Reading a CPI Print

The basket, the weights, and the arithmetic

A price index is a weighted average of category price changes, with the weights fixed to a base period’s expenditure pattern and updated periodically. Contributions are additive: each category’s weight multiplied by its change, divided by a hundred, is its contribution in percentage points, and those contributions sum to the headline rate. That additivity is what makes the release analysable rather than merely quotable — you can rebuild the number, category by category, and see which line did the work. Three features of the basket matter more than the rest. Shelter, which is rent and owners’ equivalent rent, is the largest single category and moves with a lag, because leases reprice on renewal rather than continuously; that lag means shelter inflation can be running at 4.8% when market rents are flat, and it is the reason a turn in housing takes about a year to show up in the index. Energy is the smallest category by weight for most households and the largest by volatility, so its contribution swings by half a point from one month to the next and can move the headline in either direction without saying anything about the trend. And the goods-services split matters because goods prices are set in global markets and services prices are set by domestic wages and rents, so a print driven by core goods falling is a very different statement from one driven by services slowing. The base effect completes the picture and is the most common source of a misleading headline. A 12-month rate is a ratio to the same month a year earlier, so a large change a year ago drops out of the calculation this month and the rate moves without any new price pressure. An energy price that spiked and then stabilised will show a falling year-over-year rate for twelve months after, and headline will fall with it, which is arithmetic rather than disinflation. That is why the useful comparison is month-over-month annualised — the recent pace — set beside the year-over-year figure, and why the core services line is where a lasting turn appears first. Rebuilding the headline — Shelter: 34.4% × 4.8%: 1.65pp, and 63% of the entire headline ← · Core services ex shelter: 26.5% × 3.4%: 0.90pp — the sticky, wage-linked component · Energy: 6.5% × −6.0%: −0.39pp — the line doing the visible work · Headline 2.60% · core 2.67%: core above headline, so the energy base effect is now helping The index is an average over a basket nobody buys and a geography nobody lives in. Renters and owners experience different inflation, energy and food weigh more in a lower-income budget, and a household with children faces education and child-care costs the index spreads thin. The personal rate can differ from the print by two or three points, which is why a client who disagrees with the official number is often right about their own.

What to do with the release, in order

Read it in four passes. First the headline against expectations, because that is what is priced and it produces the immediate reaction. Second, core, because that is what the committee reacts to and because the headline can be moved by a category the committee cannot influence. Third, the contribution table, to find out whether the surprise came from a broad slowing or from one line — a 0.2 point downside surprise driven entirely by used car prices is not evidence of disinflation, and a 0.1 point upside surprise driven by shelter is more worrying than its size suggests. Fourth, the details that lead: new lease rents inside the shelter calculation, which foreshadow the index a year out; wages in the same week’s labour data, because services inflation is where wages show up; and the inflation expectations series, because expectations are what make inflation persistent even when the shock that started it has gone. A release is a single observation, and the analysis is the trend, the composition and the persistence — which is why the market reaction to a print is frequently reversed the following month when the same category reverts. For a portfolio, the translation runs through the discount rate rather than through the price level. A print that reduces expected inflation lowers the expected policy path, which supports long-duration assets; a print that shows persistence above target raises it, which compresses valuations and steepens the front end of the curve. Either way, the transport mechanism is the same as in MR2 — and the asymmetry is worth remembering: uncertainty about inflation is worse for both legs of a balanced portfolio than uncertainty about growth, because a growth surprise moves equities and bonds in opposite directions while an inflation surprise moves them together. • Headline against expectations first — it is what is priced. • Core second — it is what the committee reacts to. • Contributions third — broad slowing versus one line. • Leads fourth — new lease rents, wages, expectations. • And remember the asymmetry: inflation surprises move both legs of a balanced portfolio the same way. The bond market’s own expectation is available continuously as break-even inflation, so you can see what a print did to expectations rather than only to prices. A print that moves break-evens is a regime statement; one that moves only the front end is a policy-path statement.

