Explainer · Kresmion Research
Stock Seasonality and "Sell in May": What Monthly Patterns Show
Published by Kresmion Research. Read our editorial approach and data methodology.
See it liveLive seasonality →Free, no account needed.
Stock seasonality is the pattern of some calendar months showing higher or lower average returns than others, measured over many years of month-end prices.
Market lore has a saying for several of these patterns: "sell in May and go away", the January effect, a Santa Claus stretch at the turn of the year, a weak September. This page covers how a monthly pattern is measured, why the average and the median can tell different stories, what each well-known pattern looked like in Kresmion's data for three large US index funds from 2000 to 2026, what the academic work found, and why every one of these patterns is fragile. It is descriptive throughout.
How a monthly pattern is measured
Every monthly figure from Kresmion's data on this page starts from the same arithmetic, the one Kresmion's stock seasonality tool uses:
- Monthly return: the last close of a month divided by the last close of the month before, minus one, in percent. It is labeled by the later month, so a "November" figure is the move from the end of October to the end of November. For stocks and funds the closes are adjusted for splits and dividends, so a payout does not show up as a fall.
- Per calendar month, across years: every November in the sample is collected, and from that small pile come the average (the mean), the median (the middle year), the share of years up (how many Novembers had a positive return) and the number of years counted.
- A noise check: the t-statistic divides the average by its standard error, the standard deviation of that month's returns across years over the square root of the number of years. Below about 2 in absolute value, the average sits within roughly two standard errors of zero. The test is against zero, not against the other months: stocks rose in most years, so an up month clears it more easily than a down month of the same size.
The noise check needs care because a year has twelve months. Even with no calendar pattern, extreme-looking months turn up by chance, and more than one in twenty when the market's general rise is pushing every month upward; a page that shows dozens of stocks and funds shows many such results.
Average against median: September
The S&P 500 index fund (SPY) shows why one number is not enough. Across the 27 Septembers from 2000 to 2026, September has the lowest average return of the twelve months in Kresmion's data, -1.17 percent. Its median is +0.01 percent, and September rose in 14 of the 27 years.
The two figures disagree because six Septembers fell by more than 5 percent: 2000, 2001, 2002, 2008, 2011 and 2022. Those large falls drag the average down while the typical September was roughly flat. Leave out the three worst years and the remaining 24 average -0.11 percent. A month's average carries its extremes, its median describes its middle year, and neither alone describes the month.
What the S&P 500 fund's record shows, month by month
Kresmion's stock seasonality tool computes these statistics from adjusted daily closes. For SPY, the window runs from the return for February 2000 to the return for October 2026. January has 26 years because the first return in the series is February 2000's, November and December have 26 because 2026 has no completed November or December yet, and the other nine months have 27.
| Month | Years | Average | Median | Years up | t |
|---|---|---|---|---|---|
| January | 26 (2001 to 2026) | 0.33% | 1.49% | 14 of 26 | 0.40 |
| February | 27 (2000 to 2026) | -0.27% | -0.08% | 13 of 27 | -0.32 |
| March | 27 (2000 to 2026) | 0.94% | 1.16% | 18 of 27 | 1.03 |
| April | 27 (2000 to 2026) | 2.00% | 1.26% | 19 of 27 | 2.06 |
| May | 27 (2000 to 2026) | 1.04% | 1.51% | 19 of 27 | 1.47 |
| June | 27 (2000 to 2026) | -0.01% | 0.35% | 16 of 27 | -0.01 |
| July | 27 (2000 to 2026) | 1.60% | 1.80% | 19 of 27 | 2.25 |
| August | 27 (2000 to 2026) | 0.44% | 1.28% | 18 of 27 | 0.65 |
| September | 27 (2000 to 2026) | -1.17% | 0.01% | 14 of 27 | -1.25 |
| October | 27 (2000 to 2026) | 1.29% | 1.36% | 16 of 27 | 1.22 |
| November | 26 (2000 to 2025) | 2.43% | 2.86% | 20 of 26 | 2.85 |
| December | 26 (2000 to 2025) | 0.64% | 1.01% | 17 of 26 | 0.94 |
November has the highest average and the highest share of years up in this window, and its t of 2.85 is the largest (t is given to two decimals here; the tool shows one). Three months reach an absolute t of 2 (April, July and November). That is about what the market's general rise produces on its own: the noise check measures distance from zero, and with SPY's monthly returns shuffled across the calendar about two months a year still clear 2. It is also one fund over one stretch of 26 to 27 years: November's average falls from 2.43 to 2.09 percent if the single month of November 2020 (+10.88 percent) is left out.
