Skip to main content
Seasonality Investing

Seasonality Investing · Stock Seasonality, 8 Lessons · Lesson 4

Stocks were strong on the days money flowed in

At one time 90% of the gains came from 4 days a month. Why did that pattern disappear, and why does the presidential cycle still remain?

15 min readLesson 4 / 8As of: S&P 500 daily, quarterly and annual data Jan 1950–Sep 2026 · calculated 2026-10-09✓ Read

QToday's question

If you held stocks only 4 days a month and still earned more, can you just keep doing that today?

In 1950–1989, that is exactly how it was. But since 2006 the result has been the opposite. Today we look at the numbers on why the 'days when money comes in' pattern disappeared, and why the 4-year long cycle still remains.

MAPLearning map for this lesson

We build it up in three steps. At the end of each step, a 'Quick check' lets you test your understanding before moving on.

Before you start

In Lesson 1, Why is September so hated?, we learned how to read mean, win rate and sample size against a baseline. In Lesson 2, Sell in May, and Lesson 3, Santa comes, and January speaks, we saw widely known patterns weaken. Today we explain the reason in terms of 'causes.'

  1. 1 · One number90% → 23%

    You can state in a single number how strong the was and how much it has weakened.

  2. 2 · Insight · 3 · DataIt disappears when the structure changes

    You can confirm, using a table by era and the , that even a plausible fund-flow explanation can disappear when the financial structure changes.

  3. 4 · Twist · 5 · ActionLong cycles are still alive

    You can read the together with its sample size, and set the principle of not using a pattern whose cause cannot be written in one line.

Building knowledgeConcepts built so far: 19 · 5 added in this lesson

Seasonality = tilt in probabilityMean, median, win rate, sample sizeComparing against a baselineSince 1950 vs the last 20 yearsNot 'sell' but 'lower your expectations'Nov–Apr vs May–Oct half-year comparisonHalloween effect (36 of 37 countries)The gap vanishing over the last 10 yearsFixed dates vs MACD-confirmed switchingSanta rally's 7 trading days vs the baselineJanuary barometer 2×2Small-cap January effect (excess return)The more famous the pattern, the sooner it weakensYear-end/new-year = a filterTurn-of-month effect 90% → 23%NEWWhen the structure changes, the pattern disappearsNEWThe shorter the cycle, the sooner it weakensNEW4-year presidential cycle and a sample of 18–19NEWOne-line cause testNEW

1 · One number90% → 23%

90% → 23%Share of the S&P 500's total gains from the last 1 day of the month + the first 3 days of the month1950–1989 → 2006 to Sep 2026 · these 4 days are 19% of all trading days · based on log returns

In 1950–1989, 90% of the S&P 500's gains came from 4 days a month: the last trading day of each month and the first three trading days of the next. These 4 days are only 19% of all trading days. Holding only those 4 days would have turned your principal into about 15.6x, while holding only the remaining days would have yielded just 1.36x.

But run the same calculation from 2006 on and the share falls to 23%. That means it is almost the same as the share of trading days (19%). In this period, holding only the 4 days gave 1.52x, and holding only the remaining days gave about 4x. One of the strongest calendar patterns of its time has effectively disappeared.

Quick checkSince 2006, the 'turn-of-month 4 days' 23% share of gains' has become similar to the 'share of trading days, 19%.' What does that mean?
It means those 4 days have become barely different from other days. If they were special days, they should account for a much bigger share than their share of trading days, but now they take only a share equal to their number of days.

2 · InsightThe days money comes in, and the change in structure

The turn-of-month effect had a plausible explanation. Salaries, pension contributions and fund installments enter the stock market at the same time each month, and institutions balance their books at month-end and then deploy new money at the start of the month. Lakonishok & Smidt (1988) reported that in Dow data for 1897–1986, gains were concentrated around this time (Quantpedia summary). Put simply, it is a 'payday market.' Just as restaurants fill up right after payday, buying clusters on the days money comes in.

The problem is that the way that money comes in has changed. Automatic recurring purchases are spread across dates, and algorithmic trading and ETFs deploy money in pieces throughout the day. Above all, once the pattern became known, many investors began buying one day earlier. The changes by era in the chart below fit this explanation, but what worked and how much remains an unconfirmed hypothesis.

Turn-of-month 4 trading days vs the remaining days · S&P 500

Click a bar to see the numbers for that period.

Calculated directly from Yahoo Finance ^GSPC daily closes (price index, excluding dividends and costs). Turn of month = the last 1 trading day of the month + the first 3 trading days of the month. 1950–1989 / 1990–2005 / 2006 to Sep 2026. Share of gains is based on the sum of log returns.