Why shelter is always late — and what the Fed actually watches

The single largest line in the index is also the slowest to move, and understanding why prevents a great deal of misreading. Shelter — the cost of housing services — is measured mostly by asking homeowners what their home would rent for, a figure called **owners’ equivalent rent**. The sample is surveyed on a rolling basis with a long cycle between updates, and the questions ask about current rent rather than about recent changes, so the index reflects what tenants agreed to some time ago. When market rents accelerated in 2021, the shelter component of the index did not fully reflect it until long afterwards; when market rents cooled, the official index kept rising. That lag is not a flaw in the release, it is the construction, and it means the shelter line in today’s print is describing the rental market of several quarters ago. A second divergence is institutional and more important than it looks. The Federal Reserve targets inflation in the **personal consumption expenditures** index, not the CPI. The two measure overlapping but differently weighted baskets — most notably, the PCE uses a formula that gives shelter a smaller weight and captures substitution between goods as relative prices change — so they can differ by several tenths of a point in the same quarter. A learner who checks the CPI print and then asks why the policy statement sounds calmer is usually looking at two different indices, and the one that drives the decision is the one most retail commentary quotes least. The response to both problems is the same: read the distribution, not just the average. **Median CPI** takes the middle of the price changes across the whole basket, and a **trimmed-mean** measure discards the largest and smallest changes before averaging, which removes the effect of a handful of categories having a violent month. Those measures are the reason a print that “looks like victory” can be a single volatile category doing all the work: if the headline fell while the median barely moved, the news is about one line item. If the median fell too, the disinflation is broad. That is the distinction this lesson’s counterfactual is built to make. • Shelter is measured from a slow survey of rents, so the biggest line lags the rental market by quarters. • The Fed targets PCE inflation, which weights shelter differently from the CPI that most commentary quotes. • Median and trimmed-mean measures strip out one-category months and show whether disinflation is broad. • A falling headline with a flat median is a story about one line in the basket. Do not read a base effect as a change in trend. A large fall in the twelve-month rate can come entirely from what dropped out of the comparison window last year, with the current monthly pace unchanged — which is why the monthly change, annualised, is the number to look at first.

PCE, trimmed means, and what the Fed actually targets

The CPI basket has been rebuilt from its components; the figure that policy is actually stated against is a different index, and the gap between them explains a good deal of the apparent disagreement between headlines and policy commentary. The two measures differ in four ways, and each shifts the level. **Coverage**: the personal-consumption measure includes expenditures paid for by third parties, most importantly employer-provided health care, which the CPI records only at the out-of-pocket share. **Weights**: the consumption measure updates weights with current spending patterns continuously and uses a chained formula, while the price index updates a fixed market basket periodically — which matters in a period when relative prices move sharply, because a fixed basket then overstates the cost of living. **Scope**: some items appear in one and not the other, including certain financial services and some categories of household spending. And **formula**, at the component level. The net effect is a persistent gap: the headline price index has generally run a few tenths of a percent above the consumption measure, and that gap is a known feature rather than a measurement failure. The practical consequence is that the two can tell different stories at the same moment. A price-index print and a consumption print at the same headline level imply different underlying inflation, and the reaction to the first is larger than the reaction to the second because it arrives earlier and is covered more widely. Markets trade the print they get; the policy statement refers to the other one. Both facts are true at once, and confusing them is a common source of forecasting error. On the question of what “underlying” inflation even means, the two competitors are worth distinguishing. Removing food and energy is a judgement about which categories are volatile across a business cycle; a **trimmed mean** instead discards the largest negative and positive contributors in each period, whatever they happen to be, which adapts to the episode rather than presupposing which categories are noisy. The second is more robust to a single item going haywire and harder to explain, and neither is the truth about a persistent component — both are smoothing devices with an assumption inside them. Expectations come in two families and neither is pure. Survey-based measures ask people and firms what they expect and are simple, slow to move and affected by what people are currently buying. Market-based measures are derived from index-linked bonds as **breakevens**, and they embed a risk premium and a liquidity effect alongside expectations, which means a breakeven can rise because inflation is genuinely expected to be higher or because the compensation demanded for holding the bond changed. When the two disagree, the survey is usually the slower and the breakeven the noisier, and the disagreement itself is worth naming rather than resolving. The usable routine follows: read a trimmed measure for the trend and the headline for the reaction, and when the two headline indices diverge, look at which components drive the difference before concluding that inflation is doing anything at all. • Coverage, weights, scope and formula separate the two main inflation measures. • A fixed basket overstates the cost of living when relative prices move sharply. • Markets react to the earlier print; policy is stated against the other measure. • Trimmed means adapt to the episode; survey and breakeven expectations disagree for a reason. A useful check when a headline looks alarming: decompose the gap between the two measures by category before treating it as news about inflation. Both can be reported accurately and still describe different things.

Sources

• CPI methodology, weights and the treatment of shelterBureau of Labor Statistics, “CPI Handbook of Methods”
• Core inflation as the policy signalFederal Reserve education materials and FOMC statements
• Shelter’s weight and lag in the indexBLS shelter methodology; standard rent-inflation literature