The well-known patterns, checked against the record
"Sell in May and go away"
The saying holds that stocks do worse from May to October than from November to April. Sven Bouman and Ben Jacobsen tested it in 2002 under the name the Halloween indicator: in the stock market from 31 October to 30 April, out of it for the other half of the year. They found a written reference in the Financial Times in 1964, and, using monthly index returns mostly from January 1970 to August 1998, they found November to April returns higher than May to October in 36 of the 37 countries they studied, statistically significant at the 10 percent level in 20 of them. For the UK they found evidence of the effect as far back as 1694. For the United States they reported returns more than 5 percent higher, on average, between November and April than in the rest of the year, although the US result was significant only at the 10 percent level, not once January was separated out, and not in their US series from 1802 to 1969.
In Kresmion's data for SPY, the November to April half averaged +5.98 percent across the 26 winters ending in 2001 to 2026, and the May to October half +3.31 percent across the 27 summers from 2000 to 2026. Paired within the same Halloween year (November to the following October), the winter half beat the summer that followed it in 19 of 26 years; set against the summer before it, it was ahead in 14 of 26, so the count depends on the pairing. Two things sit beside that result: the summer half was positive in 21 of its 27 years, and the year-by-year gap between the halves has a t of about 1.0, inside the noise band. For the Nasdaq-100 fund (QQQ) over the same years the order reverses: May to October averaged +5.82 percent against +4.88 percent for November to April, and the winter half came out ahead in only 14 of 26 years.
The January effect
Michael Rozeff and William Kinney documented in 1976 that January returns on an equal-weighted index of New York Stock Exchange stocks stood out against the rest of the year, and Donald Keim showed in 1983 that much of the extra return of small companies arrived in January, concentrated in its first trading days.
The Russell 2000 fund (IWM), which holds smaller US companies, shows no January premium in Kresmion's data for 2001 to 2026: an average of +0.47 percent, a median of +0.06 percent, and a positive January in 13 of 26 years, a t of 0.41. Its strongest month in that window was November, at +3.07 percent with 21 of 26 years up. IWM tracks the Russell 2000, an index of small US companies that leaves out the very smallest stocks, where the early studies found the effect largest, so this is a check of the pattern in one widely held fund, not a replication.
The Santa Claus window
Yale Hirsch, editor of the Stock Trader's Almanac, defined it in 1972: the last five trading days of December plus the first two trading days of January, seven sessions in all, measured here from the close before the window to the close of its seventh session. Over the 26 windows from December 2000 to January 2026, SPY gained in 19, with an average of +0.90 percent and a median of +0.91 percent. For scale, every seven-session stretch in the same 2000 to 2026 history averaged +0.26 percent and was positive 59.4 percent of the time. The gap over an ordinary seven-session stretch is inside the noise band: 19 or more rises in 26 tries happens about one time in nine at the 59.4 percent base rate. The three most recent windows, at the turns of 2024, 2025 and 2026, were all small losses: -0.83, -0.46 and -0.03 percent.
September
September is the month with the lowest average for all three funds in this window, and for none of them is the average clear of zero by the noise check. SPY's average is -1.17 percent with a median of +0.01 percent (t of -1.25); QQQ's is -1.69 percent with a median of -0.28 percent (t of -1.16); IWM's is -1.41 percent with a median of -1.95 percent and 13 of 27 years up (t of -1.18).
Why monthly patterns are fragile
- Small samples. Even 27 years is 27 observations per calendar month. One crash or one surge month moves an average by tenths of a point, as November 2020 does above.
