Skip to main content
Seasonality Investing

Seasonality investing · Stock seasonality, 8 lessons · Lesson 7

Why do patterns die? So how do you use them?

Known patterns weaken as they get traded. You'll build five questions to pick out the patterns that survive, and a three-line rule for using the calendar as a filter.

Reading time 15 minLesson 7 / 8As of: S&P 500 daily closes, January 1950–September 2026 · calculated 2026-10-10✓ Read

QToday's question

Why do known seasonal patterns weaken, and how should you use them anyway?

In Lessons 1–6 we saw both living and vanished patterns. This lesson sorts out where that difference comes from, and you'll write your own three-line rule for using seasonality as a 'filter', not a 'signal'.

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

Recall the four numbers (average, median, win rate, sample size) from Lesson 1, Why is September so disliked?, the six-month cycle from Lesson 2, Sell in May was real. And then it cooled, and the turn-of-the-month effect from Lesson 4, On the days money comes in, stocks were strong. This lesson gathers those patterns into one table and re-judges them.

  1. 1 · One number · 2 · Insight58% and five questions

    You can explain why known patterns weaken, and screen out in advance with five questions.

  2. 3 · Data · 4 · TwistWeakened patterns and a filter backtest

    You can judge 1950–1989 and 2006 onward side by side, and confirm with numbers that a 200-day moving average reduces drawdown rather than boosting returns.

  3. 5 · One thing to practiceCondition, action, invalidation

    Write one of your own rules in three lines, check it with the five questions, then save or print it.

Building knowledge34 concepts built so far · 5 added in this lesson

Seasonality = a tilt in probabilityAverage, median, win rate, sample sizeComparing against the baselineSince 1950 vs. the last 20 years'Lower your expectations,' not 'sell'Nov–Apr vs. May–Oct half-year comparisonHalloween effect (36 of 37 countries)The gap vanished in 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-the-month effect 90% → 23%When the structure changes, the pattern disappearsShorter cycles weaken firstPresidential four-year cycle and 18–19 observationsOne-line cause testSector half-year rotation (Stovall)Seasonality created by real demandExcess return vs. the market and number of winsTech exception, defensive reversalMonthly gold futures record and the limits of continuous futuresKOSDAQ weak in Jun–Oct, strong in JanuaryMajor-shareholder capital gains tax and the year-end record dateDisappearance of the January excess return (since 2016)A pattern made by rules is changed by rulesChecking December net selling by individuals yourselfPatterns weaken once known (−58% after publication)NEWFive questions before using oneNEWOnly patterns with an embedded cause remainNEWA filter adjusts risk, not returnsNEWThree lines: condition, action, invalidationNEW

1 · One number58% decline after publication

58%The average share by which the returns of 97 stock-return predictors shrank after the papers were publishedMcLean & Pontiff (2016), Journal of Finance · 26% decline on average in out-of-sample periods

One study collected 97 'money-making patterns' found by academics and re-measured their performance after the papers came out. The result: returns shrank by an average of 26% in out-of-sample periods and by an average of 58% after publication (McLean & Pontiff, 2016). Once a pattern is known, money chasing it piles in, and that money moves prices ahead of time and eats into what is left.

Seasonality has followed the same path. The turn-of-the-month effect from Lesson 4 explained 90% of the S&P 500's gains in 1950–1989, but since 2006 it has dropped to 23%. This lesson asks two questions: which patterns survive and which disappear, and how should the survivors be used?

Quick checkCan you say in one sentence why patterns weaken as they become famous?
Because trading aimed at the pattern grows, and prices move ahead of time. If everyone tries to buy at the start of the month, some people start buying earlier at the end of the month, and eventually the share left at the start of the month shrinks.

2 · InsightOnly patterns with an embedded cause remain

Put the vanished and surviving patterns side by side and one difference shows up. Surviving patterns have causes embedded in rules or real demand. When schedules that people cannot avoid even if they know about them move money, such as tax record dates, settlement calendars, and the heating and driving seasons, the cause remains even after the pattern becomes known. By contrast, price patterns whose cause stays at the level of a 'plausible explanation', such as the day-of-the-week effect or the turn-of-the-month effect, tend to get erased by trading the moment they become known. This distinction is an inference read from the data, though, and even patterns with a cause change when the rules change (Lesson 6, Korean major-shareholder taxation).

Easy analogy · a commuting shortcut

A shortcut nobody knows about is fast. But the moment it shows up on the navigation app, cars flood in and it becomes an ordinary road. On the other hand, the pattern 'the road in front of the school is jammed at 8 a.m.' doesn't disappear even when everyone knows it, because the cause, school start time, is still there. Seasonality also lasts longer when its cause is embedded in a schedule.

