A common seasonal story says Japanese companies repatriate foreign earnings before the fiscal year-end on 31 March, supporting the yen in February and March. It is worth testing before relying on it.
How to test:
Collect USDJPY returns for February–March over 15–20 years.
Compare with other months.
Check how many years followed the pattern.
What tests often show:
The pattern appears in some years, not others.
Other factors (rate differentials, risk sentiment) often dominate.
Sample size is small: 20 years means only 20 data points.
Seasonality can be a useful tie-breaker but not a reason to trade on its own.
Another way to test it is to look at shorter windows, for example the last two weeks of March only, and to compare the result with the same window in other months. If the effect is real, it should be visible in a narrower period, not only in a two-month average.
Have you tested any seasonal patterns in FX, and did they hold up?
FXS Prop Desk wrote: Sample size is small: 20 years means only 20 data points.
And it gets worse if the narrower windows get tested too. Once you try February to March, then the last two weeks of March, then the last week, you've run several tests on the same 20 years, and one of them showing an effect is close to expected by chance. A cleaner way is to write down the window and the direction before looking, test only that, and treat anything found afterwards as a new idea that needs fresh years. Also worth splitting by regime. In years where the rate gap between the US and Japan was widening, carry flows can easily swamp repatriation, so a pattern that only shows up in low-carry years is a different claim from a general seasonal.
Fairman's point about writing the window down first is the one I would underline. With 20 years and no effect at all, 14 or more of them showing yen strength happens about 6 percent of the time, so test four windows and the odds of one hitting are close to one in five if the windows were independent.
There is also a competing seasonal that muddies March. Japanese life insurers and pension funds often send fresh money abroad in April, at the start of the new fiscal year, so yen strength into late March can be partly reversed within weeks.
That makes the comparison window matter. A test that measures only February and March can show an effect that a February to April window would mostly cancel.
For a scalper the better documented calendar effect is much shorter. Importers buying dollars into the 9:55 Tokyo fix on gotobi days, the 5th, 10th, 15th, 20th, 25th and month end, gives roughly six observations a month instead of one a year.
Even that one has faded in my notes as more corporates moved to algorithmic execution spread through the day. Has anyone tested the gotobi effect on the last five years separately from the older data?