- Many comparisons. Twelve months, many stocks and funds, several windows and several patterns make extreme-looking cells unavoidable. Ryan Sullivan, Allan Timmermann and Halbert White (2001) tested calendar rules on about 100 years of daily US data and found that, once a rule is judged against the full set of calendar rules that could have been searched, the calendar effects were no longer significant. Bouman and Jacobsen answer that "sell in May" was an inherited saying before anyone tested it, so it was not picked out of that search.
- Patterns fade once known. R. David McLean and Jeffrey Pontiff (2016) studied 97 published stock-return predictors (cross-sectional ones, not calendar rules) and found portfolio returns 26 percent lower out of sample and 58 percent lower after publication.
- No cause in the numbers. Tax-loss selling in December, summer holidays and year-end flows are offered as explanations, but a monthly average records what happened, not why. Bouman and Jacobsen ruled out risk, cross-market correlation and the January effect, found a link with the timing of summer vacations, and left the puzzle open.
- One shared history. SPY and QQQ hold many of the same large companies, and all three funds lived through the same events (the 2000 to 2002 technology bust, 2008, 2020, 2022), so agreement among them is not three independent confirmations.
Where to see the numbers
Kresmion's stock seasonality tool lays these statistics out month by month for US index funds and large US stocks with at least ten full calendar years of history: the average, the median, the share of years up and the number of years for every calendar month, with the window stated for each ticker, months whose average sits within noise of zero marked as such, and the number of months that clear the check set against a shuffled calendar. The futures seasonality tool shows the average monthly move and the share of years up for futures contracts, where contract rolls add a caveat of their own, explained in futures seasonality: what monthly patterns tell you. Stock index funds are covered in the index fund explainer.
Honest limitations
The window is one market and one stretch of history: 26 to 27 years for SPY and QQQ, and from mid-2000 for IWM. Monthly returns run from month-end to month-end, so they miss the swings inside a month. The closes are adjusted for splits and dividends, but in the stored series a dividend paid after the history was last rebuilt is not yet folded into the older closes, so the month of the dividend's ex-date reads about one dividend low. For SPY that is June and September 2026, by roughly a quarter of a percentage point each; that moves the affected averages by about 0.01 point and changes none of the counts on this page. The Santa Claus figures use SPY's daily closes only. The academic results are quoted from the papers and cover their own samples, not Kresmion's.
Key takeaways
| Point | Detail |
|---|---|
| Definition | The tendency of some calendar months to show higher or lower average returns, measured from many years of month-end prices |
| How to measure | Average, median, share of years up and number of years per month, plus a t-statistic to separate a pattern from noise |
| November | SPY's strongest month in Kresmion's data, 2000 to 2025: average +2.43%, up in 20 of 26 years |
| Sell in May | SPY's November to April half beat the following May to October half in 19 of 26 years (14 of 26 against the summer before it), but the summer half was positive in 21 of 27 and QQQ showed the reverse |
| September | Lowest SPY average (-1.17%) with a median near zero (+0.01%); within noise of zero for all three funds |
| What it is not | A description of past years, not a forecast for the next one |
Frequently asked questions
Is "sell in May and go away" true?
Historically it has described many markets: Bouman and Jacobsen (2002) found November to April returns higher than May to October in 36 of 37 countries. In Kresmion's data for the S&P 500 fund, the winter half beat the following summer in 19 of 26 years (14 of 26 against the summer before it), yet the summer half still averaged +3.31 percent and was positive in 21 of 27 years, and the Nasdaq-100 fund showed the opposite order. It is a historical tendency with wide variation, not a rule.
What is the best month for stocks historically?
For the S&P 500 fund in Kresmion's data from 2000 to 2025, November had the highest average return (+2.43 percent), the highest median (+2.86 percent) and the most years up (20 of 26). April and July were next by average. The ranking depends on the window and the market chosen.
What is the worst month for the stock market?
September has the lowest average return for the S&P 500 fund in Kresmion's 2000 to 2026 data, -1.17 percent, but its median is +0.01 percent and it rose in 14 of 27 years. A few very bad Septembers (2002, 2008 and 2022 among them) pull the average down, and its average is within noise of zero (t of -1.25).
Does seasonality predict next month's return?
No. A seasonal average summarizes a few dozen past years at most. Academic work finds that calendar effects lose their significance once data mining is accounted for, and that published return patterns weaken after publication. A month's history says how it went in the years counted, not how the next one goes.