So before using seasonality, ask the five questions below first. If even one answer is 'no', don't give that pattern a big weight.

Five questions to ask before using one

1 Can you explain the cause in one line? 2 Is the at least 15 observations? 3 Is it still valid in the last 10 years? 4 Does it show up in other markets too? 5 Is anything left after costs and taxes?

Screening patterns with the five questions

The verdicts are an educational example compiled from the S&P 500, KOSPI and KOSDAQ figures calculated directly in Lessons 1–6 of this course and from sources cited in the plan. 'Partial' means the answer depends on the conditions.

Quick checkAt which of the five questions does the day-of-the-week effect get caught first?
Questions 1 (cause) and 3 (last 10 years). It is hard to explain in one line why Monday should be weak, and in fact the S&P 500's average Monday return changed from −0.128% a day in 1950–1989 to +0.020% since 2006.

3 · DataWeakened patterns and surviving patterns

We calculated the two periods, 1950–1989 and 2006 onward, in the same way and re-judged four patterns. All are price-index figures calculated directly from Yahoo Finance ^GSPC daily closes.

S&P 500 pattern scorecard · 1950–1989 vs. 2006 onward

Tap a row. You will see the numbers for both periods and the reason for the verdict.

Calculated directly from Yahoo Finance ^GSPC daily closes (price index, excluding dividends and costs, retrieved 2026-10-10). The turn-of-the-month share is based on log returns (4 trading days = 19% of all trading days). Santa rally: 40 observations in 1950–1989, 20 in 2006–2025. Six-month gap: 40 observations in 1950–1989, 20 in 2006–2025. September: 40 observations in 1950–1989, 21 in 2006–2026.

Of the four patterns, only September weakness kept its strength (−0.54% → −0.50%). The share of the four turn-of-the-month trading days all but vanished, from 90% to 23%. The Santa rally went from an average of +1.72% and an 82% win rate (40 observations) to +0.55% and 70% (20 observations), and the average gap between Nov–Apr and May–Oct narrowed from +5.85 percentage points to +2.11 percentage points. Looking at just the 10 years of 2016–2025, the gap was −0.15 percentage points, with May–Oct slightly higher. Notably, the shorter and easier a pattern is to trade, the sooner it weakened.

There is no confirmed explanation yet for why September remains. That it survives without a clear cause is also an exception to the principle that 'a pattern needs a cause to last.' So September weakness is also not used on its own, only as reference information for lowering expected returns.

Quick checkThe Santa rally win rate fell from 82% to 70%. It is still higher than the win rate of any random 7 trading days (58%), so is it usable right away?
Look at the size of the average and the sample too. The post-2006 average is +0.55%, so what is left after costs is small, and the sample is only 20 observations. It is judged 'partial' on questions 3 and 5 of the five.

4 · TwistThe calendar is a filter, not a signal

It is easy to jump to the conclusion "just buy in Nov–Apr," but 76 years of numbers tell a different story. From November 1950 through 2025, holding the S&P 500 only in Nov–Apr and sitting in cash the rest of the time grew about 111x, while buy-and-hold grew about 351x. Trading only the strong half-year did not beat buy-and-hold (0% interest on cash, excluding costs and taxes).

So is seasonality useless? If you add the as an invalidation condition to the same calculation, a different use appears. The rule 'even in Nov–Apr, sit out if the prior day's close is below the 200-day line' lowered the cumulative multiple (about 56x) but cut the sharply, from −56.8% (buy-and-hold) to −19.7%. Seasonality and the trend filter worked as tools to manage risk, not tools to boost returns.

Simple educational backtest · four rules

Calculated directly from Yahoo Finance ^GSPC daily closes. The 200-day line decision compares the prior day's close with the 200-day simple moving average through the prior day (no future information used). The Nov–Apr rule counts returns only on trading days in those months. Cash periods earn 0% interest; dividends, trading costs and taxes are excluded. This is a simple educational calculation and does not represent actual investment performance.

Changing the period brings one more twist. In 1950–1989 the '200-day line only' rule (8.13% a year) was slightly ahead of buy-and-hold (7.94% a year), but in 2016–2025 buy-and-hold, at 12.85% a year, was higher than any rule. Filters also depend on the era. So rather than fixing on one rule, first decide why you use that rule (to manage drawdown or to pursue returns).

Common misconceptions

"It's a verified pattern, so I can just follow it?" Even patterns verified in papers weakened by an average of 58% after publication. Verification explains the past; it is not a promise about the future.

"Do I profit by short selling in weak periods?" May–Oct also averaged +2.08% in 1950–2025, a positive number. The answer in weak periods is not to bet the other way but to cut position size or delay new purchases.

