Sell in May on the ASX: the Halloween effect, tested on 33 years of the ASX 200

"Sell in May and go away" is one of the oldest sayings in investing, and it is one of the few that has held up in peer-reviewed research. On the ASX 200 from 1993 to 2026, the November to April half averaged a rise of 4.54%, against 1.33% for the May to October half, and the winter half came out ahead in 22 of 33 years. That is the same direction the published research found, of a similar size, although on one index over 33 years it sits just short of the usual bar for statistical proof.

This is research and education, not financial advice.

The research: 36 of 37 markets

The study that put this saying on a serious footing is "The Halloween Indicator, 'Sell in May and Go Away': Another Puzzle" by Sven Bouman and Ben Jacobsen, published in the American Economic Review, volume 92, issue 5, pages 1618 to 1635, in December 2002. That is a peer-reviewed result in a mainstream economics journal, not folklore dressed up.

The authors compared two halves of the year:

• Winter half: the start of November to the end of April (hence "Halloween", 31 October, as the switch date).
• Summer half: the start of May to the end of October.

According to the CFA Institute's summary of the paper, across 37 developed and emerging markets from 1973 to 1998, returns in the November to April half were on average significantly higher than in the May to October half in 36 of the 37. The authors looked at a range of possible explanations and, as the title says, called it a puzzle: none of them settled why it happens.

The work did not stop there. A later study by Jacobsen and Cherry Zhang, as summarised by the CFA Institute, took the test to 108 stock markets and more than 300 years of UK data. They found the effect in 81 of the 108 markets, with November to April returns on average 4.52 percentage points higher than May to October.

A word on the label. In Australia, November to April is our summer and May to October our winter. The research uses Northern Hemisphere names, so we keep "winter half" for November to April to match the papers, and say the months every time.

Why test it again on the ASX

Two reasons. First, any published effect should be checked on data the authors did not use. Bouman and Jacobsen's sample ended in 1998, and their paper came out at the end of 2002. Most of our test period comes after that, so it is close to an out of sample check: if the effect were only a quirk of their data, it should have faded once everyone could read about it.

Second, Australian investors want to know about the ASX, not a global average.

How we ran the test

We used daily closes of the ASX 200, Yahoo Finance ticker ^AXJO, fetched on 12 October 2026. Yahoo's series for this ticker starts on 23 November 1992.

For each year we took two numbers: the last close in April and the last close in October. From those:

• Summer half change = last October close divided by last April close, minus one. For example, May to October 2025 ran from 8,126.2 (30 April 2025) to 8,881.9 (31 October 2025), a change of +9.30%.
• Winter half change = next April's last close divided by October's last close, minus one. November 2025 to April 2026 ran from 8,881.9 to 8,665.8 (30 April 2026), a change of -2.43%.

The test runs from the end of April 1993 to the end of April 2026. That gives 33 summer halves (1993 to 2025) and 33 winter halves (November 1993 to April 2026). The half running from May 2026 is not finished, so it is left out.

Two limits to state up front. ^AXJO is a price index, so dividends are not in these numbers. And this is a simple comparison of averages, not a copy of the paper's regression method.

What our run shows

Paired year by year, summer then the winter straight after it, the winter half did better in 22 of the 33 years.

So the ASX 200 shows the effect, in the direction the research found. The gap between the two averages is 3.21 percentage points per half year, which sits in the same range as the 4.52 point gap in the 108 market study.

The rise came from both halves. The summer half was positive on average too, and it rose in 19 of 33 years. Sitting out of the market from May to October would not have dodged losses most years, it would have given up a smaller average gain.

Is it bigger than chance?

This is the honest part, and it matters.

Each half year swings a lot. The standard deviations above are 8.5 and 9.7 points. With 33 halves on each side, the standard error of the gap between the two averages works out to about 2.2 points (the square root of 8.5² / 33 plus 9.7² / 33). The measured gap of 3.21 points is about 1.4 times that. The usual bar for "very unlikely to be luck" is about 2 times.

The year by year count is a little stronger. If winter and summer were a coin flip, we would expect winter to win about 16 or 17 times out of 33. It won 22. Using the normal approximation to the coin flip, a result that lopsided or more turns up by chance roughly 4% of the time, close to the usual 5% line.

Put plainly: on one index over 33 years, the effect is there and points the right way, but it is not strong enough to call proven from this data alone. That is not a knock on the research. It is the same lesson as every calendar effect. A gap of a few points against half year swings of eight to ten points needs many markets or very long histories to stand out, which is exactly why Bouman and Jacobsen used 37 markets and Jacobsen and Zhang used 108.

A few years carry a lot of weight

Look at the extremes and you can see why one index is not enough on its own:

• 2008: the summer half fell 28.19% as the global financial crisis hit. That one half drags the summer average down by almost a point.
• 2009: the summer half then rose 22.82% in the recovery, one of the best halves in the whole set.
• 2019 to 2020: the winter half fell 17.12%, the COVID crash of February and March 2020.
• 2020 to 2021: the winter half rose 18.53% in the rebound.

Crashes do not respect the calendar. Any seasonal effect is an average across many years, and a single bad year in the "good" half can wipe out several years of edge.

The full table

Every half, so you can check our arithmetic against the closes.

Boundary closes are the last trading day of April and of October in Yahoo's ^AXJO series. Percentages are rounded to two decimals, and the averages are taken from the rounded figures.

One more honest note from the table: the last two winter halves, 2024 to 2025 and 2025 to 2026, both lagged their summers. Two years is far too few to say the effect has gone, but it is a reminder that it never works every year.

How to read this

What the evidence supports:

• The effect shows up in the research. It was found in 36 of 37 markets in a top economics journal, and again in 81 of 108 markets in the larger follow-up.
• The ASX 200 shows it too. Over 33 years, with most of the data after the paper came out, the November to April half averaged about 3.2 points more than May to October, and won 22 years out of 33.
• It is an average, not a signal. In 11 of our 33 years the summer half did better. Nobody can tell you in advance which kind of year this will be.

What it does not support:

• It does not say the market falls in winter. On the ASX the May to October half rose more often than not.
• It does not account for dividends, tax or trading costs, all of which matter if you act on it.
• It does not tell you why. The paper itself calls it a puzzle.

This is the kind of question worth asking of any timing idea: what does the research say, does it hold on fresh data, and is the gap bigger than the noise? Our backtesting primer walks through that process, and the Bartels test post shows one way to judge whether a cycle in a price series is more than chance.

Try it yourself

1. Create a free account. The free plan includes 50 compute units a day.
2. Read the cycle scanner guide to look for repeating cycles in a price series of your choice, and check any cycle you find against chance before you trust it.
3. To repeat this test, take the last April and last October closes of the index you follow for as many years as you can get, and work out each half's change the way we did above.

Report what you find, including the years it fails. A seasonal effect only means something when you look at all of them.

This is research and education, not financial advice. Past seasonal patterns do not guarantee future returns.