When does USD/CAD actually rise?

Every hour below is a real measurement: how often USD/CAD closed higher than it opened in that hour of the week, across 142,043 hourly bars. Not signals — base rates, with the sample size on every figure.

The strongest and weakest hours

Averaged across the trading week, the hour that has leaned up hardest is 22:00 UTC — USD/CAD closed higher 54.9% of 5,924 times there, far enough from a coin flip to be beyond chance. The weakest is 20:00 UTC, closing higher only 41.8% of 5,913 times.

Hour × weekday, the whole picture

Colour is the up-rate — red leaned down, teal leaned up. The brass dot marks buckets whose lean is statistically beyond chance at |z|≥3. Most cells have no dot. That is the honest shape of the market: mostly noise, with a few stubborn corners.

The three most stubborn corners of the grid:

• Sunday 21:00 UTC — closed higher 64.6% of 811 times (z=8.3) · rollover hour, see note above
• Friday 20:00 UTC — closed lower 62.0% of 1,182 times (z=-8.3)
• Wednesday 20:00 UTC — closed lower 59.9% of 1,181 times (z=-6.8)

Days and seasons

Thursday has been the friendliest day (50.0% of 28,356), Wednesday the least (49.5%). By calendar month, January leads at 50.6% and May trails at 49.0% — month buckets carry roughly a twelfth of the sample each, so hold them more loosely than the hourly figures.

Did any of it survive a real backtest?

Colours are description. A different question is whether a simple rule built on them — enter at the open of a bucket, exit at its close, pay a realistic spread — would have made money, and kept making it on data it was never fitted to. We ran that test across 5,086 rules on the whole board: 307 finished net positive, and only 214 survived the held-out final third of the history.

3 of those survivors run on USD/CAD. The strategy board lists every one — including the losers, because a leaderboard that only shows winners is lying by omission.

Watch it live

These are the same numbers, updating as each hour begins — free to embed on your own site from the widget builder:

Read it without fooling yourself

• The percentage is the headline, not the story. 54.9% from 5,924 samples is information; the same figure from 40 samples is noise. Every number here carries its n.
• The dots matter more than the colours. A vivid cell with no dot is well inside what chance produces.
• Direction is free; getting paid is not. A reliable lean smaller than your spread costs more than it earns.
• These are frequencies, not forecasts. 19 years of "usually" contains every kind of "except when".

The guide walks through all four traps with live numbers.