Calendar effects in stocks: January, sell in May and day of week

"Sell in May and go away." "Stocks rally in January." "Never buy on a Friday." Market folklore is full of calendar rules, and some of them began as serious academic findings. They are appealing because they are simple: no forecasts, no analysis, just a date. The question is whether they still hold, whether they ever held strongly enough to use, and how to tell a real pattern from the noise that any long data set contains.

This article describes the best-known calendar effects (the January effect, "sell in May" or the Halloween effect, the turn-of-the-month effect and day-of-week effects) and what research found. Then it tests the main ones on Finabase's own SPY data, from 1993 to 2026, with the statistics needed to judge them. The goal is not a trading rule. It is to show how to read a seasonal claim.

Why calendar patterns could exist, and why they might not

There are plausible reasons some dates could differ. Investors sell losing stocks in December to realize tax losses and may buy back in January. Pension contributions and paychecks arrive around the turn of the month. Fund managers may "window dress" before quarter-end reports. Trading volume falls in the summer.

There are also strong reasons for skepticism. If a pattern is predictable and profitable, traders should buy before it and sell into it, which shifts or erases it. And with enough data, some date-based pattern will always look significant by chance. Test 12 months, 5 weekdays, the days around holidays and so on, and some of them will stand out even if returns are completely random. That is the problem of data mining, and it is the single most important idea in judging any seasonal claim.

The January effect

The claim. Stocks, especially small ones, do unusually well in January.

The research. In 1976 Michael Rozeff and William Kinney reported that January returns on an equal-weighted index of NYSE stocks were much higher than in other months. In 1983 Donald Keim showed that much of the size effect, small companies outperforming large ones, was concentrated in January, and especially its first few trading days. Tax-loss selling in December, followed by rebuying, was the leading explanation.

What our data show. For large US companies the effect is absent. Over 33 Januaries from 1994 to 2026, SPY averaged +0.76%, against +0.80% for all months. January was up in 20 of 33 years, about the same 60%-plus rate as the average month. Because the original effect lived mainly in small stocks, this does not refute it; it says the S&P 500 never carried it in a usable form. Studies have found the small-cap January effect itself much weaker in recent decades.

Sell in May and go away (the Halloween effect)

The claim. Stocks do better from November through April than from May through October, so investors should sell at the start of May and buy back at the end of October.

The research. In 2002 Sven Bouman and Ben Jacobsen tested the saying across 37 markets and found the November-to-April half beat May-to-October in most of them, over long periods. They called it the Halloween indicator and described it as a puzzle without a convincing explanation.

What our data show. We split each year in SPY's history into its two halves, using month-end closes, price only:

The pattern is there, and it is not small: $1 invested only in the winter halves grew to about $6.59, against $2.48 for the summer halves. The winter half beat the summer that followed it in 22 of 32 years.

Before treating that as a rule, read the rest of the table. The summer half was positive in 24 of 33 years, averaging +3.4%. An investor who actually sold in May would have given that up and earned only what cash paid instead. The calendar split also has bad years in both directions: the six months from November 2021 to April 2022 lost about 10%, and May to October 2025 gained 23%. In a taxable account, selling every May would realize gains each year. The honest summary is that the winter half has had better returns on average, while the summer half has still, on average, beaten cash.

September and the month-by-month picture

November (+2.44%) and April (+2.00%) were the strongest months, and September (−0.90%) the only clearly negative one. September was weak in both halves of the sample, averaging about −0.9% in 1993–2009 and again in 2010–2026, which is the kind of consistency that makes a pattern more interesting.

Now the statistics. A single month's return in this sample has a standard deviation of about 4.3 percentage points. With about 34 observations per month, the uncertainty in each month's average (its standard error) is roughly 4.3 ÷ √34 ≈ 0.73 points.

A z of about −2.3 would be notable for a single test chosen in advance. But we looked at 12 months. If returns were pure noise, the chance that at least one of 12 months lands beyond ±2 standard errors is roughly 40%, and beyond ±2.3 roughly 20%. So September's weakness is suggestive, not proof, and every other month's difference is comfortably within what chance produces.

Turn of the month

The claim. Returns cluster in the last trading day of a month and the first few days of the next.

The research. Robert Ariel reported in 1987 that US stock returns were concentrated in the first half of the month, starting on the last trading day of the prior month, and Josef Lakonishok and Seymour Smidt found a turn-of-the-month effect in 90 years of Dow Jones data in 1988. Monthly inflows from paychecks and retirement plans are the usual explanation.

