Headline CPI vs Core CPI: How Base Effects Move the 12-Month Rate

The short answer

Headline and core CPI are the same survey with different item coverage. Headline is everything; core removes food and energy. BLS’s own explanation for core is that food and energy prices are relatively volatile, so some users strip them out to look at what they call the underlying rate.

The 12-month rate for either measure is a ratio: the index this month divided by the index twelve months earlier. That is why an annual rate can fall in a month when prices rose, and why two indexes can diverge for a year without either being wrong.

The rest of this article is the arithmetic, tested against the published index levels.

Two indexes, one survey

BLS publishes CPI-U, the Consumer Price Index for All Urban Consumers, with a reference base of 1982-84 = 100. The CPI-U population covers households in core-based statistical areas and urban places of 10,000 or more, which BLS describes as over 90% of the US population. It excludes rural nonmetropolitan areas, farm households, military installations, religious communities and institutions such as prisons and hospitals.

Within that survey, several indexes exist:

• All items — the headline.
• All items less food and energy — the core.
• Components — food, energy, and the sub-aggregates beneath them.

The core index is not a different survey or a modelled series. It is a re-aggregation of the same collected prices with two categories removed. That is why it is published with the same monthly cadence and the same level of detail.

What core removes is not trivial in size. Energy is a volatile, frequently purchased category whose price is set in global markets; food is also volatile and responds to weather and supply. Removing both makes the monthly series smoother, at the cost of excluding items households actually buy every week.

The identity behind “base effects”

Let I(t) be the index level for month t. The 12-month rate is:

YoY(t) = I(t) / I(t − 12) − 1

The month-over-month rate is:

MoM(t) = I(t) / I(t − 1) − 1

Substituting the monthly ratios into the annual one gives an exact identity:

1 + YoY(t) = [1 + YoY(t − 1)] × [1 + MoM(t)] / [1 + MoM(t − 12)]

Read that in words: this month’s annual rate equals last month’s annual rate, adjusted by how this month’s price change compares with the same month a year earlier.

If this month’s change is larger than the same month last year, the annual rate rises. If it is smaller, the annual rate falls. If they are identical, the annual rate is unchanged — even if the index rose in both months.

That third case is the one that confuses people. Prices rising at the same rate as a year ago is not “no inflation”; it is inflation at a constant annual rate.

The identity, tested on nine months of data

The table below uses the CPI-U all-items index and the core index from the BLS Public Data API, both not seasonally adjusted. For each month of 2026 it shows the monthly change, the monthly change in the same month of 2025, the difference between them, and the actual 12-month change.

Headline, CPI-U all items

The last two columns track each other to within a hundredth of a point in every month. The identity is not an approximation in the way it is often presented; it is the definition, and the small gaps are rounding in the published index levels.

The March 2026 row is the clearest case. Prices rose 1.05% in the month, and the annual rate jumped 0.84 points to 3.26%. A reader who saw only the annual number would conclude that inflation had accelerated sharply. A reader who saw only the monthly number would conclude that a single strong month had happened. Both are true, and the difference between the two columns is why the annual number moved so much: March 2025 had risen only 0.22%, so the comparison base was unusually low.

Core, all items less food and energy

August 2026 is the row to remember. Core prices rose 0.27% in the month, and the 12-month core rate still fell from 2.48% to 2.45%. Nothing went down. The annual rate fell because August 2025 had risen 0.30%, a tenth of a point more than this August. The base was slightly higher than the current month, so the ratio fell.

That is the base effect in one sentence: the annual rate fell because last year’s month was slightly stronger than this year’s, not because this year’s month was weak.

Why headline and core diverged in 2026

The two indexes in the tables above started 2026 almost level and ended nine months apart.

• January 2026: headline 2.39%, core 2.50%. Headline was marginally below core.
• September 2026: headline 3.74%, core 2.56%. Headline was 1.18 points above core.

The component data explains the gap arithmetically. Energy CPI was −0.14% year over year in January 2026 and +20.45% by September 2026. Food CPI was +2.48% in September. Core was +2.56%.

Headline contains energy; core does not. When a component that is excluded from an index moves from flat to up more than 20%, the two indexes must diverge — that is not a puzzle about methodology, it is what the definitions require.

The base-effect reading adds the timing. Energy’s annual rate rose sharply in early 2026 because the comparison months in early 2025 were weak. By September the base had caught up and the energy annual rate was still above 20%, so the headline rate stayed elevated even as some monthly energy changes moderated.

A reader who tracked only headline would have called 2026 a broad inflation problem. A reader who tracked only core would have called it a stable one at around 2.5%. The data supports a narrower statement: core was stable, and an excluded, volatile component moved a great deal.

What happens next is a separate subject. A CPI print informs expectations for the policy rate, which is not the same thing as the repo rate at which cash actually trades — the repo versus policy rate guide draws that line. Through expectations it also reaches the discount rates that long-dated bonds are sensitive to, which the bond duration guide quantifies, and the shape of the curve, which the inverted yield curve guide treats on its own terms. None of that is a CPI calculation, which is why this article stops at the index.

Three ways to misread the number

First, treating the annual rate as a monthly measurement. A 3.74% annual rate does not mean prices rose 3.74% this month or will rise 3.74% next year. It means the index is 3.74% above its level twelve months ago.

Second, treating a base effect as a forecast. The identity is backwards-looking by construction. A rate that fell because last year’s month was strong tells you nothing about next month, except that next month will face a different base.

Third, mixing seasonal adjustment with unadjusted data. BLS publishes both bases and says an index like this is reported not seasonally adjusted as well as seasonally adjusted. The 12-month changes above come from the not-seasonally-adjusted series. A monthly change from the seasonally adjusted series will not multiply into the unadjusted annual rate, and the discrepancy is largest in months where seasonality is strong — BLS notes that November and December are where the aggregate seasonal pattern can visibly differ from the pattern of its components.

One data gap worth knowing

The CPI-U series retrieved for this article contains no October 2025 observation. That matters for reading 2026 releases: a 12-month change for October 2026 requires an October 2025 base, and the published series does not contain one. Anyone building a base-effect model should check for missing observations before assuming a twelve-month lag is available, because a gap turns a ratio into an error rather than a smaller number.

The same caution applies to revisions. CPI-U index levels can be corrected, and a correction to a prior-year month changes the annual rate for every month that uses it as a base.

A checklist for reading a CPI release

1. Identify the measure. All items or all items less food and energy; CPI-U, CPI-W or C-CPI-U.
2. Identify the basis. Not seasonally adjusted for the 12-month rate; seasonally adjusted for a cleaner monthly read.
3. Read the monthly change and the base together. The annual rate moves by the difference, not by the monthly number alone.
4. Check the components that were excluded. If headline and core diverge, look at food and energy before inventing a narrative.
5. Check the publication history. A missing or revised base month breaks the arithmetic quietly.
6. Do not convert an annual rate into a forecast. The identity describes the past twelve months and nothing else.

Risk note. Index levels for CPI-U all items (CUUR0000SA0), all items less food and energy (CUUR0000SA0L1E), energy (CUUR0000SA0E) and food (CUUR0000SAF1) were retrieved from the BLS Public Data API on 2026-10-28; the latest observation in each series was September 2026. Definitions and seasonal-adjustment guidance come from BLS’s CPI Questions and Answers page read the same day. Percent changes in this article are computed from the published index levels and are rounded to two decimals, so a difference of a hundredth of a point between columns is rounding, not a discrepancy.