Eleven hours ago this desk published that Statistics Canada carries two monthly CPI figures for the same month, that every preview prints one of them, and that nobody says which series the consensus is for. We asked the question out loud. The agency has now answered it, and the answer is worse than the question: September rose 0.1 per cent month over month unadjusted and 0.4 per cent seasonally adjusted. Against a consensus of plus 0.2, the first is a 10 basis point miss and the second is a 20 basis point beat. Same month, same agency, one release, and the sign of the surprise is decided by which series you score — not by the data.
The series choice flips the sign, in all four combinations
The figures, from the agency: the Consumer Price Index rose 3.1 per cent year over year in September after 3.0 per cent in August; it “increased 0.1 per cent month over month in September”; and “on a seasonally adjusted monthly basis, the CPI rose 0.4 per cent in September.” Both monthly sentences are in the same document, labelled, three-tenths of a point apart.
The consensus is itself two numbers. Three readers — two pages from one house and the release vendor — had plus 0.2. A fourth had plus 0.3. Run all four combinations and every one of them is a sign flip:
- Unadjusted plus 0.1 against consensus plus 0.2: a 10 basis point miss.
- Seasonally adjusted plus 0.4 against consensus plus 0.2: a 20 basis point beat.
- Unadjusted plus 0.1 against consensus plus 0.3: a 20 basis point miss.
- Seasonally adjusted plus 0.4 against consensus plus 0.3: a 10 basis point beat.
There is no reading of this release on which the monthly surprise has a determinate sign. The headline was unambiguous — 3.1 per cent against a 3.2 per cent consensus from three readers, a clean 10 basis point miss, and the Canadian dollar traded it as one. But the monthly number, which is the one that moves faster and the one both vendors led with, is not a surprise at all until you name the series. Neither of the two vendors reporting plus 0.1 says which series it is. The agency labels both.
We flagged the right ambiguity on the wrong axis, and we are retracting that part
This morning’s note argued that the plus 0.2 consensus against the unadjusted minus 0.1 prior reads as a 30 basis point acceleration, and that if the forecasters meant the adjusted series — whose prior was plus 0.2 — then plus 0.2 against plus 0.2 would be no acceleration at all. That framing is dead, and the print killed it in one line.
Do the arithmetic both ways. Unadjusted went from minus 0.1 to plus 0.1: an acceleration of 0.2 percentage points. Seasonally adjusted went from plus 0.2 to plus 0.4: an acceleration of 0.2 percentage points. Identical. The acceleration is the one quantity in this release that does not depend on your series choice. We spent an article on the ambiguity in the acceleration, and the acceleration was the invariant.
What is series-dependent is the surprise, which we did not flag. That is the third consecutive month in which this desk has offered a seasonal-adjustment reconciliation on this exact release and had to retire it. On 15 September we offered unadjusted-versus-adjusted as an innocent explanation of a 40 basis point consensus gap, and the print showed neither series matched. This morning we offered it as an explanation of an acceleration, and the print shows both series accelerate the same. The underlying observation — that the agency publishes two monthly series and the vendors print one without saying which — survives both retractions and is now primary-sourced twice. The reconciliations built on it keep failing. We will stop building them.
We said the Bank’s three core measures had no consensus. Two of them did.
This morning’s article stated that of the four numbers called core, the only one carrying a September consensus was ex food and energy — the basket the Bank of Canada does not target — and that the three it watches had no number to beat before the 28 October decision. Two-thirds of that is wrong, and the correction belongs under the byline that made the claim.
The release vendor’s page carries CPI-median with an expected 1.9 and CPI-trim with an expected 1.9. Only CPI-common arrived with no published forecast. So the scores exist:
- CPI-trim: 1.8 against 1.9 expected — a 10 basis point miss, down from a 1.9 prior.
- CPI-median: 1.9 against 1.9 expected — in line, down from a 2.0 prior.
- CPI-common: 2.4, no consensus published, down from 2.6.
