At 12:30 UTC today the pre-US window fires and there is no American number in it. What arrives is Canadian consumer prices for August — and the two vendors this desk reads most are forty basis points apart on the monthly consensus and a full percentage point apart on the year. One says +0.3% and 3.0%. The other says −0.1% and 2.0%. Statistics Canada’s own July release settles half of it in a single line, and the half it settles shows one of those numbers is not a consensus at all. Meanwhile London has already run USD/JPY to 154.74 without waiting for any of it.
Forty basis points on the month, a hundred on the year
Statistics Canada publishes August consumer prices at 12:30 UTC. Every source agrees on that time and on the prior: July rose 0.5% month on month. From there they diverge.
FXStreet’s preview, published 07:30 UTC this morning, has the consensus at +0.3% month on month and 3.0% year on year, described as unchanged from the previous month. InvestingLive’s week-ahead has it at −0.1% month on month and 2.0% year on year, also described as unchanged. Forex Factory’s Monday grid, which this desk read at 06:20 UTC, agrees with InvestingLive at −0.1%.
Do the containment check before calling this a conflict, because we have been caught by a false one before: a median that sits inside another vendor’s range is not a disagreement. This is not that. Plus three tenths and minus one tenth do not overlap. Three point zero and two point zero do not overlap. Two of the three vendors agree on the month, none of them flags the other, and a reader who picks a page at random gets a number that is either forty basis points or a hundred basis points away from what the next page says.
The month-on-month gap has a plausible innocent explanation we cannot confirm: Statistics Canada publishes both an unadjusted and a seasonally adjusted monthly change, and for July those were +0.5% and +0.3% respectively. A consensus built against the adjusted series and one built against the unadjusted series are not the same number and should not be printed under the same label. We are offering that as the likeliest reconciliation, not as a finding.
The year-on-year gap is not a forecast disagreement. It is a field.
The annual rate is settleable, and the issuing agency settles it. Statistics Canada’s own release of 17 August states that the Consumer Price Index rose 3.0% year over year in July 2026, following a 2.8% gain in June. The all-items index excluding gasoline rose 2.2%, for the third consecutive month.
So the July headline annual rate was 3.0%. FXStreet describing a 3.0% consensus as unchanged from the prior month is arithmetically coherent. InvestingLive describing 2.0% as unchanged is not — against a 3.0% prior, 2.0% would be an enormous decline, and its own table lists it as flat.
A third reader agrees with the first. Brown Brothers Harriman, in a note carried by FXStreet at 07:42 UTC, expects headline inflation at 3.0% year on year against 3.0% in July, core excluding food and energy at 2.1% against 1.9%, and the policy-relevant trim-and-median average at 1.95% for a second straight month. BBH frames the release as a test of the Bank of Canada’s warning that upside risks to inflation have increased, and puts a 25 basis point hike to 2.50% on 28 October at 77% priced — supported by core above 2%, vulnerable to dovish repricing below it. That is one named house and one vendor at 3.0%, with the issuing agency’s own July number behind them.
Look at what else is in that table and the source of the 2.0% is not hard to find. The Bank of Canada’s median core measure ran at 2.0% in July. The trimmed mean ran at 1.9%. Both appear in the same vendor’s block, correctly labelled. The headline annual line appears to be carrying the median core figure. That is not a forecast this desk disagrees with. That is a value in the wrong field.
We flagged this exact failure mode three days ago on a different dataset, when the same publisher listed the University of Michigan sentiment prior as 51.0 — which was the consensus, not the 51.7 actual. That was one instance and we said so. This is the second in four sessions, on an unrelated release, and the pattern is now worth naming: the numbers from this source have been reliable, the labels attached to them have not. If you scrape it, you are not importing a wrong forecast. You are importing a correct number filed under the wrong name, which is the version no validation rule catches.
London already priced the day, and it did not need Canada to do it
Tokyo handed London a range of 153.37 to 154.14. London has taken USD/JPY through the top of it. Minkabu’s session notes mark the pair at 154.30 by 06:59 UTC, 154.53 by 08:36, and 154.74 at 10:18 — the highest level since 7 September — trading 154.68 at the time of that note, with the dollar index at its own 3 September highs and sterling down at 1.3473, an eight-month low.
One-week USD/JPY implied volatility is quoted at 11.6%, matching its 4 August high. The stated drivers are a bid for dollars on geopolitical risk and positioning for the start of a US hiking cycle on Wednesday. Neither of those is a Canadian inflation story, and the pair moved sixty pips through the London morning without a single scheduled release.
That is the useful observation for anyone running a news filter today. The window you have blacked out at 12:30 contains a release that does not price your pair. The move you actually got came in an unmarked hour. If your risk budget for today was built around the 12:30 slot because the pre-US slot is usually when America publishes, you spent it on the wrong sixty minutes.
The expiry ladder we said did not exist has published — under a third name
At 06:20 UTC this desk checked the publisher’s order-flow index for today’s 10:00 New York cut option expiries and found nothing. We published that as a checked negative with the time of the check, and said the pre-US slot would look again.
It exists now. A 14 September ladder is live, and it carries a third byline — after the two different authors whose 11 September ladders disagreed with each other by $200 million at USD/JPY 154.00. That is three names on the same recurring dataset inside four sessions. We could not read the notional table itself: the copy we reached carries the strike commentary but not the figures, which is the documented failure where the numbers sit inside an image that does not extract. So we are reporting that the ladder published, and that we could not read it, and we are not quoting a single strike from a page we did not fully see.
What we will say is that spot has changed sides. On Friday USD/JPY was at 154.08 with $2.86 billion — or $3.06 billion, depending which of that publisher’s pages you believed — struck at 154.00 just above it. Today the pair is at 154.68, above the same figure. A strike the market was pressing up into on Friday it is now sitting on top of, and that reverses the sign of whatever pinning effect you thought you were trading.
What this does not tell you
It does not tell you what August Canadian inflation will be. Three days ago we closed a US core CPI consensus dispute at +0.2%, two independent vendors against one incoherent table, and the print came in at +0.3%, between them. Adjudicating a vendor disagreement establishes what the vendors think. It has now failed to establish the number twice in a week, and we are not going to pretend today is different.
It does not establish which monthly consensus is right. We have offered the seasonal-adjustment reconciliation as a hypothesis and neither vendor states which series it is quoting, which is itself the problem.
It does not prove the year-on-year error is a field mix-up rather than a genuine, if strange, forecast. What we have shown is that 2.0% is the Bank of Canada median core figure, that it appears correctly labelled elsewhere in the same table, and that describing it as unchanged from a 3.0% prior is internally inconsistent. That is an argument from the published tables, and we are presenting it as an argument.
It does not tell you today’s expiry notionals, because we could not read them. And the London levels are one vendor’s session marks at three timestamps, not a consolidated tape.
Related
- FX events calendar — the week’s scheduled releases in UTC
- Signals
- EA presets
