Study I · Measuring the Economy

Which Inflation?

Nine published measures of inflation: in July 2026 they disagreed by 1.22 percentage points.

5 min read · 5 exhibits

Every month the Bureau of Labor Statistics publishes a number, and every month that number is reported as inflation. It is not inflation. It is one instrument’s reading of inflation, and federal agencies and the Federal Reserve publish at least nine others. In July 2026 the nine disagreed with each other by 1.22 percentage points.

This is not a scandal and nobody is hiding it. The measures differ because they were built to answer different questions, and the differences are documented in detail by the agencies themselves. But the practical consequence is real: the choice of instrument decides the story, and almost nobody reporting the number names the instrument.

Exhibit 1

Nine measures of the same month

Source. BLS Consumer Price Index release, August 12, 2026 (series CUUR0000SA0L1E, CUUR0000SA0, CWUR0000SA0, SUUR0000SA0, CUUR0000SASL2RS); BEA Personal Income and Outlays, August 26, 2026 (PCEPI, PCEPILFE); Federal Reserve Bank of Cleveland median and 16% trimmed-mean CPI, August 12, 2026.

BEA’s annual update of September 30, 2026 revised PCE to 3.36 and core PCE to 2.98.

Full size

Method. Twelve-month changes computed from published index levels rather than from rounded headline prints, so values carry two decimals. All nine describe July 2026. No seasonal-adjustment mixing: NSA where the agency publishes NSA for the twelve-month change. The five BLS rows are not seasonally adjusted; the Cleveland Fed median and trimmed mean and BEA’s two PCE indexes are published only seasonally adjusted. PCE and the Cleveland Fed measures are at the September 4, 2026 vintage (BEA’s August 26 release; the Cleveland Fed’s August 12 release).

The spread is not noise around a true value. Each end of it is a defensible answer to a different question. Core CPI at 2.48% answers what is the underlying trend, stripped of the volatile parts. Headline PCE at 3.70% answers what happened to the prices of everything consumed on households’ behalf, including what their insurers and the government paid. A reader told only that “inflation is 3.4%” has been handed one of nine answers with no indication that the others exist.

The sharpest demonstration came a month earlier.

Exhibit 2

June 2026: the same month, in opposite directions

Source. BLS CPI release for June 2026, published July 14, 2026; Federal Reserve Bank of Cleveland median and trimmed-mean CPI for June 2026.

Full size

Method. The trimmed-mean and median figures are calculated by the firm from the Cleveland Fed’s published monthly changes for June 2026 (trimmed mean 0.0106%, median 0.1741%) as (1+m)¹²−1. The CPI-U and core figures are annualized by the firm from BLS’s published seasonally adjusted index levels as (June/May)¹²−1: CPI-U (CUSR0000SA0) 333.979 in May to 332.568 in June 2026, core (CUSR0000SA0L1E) 336.121 to 336.065. BLS publishes these index levels, so the annualized values do not carry the rounding of the one-decimal monthly rates (−0.4% and 0.0%).

In June 2026 the headline index fell at an annual rate of 4.95 percent while the median price change in the same basket, computed from the same underlying microdata by the Cleveland Fed, rose at 2.11 percent. Gasoline fell 9.7 percent in the month (BLS, CPI Summary, Table A, seasonally adjusted) and dragged the weighted average down; the typical price in the basket did not move that way at all. Both numbers are correct. They point in opposite directions because a weighted mean and a median are different instruments, and one of them was being pulled by a single volatile component.

Where the largest gap comes from

The most consequential disagreement is between the two indexes that actually drive policy and contracts: CPI, which indexes Social Security and tax brackets, and PCE, which the Federal Reserve targets. Over 1959 to July 2026, CPI ran 0.43 percentage points a year above PCE. Both agencies have published decompositions of why.

Source. The seasonally adjusted CPI-U (BLS, CUSR0000SA0) rose from 29.01 in January 1959 to 332.813 in July 2026, a compound annual rate of 3.68% over 67.5 years; the PCE price index (BEA, PCEPI (August 26, 2026 vintage, via ALFRED)) rose from 15.164 to 131.659, a compound annual rate of 3.25%. The difference is 0.43 points a year.

Exhibit 3

Why CPI runs above PCE — two official decompositions

Source. McCully, Moyer and Stewart, “Comparing the Consumer Price Index and the Personal Consumption Expenditures Price Index,” BEA Survey of Current Business, November 2007; BLS, Focus on Prices and Spending, Vol. 2 No. 3, May 2011.

Full size

Method. Contributions to the CPI-over-PCE gap in percentage points. The two studies cover different periods and are not directly comparable; they are shown together because they are the two official decompositions on the record. Note a live disagreement: BEA’s own text says the formula effect “explains almost half” of the gap, while the arithmetic in both tables makes the weight effect larger. We present both as published rather than resolving it.