The same thing happened by day of the week. In 1950–1989, Monday's daily average was −0.128%, the only negative one among the five weekdays. The explanation attached was that bad news accumulated over the weekend is reflected on Monday, and this is the . Since 2006, Monday's daily average has changed to +0.020%.

Quick checkThe explanation for the turn-of-month effect was plausible, so why did it disappear?
Because the 'structure' the explanation relied on changed (a hypothesis). The days when money comes in became scattered, and as people who knew the pattern bought ahead of time, the buying that used to cluster on particular dates spread out evenly.

3 · DataShort cycles have all weakened

It is not just the turn of the month and Mondays. When you split cycles of a day, a week or a month into three periods and calculate them the same way, the differences that were clear in 1950–1989 have mostly disappeared since 2006. The baseline in the table below is the daily average of all trading days in the same period (+0.034% in 1950–1989, +0.042% since 2006).

S&P 500 daily average1950–19891990–20052006 onward
Turn-of-month 4 trading days+0.147%+0.103%+0.049%
Remaining days+0.007%+0.021%+0.041%
Monday−0.128%+0.085%+0.020%
Day before a market holiday+0.251% (417 times)+0.048% (152 times)+0.127% (212 times)
All trading days+0.034%+0.036%+0.042%

The trading day before a market holiday is defined as 'a day when one or more weekdays are missing before the next trading day,' and was found from gaps in the daily data. It includes irregular closures such as the 1968 Wednesday closures, so it differs slightly from the course plan's figures (+0.276% / +0.021% / +0.138%). The pre-holiday effect varies in size from period to period, so its cause is hard to state in one line.

Daily average return by weekday · three periods

Click a cell. The darker the color, the more it was above or below the 'average of all trading days' for that period.

Calculated directly from Yahoo Finance ^GSPC daily closes. 1950–1989 / 1990–2005 / 2006 to Sep 2026.

The message is one. The shorter the cycle and the easier a pattern is to trade, the sooner it weakened. A pattern that repeats four times a month or once a week is easy for anyone to test and copy. When many people buy first on the same day, that day's edge shrinks. This is called , and Lesson 7 covers it in detail.

Quick checkSince 2006, Monday's daily average is +0.020% and the average of all trading days is +0.042%. Is Monday a 'weak day'?
The gap is too small to say so. A 0.02 percentage point daily difference is small enough to vanish with a single trading cost. It carries nothing like the weight of −0.128% in 1950–1989.

4 · TwistLong cycles are still alive

The twist. Short cycles have vanished, but long cycles remain. In 1951–2025, the year before a US presidential election (year 3 of the term) ended positive 17 times out of 19, with an average of +17.2%. Midterm years (year 2) averaged +3.7%, the lowest of the four years of the term.

The divides a president's 4-year term into year 1 (inauguration year), year 2 (midterm year), year 3 (pre-election year) and year 4 (election year) and compares annual returns. It is a pattern the Stock Trader's Almanac popularized. There is an explanation that economic stimulus measures cluster in the second half of the term, ahead of reelection and elections, but this is a hypothesis.

S&P 500 annual return by year of the term · 1951–2025

Click a bar to see that year's sample size and the best and worst years. Each dot is one year.

Calculated directly from Yahoo Finance ^GSPC year-end closes (price index). Presidential election years 1952, 1956 … 2024 = year 4. Because 1951 was year 3, year 2 has only 18 observations and the rest have 19.

Splitting by quarter makes the picture sharper. Of the 16 quarters, the lowest average was Q2 of a midterm year (−3.12%, 18 observations), and the highest were Q1 of the following year (the pre-election year) (+7.38%, positive in 18 of 19) and Q4 of a midterm year (+6.72%, positive in 15 of 18). 2026 is a midterm year. The chart below puts the year 2 average path next to what 2026 actually did (through Q3). It is a comparison, not a forecast.

Average cumulative path of the 16 quarters vs. actual 2026

Tap a dot to see that quarter's average and how many times it was positive. The average path is a hypothetical line made by chaining quarterly averages, so no actual year moved exactly like this.

Based on quarter-end closes. 2026: Q1 −4.63%, Q2 +14.87%, Q3 +2.03% (+11.77% at end of September vs. end of 2025). The year 2 average cumulative return for Q1–Q3 is about −3.0%.

Through Q3, 2026 moved very differently from the average path. This is the weight of a statistic with only 18 observations. With just 18–19 observations per year of the cycle, one big year easily shifts the average, and the gap between the year 2 average (+3.68%) and median (+0.58%) is also large. Looking only at 1990 onward, year 2 was positive in only 4 of 9 cases.

Quick check"This is a midterm year, so stocks will rise in Q4." What is wrong with that statement?
The prediction "will rise" is what's wrong. What the data says is only this: "Since 1951, the average Q4 return in midterm years was +6.72%, positive in 15 of 18 cases." Attach the period and sample size, and read it as a probability.

