US CPI is published rounded to one decimal, but the market quickly calculates the unrounded figure. A core m/m reading of 0.3 can be 0.25 or 0.34, and the difference matters.
Why the detail counts:
Consensus is often at 0.3, so an unrounded 0.26 is effectively a soft print and 0.34 a hot one.
Annualised, 0.25 vs 0.35 is roughly 3% vs 4.2% inflation — a big difference for rate expectations.
Components like shelter or services excluding shelter can change the interpretation.
Practical implications:
The first move reacts to the rounded figure; a second move often follows once the unrounded number circulates.
Reversals within the first few minutes are common when the rounded and unrounded readings tell different stories.
Do you watch the unrounded CPI figure, and have you seen the second move it causes?
FXS Prop Desk wrote: Annualised, 0.25 vs 0.35 is roughly 3% vs 4.2% inflation
Small addition on the annualised numbers. Compounding 0.25 monthly gives about 3.0% and 0.35 about 4.3%, so the gap is a touch wider than simple multiplication suggests. More practically, the unrounded figure isn't printed in the release itself. People back it out from the index levels, so the first couple of minutes depend on how fast that calculation circulates, which is part of why the second move happens. It's also worth knowing the consensus is usually a rounded number too, while some forecasters publish to two decimals. Say the median forecast is 0.3 but most of the sell side was at 0.28. An unrounded 0.31 then reads hotter than the headline in line suggests.
FXS Prop Desk, watching the unrounded core m/m is part of my CPI card now. A printed 0.3 that backs out to 0.26 versus 0.34 has produced opposite second moves often enough that I no longer trust the first spike alone.
Fairman's point about backing the figure from index levels is why I wait for the calculation to circulate before I size up. The first two minutes are mostly the rounded headline and algorithms.
I treat a soft unrounded print after a hot rounded one as a reason to stand aside rather than fade immediately. Spreads are still wide, and the second move can reverse again once shelter detail lands.
Annualising 0.25 against 0.35 still frames how hard the dollar can reprice rate expectations. That gap is large enough that I cut size in half until the unrounded number is clear.
I log both the printed and the backed out figures in the journal. Have you seen the second move fail when the components disagreed with the headline story?
FXS Prop Desk, the year on year core figure deserves a place next to the unrounded monthly, because the two can point in opposite directions. The annual rate depends as much on the month that drops out of the window as on the new one.
Say the month leaving the window was 0.45 and the new print backs out to 0.34. The monthly reads hot against a 0.3 consensus, yet the annual rate falls by roughly 0.11 points, and a headline saying core inflation slowed hits the wires at the same moment.
On mornings like that the first spike can follow either number. I have seen the dollar jump on the monthly and then give it all back within a few minutes as the annual headline spread.
Revisions to the prior month are the other piece. A previous month revised from 0.2 up to 0.3 can make an in line print look hotter in level terms, and that detail usually takes even longer to circulate than the unrounded figure.
So I write down the drop out month in my prep the night before. Do you watch base effects on CPI days, or only the monthly surprise?
Consumer price inflation reports contain a headline figure and a core figure, plus breakdowns by category. Markets react to the headline first, then adjust as the components are read.
A headline that matches forecasts can still move the market if core inflation or a key component, like shelter or services, surprises. Those surprises change expectations for central bank policy, which is what currencies actually trade on.
The result is a two-stage reaction. A first spike on the headline, then a second move, sometimes in the opposite direction, once the details are digested. The gap between those moves can be only seconds or several minutes.
For scalpers, that means CPI days reward patience. The most dangerous moment is the first few seconds, when spreads are wide and prices unreliable.
A written playbook for event days, with rules about size, timing and which pairs to avoid, helps avoid improvising under pressure.
CPI releases also come in cycles: one month the market cares about shelter, another about energy, another about goods. The component that matters is the one the central bank is currently worried about, which is knowable by reading the latest meeting statement.
Day-of preparation might include checking forecasts for headline and core, reading what analysts say about the risks around each, and marking the levels where USD pairs would be if the surprise went either way. None of this predicts the result, but it prevents being startled by it.
Afterwards, recording the price at 10 seconds, 1 minute, 5 minutes and 30 minutes builds the personal database that turns anxiety into pattern recognition.
Is there a written playbook for CPI days, and what's the central rule in it?
Fairman wrote: Say the median forecast is 0.3 but most of the sell side was at 0.28. An unrounded 0.31 then reads hotter than the headline in line suggests.
Compounding is the proper way to annualise, thanks. The two-decimal forecasts are the practical point: a headline matching a rounded consensus can still be a surprise against where the desks actually sat. Without access to those forecasts, the direction of the first minute is a reasonable clue that the unrounded number has been worked out and differs from the headline story, which is a reason not to fade the first move on an "in line" print.
That may be a clue about information flow, but I would be cautious about treating the first minute's direction as a reliable proxy for an unrounded value. The move can reflect the headline, positioning, revisions, or other release details as well. If someone reconstructs a two-decimal monthly figure from index levels, the calculation method and precision of the inputs matter; it is useful to label that estimate as a reconstruction rather than an official value. The annualisation also illustrates scale, not a forecast: compounding 0.25% monthly for twelve months is about 3.0%, while 0.35% is about 4.3%. For a CPI playbook, I would log the release timestamp, which figures were actually available at each decision point, spread and execution conditions, and the rule for standing aside. Then compare a predeclared response to the headline alone with one that waits for the detailed table, using executable prices and costs rather than chart midpoints. That can test whether the later move is consistently tradable without assuming it was caused by one hidden decimal. If the data source or timing is uncertain, waiting for the verified components and normal execution conditions is a more defensible rule than guessing from the initial spike.