Exhibit 4

The same two categories, weighted three times differently

Source. Johnson, “A comparison of PCE and CPI: Methodological differences in U.S. inflation calculation and their implications,” BLS working paper, November 2017, Table 6. Weights for 2014–2016.

Full size

Method. Expenditure weights, percent of index. The gap is not an error in either index. CPI weights come from what households report spending; PCE weights come from business and administrative data and count spending made on behalf of households — employer-paid premiums, Medicare, Medicaid. That single difference is why medical care carries roughly three times the weight in the index the Federal Reserve targets.

And none of them is your number

A national index describes no actual household. BLS itself has measured the spread: between 2006 and 2023, households in the lowest income quintile experienced inflation 0.28 percentage points a year higher than the highest quintile, compounding to a 7.7-point gap over eighteen years. “In 2023, rent increases explained over 25 percent of the inflation gap” (Klick and Stockburger, Monthly Labor Review, July 2024); rent is 14.8 percent of spending at the bottom and 3.8 percent at the top (2021 weights).

For a business the divergence is larger still, and it runs the other way.

Exhibit 5

What a contractor pays, against the national index

Source. BLS Producer Price Index, series WPUIP2311001, inputs to residential construction, goods, released August 13, 2026 (index 323.198 in July 2025 to 344.861 in July 2026); BLS CPI release, August 12, 2026.

Full size

Method. Twelve-month changes computed from published index levels, all three not seasonally adjusted, the PPI at the September 4, 2026 vintage (BLS’s August 13 release). PPI and CPI measure different things — producer input costs against consumer prices — and are shown together not as rivals but to make one point: a firm buying materials in July 2026 faced cost growth 4.2 points above the core index that anchors the public conversation about whether inflation is beaten.

The honest sentence is not “inflation is 3.4 per cent.” It is: the all-items consumer index rose 3.4 percent over the year to July, the measure the Federal Reserve actually targets rose 3.7, the trimmed measures say 2.6, and the materials a builder buys rose 6.7. That sentence is nine words longer and it is the only one that survives contact with the data.

What follows

Three practical rules, and none of them requires taking a position on monetary policy.

Name the instrument before the number. Anyone who cites an inflation figure without saying which series it comes from is either not checking or counting on you not to.

Match the instrument to the decision. A contract escalator should be indexed to the costs the contract actually bears. Indexing a materials-heavy contract to headline CPI is a choice to absorb the gap in Exhibit 5, and in the year to July 2026 that gap was 3.3 points.

Treat a single month as a single month. June 2026 produced a falling headline and a rising median in the same release. Either could have been reported alone, truthfully, to opposite effect.

What we could not verify

There is no official “supercore” series at BLS or BEA. The 2.99% figure in Exhibit 1 is the firm’s calculation from CUUR0000SASL2RS (services less rent of shelter) and is labeled as such.
No BEA or BLS decomposition exists for July 2026, so we do not attribute a cause to the current inversion in which PCE runs above CPI — the reverse of the historical pattern. The mechanism is reasoning, not sourced attribution, and is therefore not charted.
BLS’s own thirty-year retrospective on the 1983 shift to rental equivalence gives no estimate of that change’s effect on measured inflation. We give no figure.
Lebow and Rudd’s CPI bias estimate appears as 0.6 percentage points a year (0.1–1.2) and as 0.9 (0.3–1.4). The 0.6 is their 2001 Federal Reserve working paper (FEDS 2001-61); the 0.9 is the 2003 published version (Journal of Economic Literature 41(1)). We have not averaged them.
A structural break. BLS collected no CPI data from October 1 to November 12, 2025 during the lapse in appropriations. The October 2025 observation does not exist. Any series crossing that month is discontinuous, and we have not plotted across it.

Sources

Corrections

The July 2026 spread is 1.22 percentage points, not 1.23, with every BLS row in Exhibit 1 on one not-seasonally-adjusted basis (core CPI 2.48, CPI-U 3.36, trimmed mean 2.60).
Exhibit 5 is on the same basis: CPI-U 3.36 and core CPI 2.48.
In Exhibit 2, June CPI-U and core CPI are annualized from index levels (−4.95 and −0.20), not from the rounded monthly rates.
The 14.8 and 3.8 percent rent shares are 2021 weights, and “the driver is rent” is replaced by BLS’s own finding.

Data for every exhibit: download (CSV)

Cite this study

Higgins Strategy. “Which Inflation?” Study I, Measuring the Economy. October 12, 2026. https://higginsstrategy.com/studies/which-inflation/

@techreport{higginsstrategy2026whichinflation,
  author      = {{Higgins Strategy}},
  title       = {{Which Inflation?}},
  institution = {Higgins Strategy},
  type        = {Study},
  number      = {I},
  series      = {Measuring the Economy},
  year        = {2026},
  url         = {https://higginsstrategy.com/studies/which-inflation/},
  month       = oct
}

Next: Study II, Two Reports, One Economy

Which Inflation?

Where the largest gap comes from

And none of them is your number

What follows

What we could not verify

Sources

Read next