MEASURED · 16 SEPTEMBER 2026

Is gold seasonality real?

For direction, no. January averages +3.61% and rose in nine of the last twelve years, the strongest gold seasonality claim there is, and it fails a significance test once you count the twelve months tested. How far gold moves does change with the month, and that result holds up.

The usual gold seasonality chart hides its sample size

Search for gold seasonality and you get a bar chart of average returns by month. The taller bars get a story: January buying from Asian physical demand, a weak September, a year-end rally. The chart is usually arithmetically correct and almost never states how many observations sit behind each bar, or whether the differences between bars mean anything.

Behind each bar sit about twelve. Even with 3,901 daily bars going back to 2014, splitting by calendar month leaves about 12.6 observations per bar. Monthly gold returns have a standard deviation over 4%, so the standard error on a twelve-sample average is more than a full percentage point. Most of the height difference between those bars is measurement noise.

Every month’s return, with its error bar

151 complete months, 2014-01-14 to 2026-09-16. We dropped the partial month at each end, because a half-month counted as a full one distorts its calendar slot. Returns are logarithmic so they add up correctly.

MONTHNMEANMEDIANWIN RATE±SEDAILY RANGE
January12+3.61%+4.23%75%1.221.208%
February13+1.29%-0.21%46%1.581.313%
March13+0.09%-0.12%31%1.581.524%
April13+1.31%+1.51%62%0.801.317%
May13+0.06%+0.15%54%0.921.184%
June13-0.52%-1.39%38%1.511.208%
July13+0.80%+0.61%62%1.221.147%
August13+1.31%+0.04%54%1.001.214%
September12-1.27%-2.76%25%1.361.074%
October12+0.97%+1.58%58%0.891.182%
November12-1.28%-1.20%25%1.421.212%
December12+1.26%+1.47%58%0.701.104%

Read the SE column next to the mean column. January’s +3.61% carries an error bar of ±1.22 points. Most months cannot be told apart from zero, let alone from each other.

A permutation test, because monthly returns are not normal

Monthly returns are not normally distributed, so instead of assuming a bell curve we used a permutation test. Shuffle the calendar labels across the 151 observed returns, recompute every monthly average, and repeat 20,000 times. That shows which seasonality patterns data with no seasonality in it produces, and the real result is compared against them.

Tested on its own, January returns p = 0.006. By the conventional 5% threshold that is a strong result, and published as it stands it would make a convincing article.

It would also be wrong, for a reason that has nothing to do with gold.

January was picked from twelve tests

January was picked after looking at all twelve months and taking the most extreme. Twelve tests at a 5% threshold should produce 0.6 false positives from noise alone. Finding one is exactly what a dataset with no pattern produces.

So the right test asks whether January is more unusual than the most unusual month that noise hands us. The permutation answers that directly, by recording the strongest monthly average in every shuffle:

January, tested alone

p = 0.006

Looks significant

Corrected for 12 tests

p = 0.073

Not significant

Months passing raw test

1 of 12

0.6 expected by chance

Gold return seasonality does not survive its own sample size, at least in this data over 12.6 years. The effect could exist and be too small to detect with twelve samples per month. That is a fair objection, and a pattern you cannot detect is also one you cannot trade.

Volatility seasonality is real

Ask how far gold moves instead of which way, and the statistics change completely. Daily range is measured from every one of the 3,871 days in the sample, roughly 323 observations per calendar month instead of twelve.

Widest month

March 1.524%

Calmest month

September 1.074%

Permutation p

0.0000

No shuffle matched it

March runs 42% wider than September, and the permutation test on that spread never once produced a gap that large across 20,000 shuffles. The same data that cannot tell you which month gold rises in tells you very clearly which months it moves in.

This result changes how you size a trade. A stop set at a fixed dollar distance is a far tighter stop in March than the same number is in September, and a position sized without reference to volatility carries different risk in each. Size from current measured volatility instead of a seasonal calendar. That gets this right automatically and keeps working in a month that breaks the pattern. The same reasoning is behind measuring how the gold spread and range move by hour.

Why seasonality produces so many false positives

The data is free, the twelve buckets are handed to you, and every bucket comes with a ready-made story: Indian wedding season, summer liquidity, year-end positioning. When a month looks strong there is always an explanation available, and an explanation feels like evidence.

The defence is deciding what counts as proof before looking. Here that meant fixing the correction for twelve simultaneous tests in advance, which turned a publishable January result into a null one. We apply the same rule to our own trading system. An earlier version scaled its risk-reward ratio by market regime, validated well in-sample, and was removed after out-of-sample testing showed it was fitting noise. Removing a feature improved results. We also wrote a test down in advance and ran the four strategies in the bot on their own, on gold from 2019 to 2024, years they were not tuned on, and none of the six years was positive. How many trades a track record needs before it means anything is the same problem in another form.

Limits

One symbol, one broker, 2014-01-14 to 2026-09-16. Gold spent much of this window in a strong bull trend, which lifts most months and could hide a genuine seasonal effect underneath. A longer history would test more. It would also span periods when gold traded under entirely different monetary regimes, so more years would not simply mean more evidence.

The volatility result is much better powered than the return result, and it is still one instrument. None of this says seasonality is impossible. In this sample, the return pattern cannot be told apart from chance once the multiple-testing problem is handled. The full numbers for every month are published alongside our other studies, including the columns that would let anyone prove this argument wrong.

Common questions

Is there a best month to trade gold?

Not for direction. Across 151 complete months of XAUUSD, January had the strongest average return at +3.61% and won 75% of the time. That rests on twelve observations, and once you correct for testing all twelve months at once it is indistinguishable from chance (family-wise p = 0.073). Exactly one month passed the uncorrected test, and pure noise produces 0.6 across twelve tests. The months do differ clearly in how VOLATILE they are, and that difference holds up statistically.

Does gold really go up in January?

It rose in nine of the last twelve Januaries, averaging +3.61%, and that is the most common gold seasonality claim. Twelve samples of a series whose monthly standard deviation is over 4% give a standard error of 1.22 percentage points, and January was picked out of twelve candidates after seeing the data. Shuffle the month labels at random and a month this strong appears 7.3% of the time by luck alone.

Is gold weak in September?

Not by more than chance. September was the weakest month in our sample at -1.27% with only a 25% win rate, which matches the folklore, but its uncorrected p-value is 0.359. The more useful and far better-evidenced fact is that September is the CALMEST month by daily range, at 1.074%.

Which month is gold most volatile?

March, at an average daily range of 1.524% of price, against September at 1.074%: 42% wider. This result is solid because volatility is measured from 3,871 individual days instead of 151 months, so roughly 322 observations back each month instead of twelve. The permutation test on the spread returns p = 0.0000.

Why do so many gold seasonality charts disagree with this?

Most of them plot average monthly returns without stating the sample size or running any significance test. With twelve observations per bar, such a chart will always show some months clearly higher than others, because that is what random data looks like when you average it in small groups. The arithmetic on those charts is usually right. They never say whether the pattern means anything.

Can I use volatility seasonality in a trading system?

For sizing, yes, and on better evidence than return seasonality. It gives you no direction. A 42% difference in average daily range between March and September changes position size and stop distance: a stop set in fixed dollars is a much tighter stop in March than the same number is in September. Sizing from current measured volatility does the same job without depending on the month at all.