What our data show. Daily bars in our snapshot only go back to October 2024, so we can test only two years, about 500 trading days. On the last trading day of each month and the first three of the next, SPY averaged +0.15% a day (96 days), against +0.04% on all other days (404 days). That is the direction the research predicts, but daily returns vary by about a percentage point, so with 96 days the uncertainty in that average is about ±0.1 point. Two years can't confirm or reject it.

Day of the week

The claim. Monday is bad for stocks (the "weekend effect").

The research. Kenneth French found in 1980 that from 1953 to 1977 the S&P 500's average return from Friday's close to Monday's close was negative, while other days were positive. Later studies found the effect weakened sharply from the late 1980s on, and in some later periods reversed.

What our data show. Over the last two years, Monday was the best day (+0.22%, 96 days) and Thursday the worst (−0.12%). That is the opposite of the classic finding, and it means nothing: with about 100 days per weekday and a daily standard deviation near 1%, each average is uncertain by about ±0.1 point, and these differences sit inside that band. A two-year sample is a lesson in how easily noise produces a "pattern."

Other calendar lore

• The Santa Claus rally: gains in the last five trading days of December and the first two of January.
• The presidential cycle: weaker markets in the first two years of a US presidential term and stronger ones in the third.
• Pre-holiday effects: higher average returns on the trading day before a market holiday.
• Options-expiration and index-rebalance days: unusual volume and moves on known dates.

Each has some supporting research and many critics. All share the same problems: few independent observations (there are only about two dozen presidential terms in modern market history), many possible patterns to choose from, and effects that tend to fade once widely known.

Why published anomalies fade

David McLean and Jeffrey Pontiff studied 97 return patterns documented in academic journals and found that their returns were substantially lower after publication, by more than half on average. Part of the drop reflects data mining (some patterns were never real), and part reflects traders acting on them once they were known. Calendar effects fit that story. The weekend effect and the large-cap January effect largely disappeared in the decades after they became famous.

How to test a seasonal claim yourself

• Use a long sample, and count observations, not days. Thirty-three Januaries are 33 data points, not 700 trading days.
• Compute the noise. Divide the standard deviation by the square root of the number of observations before comparing averages.
• Adjust for how many things you looked at. One striking month out of 12 is expected.
• Split the sample. A real effect should appear in both halves of the history.
• Count costs and taxes. A strategy that trades on the calendar pays spreads every time, and in a taxable account realizes gains.
• Compare with the alternative. Being out of the market for six months means earning cash returns instead of stock returns, which on average were still higher.

How to explore seasonality in Finabase

• Charts. Any symbol's chart, for example SPY, can be switched to weekly or monthly bars over long ranges to look at the same history used here.
• Backtesting. The backtester tests indicator rules on historical prices; pairing it with the questions above is a good guard against fitting noise. The strategy builder builds rule-based allocations.
• Indexes. The indices page tracks the S&P 500, Nasdaq, Dow and others in real time.

Frequently asked questions

Is "sell in May and go away" true?

In SPY from 1993 to 2026, the November-to-April half returned more on average (+6.3%) than May-to-October (+3.4%), and beat the following summer in 22 of 32 years. But the summer half was still positive in 24 of 33 years, so selling in May usually meant giving up gains in exchange for cash returns.

What is the worst month for the stock market?

In SPY's history from 1993 to 2026, September, with an average price return of −0.90% and gains in only 17 of 34 years. Its weakness has been consistent across decades, but given how noisy monthly returns are, it is not strong enough to rely on.

Does the January effect still exist?

Not in large US stocks: SPY's average January return (+0.76%) was in line with the average month. The original effect was concentrated in small stocks, where research suggests it has weakened a lot since it was first published.

Is Monday a bad day to buy stocks?

The classic weekend effect, negative average Monday returns, was found in data up to the 1970s and has largely disappeared since. In the two years to October 2026, Monday was SPY's best day on average, which is also most likely noise.

Summary

Calendar effects are patterns tied to dates: January, the six months from November to April, the turn of the month, the day of the week. Some began as solid research findings, but in 33 years of SPY data most look weak or absent. January was an average month for large stocks, Monday's famous weakness has vanished, and the turn-of-the-month pattern can't be judged from two years of daily data. The strongest pattern, better returns from November to April, is real in the data, but the summer half still usually beat cash. Monthly returns vary by about ±4.3 points, so with a dozen months to compare, a standout or two is expected by chance. The habit worth keeping is the method: count the observations, measure the noise, split the sample and remember how many patterns were tried.