Core Inflation Is 2.5 Percent or 3.3 Percent or Whatever Number You Trust Most

The Number

The Bureau of Economic Analysis released the July reading for the Personal Consumption Expenditures price index this morning. Core prices rose 0.2 percent versus June and 3.3 percent versus a year ago, the latter of which is identical to the figure in June. It held steady, and it held steady a full 1.3 points above target. Headline PCE rose by 0.2 percent on the month and 3.7 percent on the year, beating forecasts by a tenth of a point.

Core CPI, which was released two weeks ago, tells a different story. Headline CPI was 3.4 percent on the year, against 3.7 percent for headline PCE over the same month, and core CPI was 2.5 percent, down from 2.6 percent in June. The core number actually decreased by a tenth over the last month. So we have two measures of inflation, both supposedly measuring the same thirty one day period but coming to results that differ by 80 basis points, and one of them is right around the target while the other is well above it.

The Federal Reserve targets the one that stayed steady.

Two Measures, One Economy

The two measures of inflation are meant to capture the same concept: how many more dollars must be spent this month to buy the same goods and services that were bought last month. The reason both numbers disagree is that both are constructed differently.

The CPI, from the Bureau of Labor Statistics, is built from a survey of households and updated periodically, while the PCE, from the Bureau of Economic Analysis, is built from business spending and inherently includes more spending on behalf of consumers.

The latter point matters more than you might think. PCE captures spending on health care benefits, which is a significant portion of the economy and has been rising as a share of total compensation for decades. That spending is effectively invisible to the CPI, because the spending is made by an employer, not a household.

The other reason the two indexes behave differently is substitution. When the price of beef rises, consumers buy less beef and more chicken. The PCE index automatically incorporates these changes within the span of a month, while the CPI takes longer to reflect them, resulting in a phenomenon called substitution bias. The result is that, over long stretches of time, the CPI tends to increase faster than the PCE. That has been the case for much of the last twenty years, with core PCE consistently running two to three tenths of a point below core CPI.

Why Core CPI Is Falling, But Core PCE Is Not

That trend appears to have reversed in 2026. The reason has to do with the weight of shelter in the two indexes.

Shelter makes up approximately 33 percent of the core CPI basket versus 15 percent in the PCE, because the latter uses a different formula that essentially takes into account a much larger economy. When the rate of increase in the price of rent slowed, it had a much larger effect on the CPI than on the PCE. It has subsequently been the single largest factor depressing core CPI growth this year, preventing it from matching the increase in core PCE.

Meanwhile, core PCE has seen little slowdown across the rest of the economy.

Viewed in this light, all of them but core CPI are well above the 2 percent inflation target, and none of them are showing signs of improvement from last month.

Where Did the Spending Go?

The other part of the report details the change in spending, which is an interesting look at the economy in itself. Consumer spending rose 36.3 billion dollars, or 0.2 percent, in July. A closer examination shows that services spending increased by 86.2 billion dollars while goods spending decreased by 49.9 billion.

The headline number is positive because one large category gained while the other lost ground. After adjusting for rising prices, however, the picture becomes far less positive. With prices rising by 0.2 percent, as reflected in the headline PCE, real spending was flat in July. Real spending increased by 0.3 percent in June.

Americans continued to spend more dollars on the same amount of goods and services.

The Savings Rate Tells Us How Much Confidence People Have

Personal income increased by 115.1 billion dollars, or 0.4 percent, and disposable income, which is after taxes, increased by 0.5 percent. Some of that increase was due to wages, some due to direct payments from the government, and some due to tax cuts enacted at the state level.

Consider the saving rate, which measures the share of after tax income that is not spent on consumption. It increased to 3.0 percent in July from a downwardly revised 2.6 percent in June, the highest it has been since March.

Income increased, spending increased at a lower pace, and the difference went into savings.

Households that expect income to increase in the next twelve months will spend some of that increase. Households that expect income to decrease are likely to save some of their current income. A jump in the saving rate of 40 basis points in a single month while real spending is stagnant reflects confidence, or the lack thereof, and is a useful indicator of the outlook for personal consumption.

The saving rate of 3.0 percent is not high by historical standards. It sits well below the five to seven percent range typical for much of the 1990s. The point is not that households are saving an unprecedented amount, but that they have chosen to save more in the last few months.

Supercore Is What the Fed Watches Most Closely

Within the core PCE, there is a number that is watched even more closely than the number itself. This is supercore, a measure of services spending that excludes energy and housing.

Supercore is supposed to be a barometer of inflation that is less subject to supply shocks, because energy prices are notoriously volatile, and housing prices are subject to sticky adjustments. By removing those two factors, what remains is services inflation, which reflects domestic labor costs and is thus theoretically influenced by monetary policy.

