An honest defence of the core measure, followed by the reason the standard critique of it does not hold this month: the alarming annual energy figure is a base effect from a spike that is already unwinding. Plus the gap that does matter — headline is what a household pays, core is what the Fed acts on, and right now they are 0.9 points apart.
The July CPI report, released 12 August, was read as uneventful. Headline prices rose 0.1% on the month and 3.4% over the year. Core CPI, which strips food and energy, rose 0.2% and 2.5% annually — reported as the slowest annual core pace since March 2021. Both came in roughly where forecasters expected, and coverage moved on inside a day.
Inside the same release are two numbers that point in opposite directions and are almost never printed together. Core inflation is at a four-and-a-half-year low. Energy prices are 14.7% above where they were a year ago. Core is low partly because it excludes energy.
That sounds like an indictment of the core measure. It is not, quite — and the more useful version of this story is what the monthly data does to the conclusion you would otherwise reach.
Core is not a trick and it is not a way of hiding bad news. Food and energy prices are volatile and substantially mean-reverting: they respond to weather, harvests, OPEC decisions, refinery outages and shipping disruptions, almost none of which says anything about the persistent, demand-driven component of inflation that monetary policy can actually influence. Stripping them produces a cleaner read on the trend. Central banks use some version of it for exactly that reason, and the practice has a decent empirical record at predicting where headline inflation ends up.
The real critique of core is therefore not that excluding volatile components is wrong. It is that the exclusion is only valid while the excluded component is behaving volatilely rather than trending. A supply-driven energy shock that persists is not the noise core was designed to filter out. If it lasts, it enters core anyway — through freight, packaging, plastics, utilities and every service with an energy input — on a lag of several months. When that is happening, core understates the problem for two or three quarters and then catches up abruptly.
This is where the standard version of the argument meets the actual report, and it is worth being precise rather than dramatic.
Energy prices fell 1.5% in July, after falling 5.7% in June. The 14.7% annual figure is a base effect. It measures July 2026 against a much lower July 2025 and is dominated by a spike that has already happened and is now unwinding.
That materially changes the inference. A sustained supply shock feeding into core over the next two quarters, and a spike that peaked and is rolling over, produce the same year-over-year number today and opposite implications for next year. The monthly direction says the second is closer to what is happening. If energy keeps falling month over month, the annual comparison collapses mechanically as the 2025 base rises — and the pass-through pressure on core weakens rather than builds.
So the honest conclusion is not that core is hiding an inflation problem. It is narrower and more interesting: the annual energy figure describes the past, the monthly figure describes the present, and a reader who sees only the first will draw the wrong conclusion about the second. The 14.7% is real, and it is also stale.
This is the general failure mode of year-over-year statistics. They are the default in inflation coverage because they smooth seasonality, and the price of that smoothing is a twelve-month memory. In any month where the trend has recently turned, the annual number is describing a world that no longer exists.
The reader-useful part of the divergence survives all of that intact. Nobody buys a basket with food and energy removed. Household experienced inflation is headline — 3.4% — not core — 2.5%.
The 0.9-point gap between them is not a rounding artefact. It is the portion of your cost of living that the number the Fed acts on deliberately does not contain. That is a defensible methodological choice by the statistical agency and an unhelpful one at the till, and both things can be true.
The practical framing: use core to anticipate policy, use headline to budget. When the two diverge, the size of the divergence is telling you how much of your own price experience is currently classified as transitory.
Shelter is the other half of the picture and gets less attention than it deserves. It rose only 0.1% in July, but accounted for roughly two-thirds of the entire headline monthly increase — a reminder that the largest single component of the index moves slowly and dominates the total anyway. A tenth of a point on shelter matters more to the headline than a full point on almost anything else.
The FOMC meets 15–16 September, and a report with cooling core and falling monthly energy gives it room to sit still; post-release, CME FedWatch put September hike odds near 42%. That is context for the print, not the point of it.
Two numbers in one report, pointing opposite ways, and the reconciliation is a single line of methodology plus a base effect. That is not a scandal about how inflation is measured. It is a reminder that a summary statistic is a compression, and that the thing compressed out is usually the thing you were trying to find.
Headlines quote the annual rate of change, not the price level. Prices rising 3.4% more slowly than before are still rising, and they are rising on top of everything that came before. The core measure many stories lead with also excludes food and energy — two categories where households notice changes most directly — which widens the gap between the reported number and the experienced one.
DrafterDaily reads economic releases for what they actually measure. One analytical brief a day, no hype.
Browse Finance coverageGrad PLUS lent up to the full cost of attendance. Its replacement caps professional students at $50,000 a year — against a medical-school year that averages $61,110. The difference does not disappear.
Fitch tracks sixteen perpetual non-traded BDCs. In Q2 2026, redemption requests averaged 10.3% of shares against a standard 5% cap, and ten of the sixteen breached it. Nothing broke — which is the uncomfortable part.
Nationally, asking prices are down about 2% and inventory is up 3.2% — a picture of orderly rebalancing. Underneath it, Austin's median list price is down 12.2% while San Francisco's inventory has fallen nearly 20%. Those are not variations around a mean.