What the CPI release actually measures

Three indexes, one name, different jobs

CPI-U covers all urban consumers and is what the headline number refers to. CPI-W covers urban wage earners and clerical workers, a narrower population that weights transportation and food more heavily, and it is the index Social Security cost-of-living adjustments are computed from. Chained CPI-U, C-CPI-U, allows for substitution between categories as relative prices change, runs slightly lower over time, and is what federal income tax brackets are indexed to.

The distinction is not academic. A retiree’s benefit and a taxpayer’s bracket move on two different indexes, neither of which is the number on the front page. When a contract, a lease, or a pension formula references "CPI", the first thing to establish is which one, and whether it uses the seasonally adjusted series or not.

Shelter is most of the story and it arrives late

Shelter is about a third of the CPI-U basket, and the largest piece of it is owners’ equivalent rent, which is not a price anyone pays. It is an imputation: what an owner-occupied home would rent for, estimated from observed rents on comparable units. Because leases turn over gradually and the survey samples each unit every six months, this component tracks the market with a lag measured in quarters, often around a year.

That single fact explains most of the confusion in commentary about CPI. When market rents are falling fast and the CPI shelter component is still rising, neither the market data nor the official index is wrong. They are measuring different things at different points in a pipeline. If you want a current read on housing costs, CPI shelter is the wrong instrument, and if you want to know what CPI will do over the next few quarters, market rents are one of the better leading inputs you have.

What gets revised, and why that decides which series to reference

The unadjusted CPI index levels are not revised. The seasonally adjusted series are: seasonal factors are recalculated annually and the adjusted data for the prior several years is revised with them. This has a practical consequence people discover at the worst moment. Any contract, escalation clause, or settlement rule that references a seasonally adjusted CPI figure references a number that can change after the fact. Reference the unadjusted index level and the release month, and the value is permanent.

The same logic drives how JMM writes resolution rules against BLS data. Naming the series, the vintage, and whether an initial release or a later revision controls is the difference between a scoreable question and an argument.

• Confirm which index the number refers to: CPI-U, CPI-W, or C-CPI-U.
• Read the unadjusted index level and the twelve-month unadjusted change.
• Check shelter and core separately before drawing a conclusion about either.
• For any contract or rule, reference the unadjusted series and name the release month.

Release morning, in the order that helps

The release lands at 8:30 in the morning Eastern time on a schedule published a year in advance. The market reaction happens in the first seconds and is driven by core against consensus, so if you are reading the release you have already missed the trade and should be reading it for something else.

Read core, which excludes food and energy, not because those do not matter to anyone’s budget but because they are volatile enough to drown the signal in any single month. Then read shelter, then read services excluding shelter, which is the component most closely tied to wages. Our position: the annualised one-month change is the single most over-quoted number in economic data. One month of a survey-based index, seasonally adjusted with factors that will later be revised, multiplied by twelve, is not a measurement of anything. The twelve-month unadjusted change is slower and it is real.