window lost 0.11%.
Who named it? Yale Hirsch, in the Stock Trader's Almanac. The rhyme about Santa failing to call was in print by December 1972.
Does it work on the Nasdaq and small caps? The Nasdaq Composite averaged +1.64% in the window since 1971 and passed the gate. The Russell 2000 averaged +1.37% since 1987 but did not pass, on 38 years.
Is a failed Santa rally bearish? Not in this data. After a negative window the rest of January averaged +1.31% and the next 12 months +8.17%. Random December windows predict just as well.
Can I trade it? Holding SPY only in the window from 1993 to 2025 earned +0.73% per window after costs. With T-bills the rest of the time, it compounded at 3.1% a year, against 10.7% for holding SPY all year.
Sources
- Stock Trader's Almanac: Santa Claus Rally Comes Up Hair Short, Focus on January Barometer (January 5, 2026)
- Stock Trader's Almanac: Santa Fails to Call, But Trifecta 2 Out of 3 Ain't Bad (January 3, 2024)
- Barry Popik: If Santa Claus should fail to call, bears may come to Broad and Wall
- Wikipedia: Santa Claus rally
- Business Insider via Yahoo Finance: History says last year's weak finish for stocks could be a drag on returns in 2026 (LPL Financial data)
- The Motley Fool: The stock market just did something it hasn't done since 1950
- McLean and Pontiff, Does Academic Research Destroy Stock Return Predictability? (Journal of Finance, 2016)
- Kenneth R. French Data Library
- Yahoo Finance: S&P 500 history
- Alpaca Market Data, historical stock bars
- pandas_market_calendars (NYSE calendar)
Related stocks
SPY