"Does stacking several filters make it better?" The more conditions you add, the more the rule fits only the past. This is . Keep only conditions whose cause you can explain.

Quick checkAfter adding the 200-day invalidation condition, returns fell and maximum drawdown fell too. Is this rule a failure?
It depends on the purpose. If the goal was to maximize returns, buy-and-hold was better, but for an investor who has trouble tolerating big losses, the rule that cut drawdown from −56.8% to −19.7% may fit better. Decide the purpose before you build the rule.

5 · One thing to practiceCondition · action · invalidation: three lines

Now write your own rule in three lines. The condition is the part where you set a period with the calendar and confirm with a trend signal; the action is the part expressed as position size or pace rather than an all-in trade; and is the situation in which you don't use the rule. For example: "If it is Nov–Apr and the index is above the 200-day line → raise the contribution to 1.2x the usual → if it drops below the 200-day line, go back to the usual amount." Pick from the choices below or write your own to create a rule card, then check it against the five questions. Today's assignment is to save this card and compare it with the last 10 years (2016–2025) in the six-month cycle table from Lesson 2.

Build a three-line rule

The rules you enter are stored only in this device's browser. The examples are for education and do not recommend any particular trade.

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

Quick checkHow would you rewrite "sell everything in September" as a three-line rule?
Turn the action into position size and add an invalidation. For example: "If it is September (condition) → don't buy new positions all at once; split purchases over four weeks (action) → if the index is above its 200-day line and earnings outlooks are improving, buy as usual (invalidation)."
The one sentence left from Lesson 7Three lines: condition, action, invalidation

✓What we learned today

  • The returns of 97 stock-return predictors shrank by an average of 26% in out-of-sample periods and by an average of 58% after publication (McLean & Pontiff, 2016). Seasonality is no exception.
  • The S&P 500's four turn-of-the-month trading days' share of gains fell from 90% (1950–1989) to 23% (since 2006), the Santa rally from +1.72% and 82% to +0.55% and 70%, and the six-month cycle gap from +5.85 percentage points to +2.11 percentage points. Only the September average (−0.54% → −0.50%) stayed the same.
  • Before using a pattern, ask the five questions: Can you explain the cause? Is the sample at least 15 observations? Is it still valid in the last 10 years? Does it exist in other markets? Is anything left after costs and taxes?
  • The rule of holding only in Nov–Apr and sitting out when below the 200-day line cut the maximum drawdown from −56.8% (buy-and-hold) to −19.7% for November 1950–2025, but its cumulative multiple was lower, 56x versus 351x. The filter was a tool for managing risk, not for boosting returns.
  • Write seasonality as three lines: condition, action, invalidation. The calendar only sets the window, entry is confirmed by trend, and the action is expressed as position size rather than an all-in trade.
Continues in the next lesson · Lesson 8People who made money with the calendar

You've built the frame for the three-line rule. In the final Lesson 8, we look at how six masters known for the calendar each filled in this frame, and the one thing they have in common.

Quiz5-question quiz

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

Q1. Which statement matches what McLean & Pontiff (2016) reported?

Q2. Comparing 1950–1989 with 2006 onward, which S&P 500 pattern kept almost the same strength?

Q3. Which of these is not one of the five questions?

Q4. Which statement is correct about the 1950 November–2025 result of the rule that holds only in Nov–Apr and sits out when below the 200-day line?

Q5. Which is the most fitting 'action' for a three-line rule?

NextTerms · Next lesson · Related tools

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.

Invalidation conditionInvalidation rule

A condition, set in advance, for situations in which a trading rule is not applied. For example: "Even in a strong period, don't use the seasonality rule if the index is below its 200-day moving average."

200-day moving average200-day moving average

A line connecting the average closing price over the most recent 200 trading days. It is a standard trend gauge: when an index is above this line, the long-term trend is read as up; when below, as down.

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.

Out-of-sample testingOut-of-sample test

This means checking whether the same pattern also shows up in a period other than the one where it was found (especially a recent period) or in another market. It is the basic way to filter out data-mining bias.

Data mining biasData-mining bias

The phenomenon where, if you search through many periods and indices, a plausible-looking pattern appears purely by chance. It is filtered out with a causal explanation, out-of-sample periods, and data from other countries.

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.

Maximum drawdownMDD, Maximum Drawdown

The percentage a price falls from its previous peak to its lowest point. Calculated as (bottom ÷ peak − 1). In the 2008 crisis, KOSPI fell 54.5% and the S&P 500 fell 56.8%.

Put what you learned to use right away

Return & CAGR Calculator→Position Sizing Calculator→Macro Map→

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 2026-10-10). The backtests are simple educational calculations, and interest on cash periods is set to 0%.