What is the January effect?
It is the finding, first documented for an equal-weighted index of NYSE stocks by Rozeff and Kinney (1976) and tied to small companies by Keim (1983), that January returns stood out, especially for small stocks. In Kresmion's data the Russell 2000 fund showed no January premium from 2001 to 2026: an average of +0.47 percent and a median of +0.06 percent, up in 13 of 26 years.
This page is information, not investment advice.
---
Source: Kresmion stock seasonality tool, SPY, QQQ and IWM monthly returns from closes adjusted for splits and dividends, February 2000 (IWM June 2000) to October 2026, computed 7 November 2026 ; Santa Claus windows and seven-session comparison: Kresmion computation from the adjusted daily closes behind the stock seasonality tool (derived returns only), December 2000 to January 2026 ; Bouman, S. and Jacobsen, B. (2002), "The Halloween Indicator, 'Sell in May and Go Away': Another Puzzle", American Economic Review 92(5), 1618 to 1635, https://doi.org/10.1257/000282802762024683 ; Rozeff, M. and Kinney, W. (1976), "Capital Market Seasonality: The Case of Stock Returns", Journal of Financial Economics 3(4), 379 to 402 ; Keim, D. (1983), "Size-Related Anomalies and Stock Return Seasonality: Further Empirical Evidence", Journal of Financial Economics 12(1), 13 to 32 ; Sullivan, R., Timmermann, A. and White, H. (2001), "Dangers of Data Mining: The Case of Calendar Effects in Stock Returns", Journal of Econometrics 105(1), 249 to 286 ; McLean, R. D. and Pontiff, J. (2016), "Does Academic Research Destroy Stock Return Predictability?", Journal of Finance 71(1), 5 to 32 ; Stock Trader's Almanac, the Santa Claus window as defined by Yale Hirsch in 1972, https://stocktradersalmanac.com/Alert/20211123_2.aspx ; label-shuffle check of SPY monthly returns: Kresmion computation, 7 November 2026.
Kresmion Research.
- · Kresmion stock seasonality tool, SPY, QQQ and IWM monthly returns from closes adjusted for splits and dividends, February 2000 (IWM June 2000) to October 2026, computed 7 November 2026: https://kresmion.com/tools/stock-seasonality
- · Kresmion computation from the adjusted daily closes behind the stock seasonality tool (derived returns only): Santa Claus windows December 2000 to January 2026 and all seven-session stretches 2000 to 2026
- · Bouman, S. and Jacobsen, B. (2002), The Halloween Indicator, Sell in May and Go Away: Another Puzzle, American Economic Review 92(5), 1618 to 1635: https://doi.org/10.1257/000282802762024683
- · Rozeff, M. and Kinney, W. (1976), Capital Market Seasonality: The Case of Stock Returns, Journal of Financial Economics 3(4), 379 to 402
- · Keim, D. (1983), Size-Related Anomalies and Stock Return Seasonality: Further Empirical Evidence, Journal of Financial Economics 12(1), 13 to 32
- · Sullivan, R., Timmermann, A. and White, H. (2001), Dangers of Data Mining: The Case of Calendar Effects in Stock Returns, Journal of Econometrics 105(1), 249 to 286
- · McLean, R. D. and Pontiff, J. (2016), Does Academic Research Destroy Stock Return Predictability?, Journal of Finance 71(1), 5 to 32: https://doi.org/10.1111/jofi.12365
- · Stock Trader's Almanac, the Santa Claus window as defined by Yale Hirsch in 1972: https://stocktradersalmanac.com/Alert/20211123_2.aspx
- · Kresmion futures seasonality tool: https://kresmion.com/tools/futures-seasonality
Average return by calendar month for each futures market, with the number of years behind every bucket.
Free to view, no account needed.
Put this to work
Real filings, 13F flows, and positioning reads with their sources, in your inbox when there is something in the data, or live on Telegram. Free, no account.
No fixed schedule: it goes out when the data has something in it. Unsubscribe anytime.
Kresmion publishes information, not investment advice. See our methodology and the latest research notes.