Our claim was bounded — it said no consensus appeared “anywhere in this sweep” — so it was not a fabrication, and that is the most generous thing we are going to say about it. The process lesson is specific and worth more than the correction: a negative claim about whether a consensus exists cannot be closed from previews, because the release vendor publishes the consensus alongside the print. We were looking in the wrong class of document four hours before the release, not at too few sources. A preview page tells you what the forecasters said about the headline; the release page is where the forecasts for the secondary fields turn up.
And the substance matters more than the bookkeeping. Trim at 1.8 and median at 1.9 average 1.85 per cent, below the two per cent midpoint the Bank targets, with both measures lower than the month before. That is the pair of numbers the 28 October decision runs on, and they went the dovish way.
The headline went up and all four core baskets went down
Put the month in one line. Headline: 3.0 to 3.1 per cent, up 0.1 points. CPI-common: 2.6 to 2.4, down 0.2. CPI-trim: 1.9 to 1.8, down 0.1. CPI-median: 2.0 to 1.9, down 0.1. And from the agency itself, excluding gasoline: “the CPI rose 2.2 per cent, after increasing 2.4 per cent in August” — down 0.2.
So the headline accelerated while every underlying measure decelerated, and the divergence between headline and ex-gasoline is 0.3 percentage points in the same month. The entire acceleration, and 0.2 points more, is gasoline. September’s core spread, across the three measures the Bank actually watches, is 0.6 percentage points — 2.4 against 1.8 — which is narrower than August’s 70 basis point four-basket spread and still wide enough that “core inflation” is not a number.
If you trade the Canadian dollar off the headline you traded a 10 basis point miss, correctly. If you are positioning for 28 October off core, you are looking at a month in which the Bank’s own preferred gauges fell on every one of the three and the energy component did all the damage to the top line. Those are not the same trade, and this release separates them more cleanly than most.
The core the vendors printed is not one the Bank targets
Both vendors reported a core figure of 2.2 per cent, easing from 2.4. One called it “BOC core” and gave a monthly reading of 0.0 against a plus 0.1 prior; the other wrote that annual core eased to 2.2 from 2.4. Neither names the underlying measure.
It is not any of the three the Bank of Canada targets. Those printed 2.4, 1.8 and 1.9. It is, however, numerically identical to the agency’s ex-gasoline series in both September and August — 2.2 after 2.4 — which is the second consecutive month in which that coincidence has held, after the 2.4-equals-2.4 pairing we named in August and declined to build on.
We are declining again, and the reason is unchanged: ex food and energy is also a candidate, we have no September reading for it, and two months of identical values across two different exclusions is suggestive rather than conclusive. What we will say flatly is the part that costs a reader money. A trader who sees “core eased to 2.2 per cent” and concludes that this is what the Bank watches is wrong by between 30 and 40 basis points in either direction, and the label on the field is what misled them.
What this does not tell you
Both monthly figures, the year-over-year figure and the ex-gasoline series are from the issuing agency and need no second reader. Everything else here is secondary. The three Bank of Canada core measures and their two consensuses come from one release page; we did not reach the agency’s own table for them, which is the second consecutive month that table has failed to extract for us. Until a second reader confirms them, treat 1.8, 1.9 and 2.4 as one vendor’s numbers.
The plus 0.3 monthly consensus is a single reader against three at plus 0.2, and we have not established that it is an error rather than a different survey. We cannot tell you which series any of those forecasts covers, which is the whole point and is not something a print can settle. We have not identified what the vendors’ 2.2 per cent core actually is. And nothing here forecasts 28 October: softer core measures are consistent with a hold at 2.25 per cent, which is what the readers in our sweep expect, but we are not putting a probability on it — one publisher priced the Bank’s year end twice in three hours this morning, eleven basis points apart, and that spread has not been resolved either.
Related
- FX events calendar — release times in UTC, including the 28 October Bank of Canada decision and the 16 November October CPI.
- Signals — what a release with two legitimate monthly prints does to a beat-or-miss rule.
- EA presets — news-window and field-definition filters.
- Canada’s Prior Is Minus 0.1 and Plus 0.2. The Consensus Only Fits One of Them. — the article this one corrects and scores.
- Our September note on the Canadian consensus gap — where this argument started, five weeks ago.