5 · One thing to practiceOne-line cause test

The patterns we saw today split in two: the short cycles that disappeared and the long cycle that remains. We turn one criterion that separates them into a practice task: do not trade a pattern if you cannot write 'why it happens' in one line. On each card below, pick the most fitting cause. Once you choose, you will see how solid that pattern's cause is, along with its data record.

One-line cause test

0 / 6 cards checked

The answers follow this course's classification, and most of the cause explanations are unconfirmed hypotheses.

The figures in this lesson are on a price-index basis, so dividends, trading costs and taxes are excluded. The fact that things leaned one way in the past is no guarantee they will lean the same way going forward.

Quick checkWhat should you do with a pattern whose cause you can write in one line, but whose record over the last 20 years has vanished?
First doubt whether the cause is still at work. If the cause (days when money piles in), as with the turn of the month, has dispersed due to structural change, don't use the pattern even if the explanation survives. It has to pass both tests: the cause and the recent data.
The one sentence left from Lesson 4Money-flow patterns disappear when the structure changes

✓What we learned today

  • In 1950–1989, 90% of the S&P 500's gains came from the last 1 and first 3 trading days of the month, which are 19% of trading days. Since 2006, that share is 23%.
  • The money-flow explanation that salaries and pension contributions arrive at the start of the month sounded plausible, but once automatic contributions and algorithmic trading spread the timing of inflows evenly, the pattern weakened too (a hypothesis). The average Monday return also changed from −0.128% to +0.020%.
  • Twist: the long cycle remains. In 1951–2025, the pre-election year (year 3) was positive in 17 of 19 cases, with an average of +17.2%, while the midterm year (year 2) was the lowest, with an average of +3.7%.
  • Q4 of the midterm year (+6.72%, 18 observations) and Q1 of the following year (+7.38%, 19 observations) had the highest averages of the 16 quarters. Still, each year of the cycle has only 18–19 observations.
  • Principle: do not trade a pattern if you cannot write 'why it happens' in one line.
Continues in the next lesson · Lesson 5Each industry has its own seasons

Today we looked at the calendar of the whole market. In Lesson 5, we look at how seasons differ by industry, such as energy in winter and tech as the exception, along with the gold calendar.

Quiz5-question quiz

Pick an answer and it's graded right away with an explanation.

Q1. The 'four turn-of-the-month trading days' that produced 90% of the S&P 500's gains in 1950–1989 were about what percent of all trading days?

Q2. Since 2006, the four turn-of-the-month trading days' share of gains is 23%. What is the most fitting interpretation?

Q3. What explanation does this lesson give for why short-cycle patterns weakened first?

Q4. Which statement is correct about the S&P 500's record in the pre-election year (year 3 of the term) in 1951–2025?

Q5. Which is the most fitting principle proposed by this lesson?

NextTerms · Next lesson · Related tools

Turn-of-month effectTurn-of-the-month effect

A calendar effect in which stock gains cluster on the last trading day of each month and the first few trading days of the next. It is explained by inflows of salaries and pension contributions, and it has weakened sharply in the S&P 500 since 2006.

Day-of-the-week effectDay-of-the-week effect

A calendar effect in which returns on certain weekdays differ from other days. Monday weakness was the classic case, but no cause was ever confirmed, and in the S&P 500 it has virtually disappeared since 2006.

Presidential four-year cyclePresidential cycle

A pattern that splits the four-year US presidential term into year 1 (inauguration year), year 2 (midterm year), year 3 (pre-election year) and year 4 (election year) and compares stock returns across them. From 1951 to 2025, the S&P 500's average was highest in year 3 and lowest in year 2, but each year of the cycle has only 18–19 observations.

Midterm election yearMidterm election year

The second year of a US presidential term, when the entire House and one-third of the Senate are elected. From 1951 to 2025, the S&P 500's average return in midterm years was +3.7% (18 instances), the lowest of the four years of the cycle, and the average for Q4 of that year was +6.72%.

Pattern decayPattern decay

This is the phenomenon where a past return pattern weakens or disappears as it becomes widely known or as market structure changes. Patterns with short cycles that are easy to trade, such as the turn-of-the-month effect and the Monday effect, faded first.

Sample sizeSample size (n)

The number of observations used to calculate a statistic. In monthly seasonality it means how many times that month has occurred; the smaller the number, the more likely the result is due to chance.

Put what you learned to use right away

Return & CAGR Calculator→Calendar→

For educational purposes only; this is not investment advice. Past statistics do not guarantee future returns. The S&P 500 figures were calculated directly from Yahoo Finance ^GSPC daily closes (price index, excluding dividends, trading costs and taxes; January 1950 to September 2026; retrieved 2026-10-09). Yearly and quarterly returns are based on year-end and quarter-end closes.