Why does every currency pair fall at the New York close?

Because it is 5pm in New York, not because the market is going down. Across 5,952 trading days, 12 of the 12 markets that trade the whole clock close lower in that one hour — USD/CHF closes up just 33.3% of the time, a z of -18.6. Those are the largest numbers anywhere in this archive, by an order of magnitude. They are also worth about one pip, which is the whole story.

First: which hour is it?

Almost every version of this claim you will read pins the New York close to a fixed UTC hour, and that is wrong for months of every year. New York keeps its own daylight-saving calendar, so we resolve 17:00 New York through that city’s clock for every single day in the archive. The result: the New York close falls at 20:00 UTC on 64.6% of days and 21:00 UTC on 35.4% of days. Pin it to one and you are testing the wrong hour for a third of the year.

Every market, and the column that kills it

The last column is the one to read: the size of that hour’s typical move against one round trip in that market. Under 1.00 means the effect is smaller than the cost of taking it.

z is measured against each market’s own up-rate across every hour it trades, never a flat 50% — two of these markets close up more than half of all hours, and scoring them against 50 would invent an effect.

So what is it?

Put the pieces beside each other. The effect appears in every instrument. It runs in the same direction in all of them, including pairs that move opposite each other the rest of the day. It happens at one named minute. And it lands on the same round number — the median bar is about one pip, and exactly −1.00 in several markets.

And it has since been checked from the other side. Our hourly feed is bid-only, but the minute archive we later added carries the ask for 10 FX pairs, so the clean test can be run: in the same hour, does the ask fall too? It does the opposite. The bid closes up in a median 42.55% of New York close hours and the ask in 62.3% — the bid below its own average past our bar in 10 of 10 pairs, the ask above its own in 10 of 10. Two quotes of one price moving apart is the spread widening. Three hours earlier, where nothing settles, their up-rates sit a median 0.7 points apart; at the close, 18.2. The metals have no minute archive here, so for them this is still an inference. The test is in section 9.

⚠ We also measured how much of the move comes back in the following hour, which is the other thing you would expect of an artefact. It does, in most markets — but our estimator for it saturates at 0 and 100%, so the individual figures are unstable and we are not publishing them as numbers. A measurement we do not trust is worth less than the sentence saying we do not trust it.

The hours that DO matter at those times, and what they are good for

The named hours are not all alike, and the split is the same one that runs through this whole site: size is real, direction is not. The two opens are the biggest hours of the trading day — the New York open moves 2.53× a typical hour and the London open 2.23× — and they lean directionally in 1 of 12 and 1 of 12 markets respectively. That is nothing. The London 4pm fix moves 2× and leans in 3 of 12.

So the useful reading of the named hours is a timetable, not a signal: they tell you when the market is awake, which is what you size a stop against. The one hour that produces a huge directional number is the one hour where that number is an accounting entry.

What we would actually say

• Do not trade the New York close on this. 11 of 12 markets lean by less than one round trip, and the median across all of them is 0.66× the spread. The strongest statistical effect we hold is also one of the least tradeable.
• Beware anything sold on a z-score alone. This is what a huge z looks like when the underlying move is one tick. Sample size makes a small, meaningless difference detectable; it does not make it worth having.
• Resolve the hour through the city’s clock. Pinned to a fixed UTC hour, this effect is measurably weaker, because for months of the year the pin is on the wrong hour.
• Use those hours for size instead. The opens are 2×+ a typical hour and say nothing about direction — that is the axis that survives being tested on years it was not chosen from.

Full method and every figure in section 9 of the evidence page, which qualifies section 8 — and the rest of the tested claims are on the reports index. The data is free to download if you want to redo any of it.

More on when a market actually moves

Every claim we have tested, including the ones that held, is on the reports index.