Supercore increased by 0.3 percent in July and 3.9 percent on an annual basis.

That number represents a significant challenge to the view that the Federal Reserve has engineered a restrictive policy environment, because nearly 4 percent inflation in a category that is supposed to react directly to monetary policy changes has not resulted in the anticipated dampening of economic activity. Put another way, the economy has adapted to higher interest rates, and prices continue to increase within the economy.

Two subcategories within supercore deserve special attention, because they provide context for why inflation has proven to be so persistent. Child-care and elder-care expenses increased sharply in July, with both categories seeing widespread price pressures across the country. Both represent labor intensive industries within services, and both are seeing shortages at a time when demand is rising. The supply constraints cannot easily be remedied by adjusting interest rates, because there is no immediate response to demographics within labor markets.

The Same Supply Shock Is Showing Up in Memory Prices

The most surprising detail within the report concerns prices of consumer electronics rising due to a supply chain issue related to memory production.

It is a detail that merits special attention, because consumer electronics have been a consistent deflator for the broader economy for decades. Televisions, personal computers, and mobile phones have steadily grown more capable and cheaper to buy over that period of time, allowing households to increase consumption without increasing spending.

This process has reversed in 2026, and the reason has to do with artificial intelligence. The demand for memory within data centers has created a situation where manufacturers are prioritizing high bandwidth memory over traditional memory, reducing the supply of the latter and increasing its price.

This price pressure is passed on to consumers in the form of higher prices for personal computers, and by extension, higher prices within the PCE. It is the first time that a specific price increase within a federal economic report has been directly attributable to investment in artificial intelligence servers, and it is likely to be the first of many.

The Disinflation That Was Not Really Disinflation

It is worth exploring why the figures for July were not higher, because there were two specific factors that pushed prices down. One was energy, where lower prices reflected a reduction in risk premia associated with the potential closure of the Strait of Hormuz and a disruption to oil exports.

The other factor was groceries, where discounts have been supported by a combination of tariffs being removed for low and middle income households.

Those forces, however, are not really disinflationary in the way that central bankers might hope, because they represented temporary factors that will unwind at some point.

The removal of a risk premium in energy prices is unwinding as the Strait of Hormuz closes, and lower import costs for groceries are unwinding as tariffs are reinstated. In neither case is there a change in the cost of production or inflation within the economy as a whole.

A tariff-induced discount lowers the amount a consumer spends on a good without lowering the amount a producer is paid for the good. It is a transfer of income from one party to another and has no effect on the aggregate supply of the good. Similarly, changes in risk premia alter asset prices without really affecting the value of the underlying asset.

Inflation targeting works when inflation reflects changes in demand relative to supply, because monetary policy is a demand-side tool. Temporary supply shocks push prices higher or lower, but they do not alter the amount of spending within an economy.

That is the crucial distinction that separates a healthy inflation report from an unproductive one, and it is the reason why a large chunk of the discussion of inflation should concern temporary supply shocks rather than persistent ones. With that context, the figures for July reflect a period of persistent inflation in supercore and a pause in disinflation in core CPI, both of which contain persistent supply shocks.

What the Fed Sees Three Weeks From Now

The Federal Open Market Committee will be meeting on September 16, and when it does, this is the report that will be on its desks.

Core PCE stopped falling. Headline PCE is well above target. Supercore is running at 3.9 percent, real spending is stagnant, the saving rate is rising, and the interest rate is 3.5 to 3.75 percent, not far above 3.7 percent inflation.

It is the last figure that is likely to keep committee members up at night, because a central bank that sees itself as inflation fighters finds itself with a real monetary policy that is near zero in its ability to slow the economy.

Two Events This Week Are Worth Paying Attention To

Nvidia will be reporting earnings after the close today, and the most important number will be the guidance for the next quarter, because it reflects the market's view of how much spending on artificial intelligence infrastructure will accelerate.

That is followed by Kevin Warsh's Jackson Hole speech, which is his first as Federal Reserve Chair.

Both events were likely to dominate the week regardless, but the inflation report this morning has changed the context for each.

Nvidia has to justify building more servers if its earnings are supposedly responsible for higher prices for personal computers. Meanwhile, Warsh has to explain why the Fed's explicit commitment to fighting inflation does not include adjusting the interest rate when it is barely above inflation itself.

They have seventy two hours between them.

The Bottom Line

When two measures of inflation contradict each other by more than 80 basis points, the responsible reaction is not to embrace whichever number you prefer, but to understand why they differ.

In this case, it is explained largely by shelter and labor intensive services: the former receives a higher weight in the CPI and has been cooling, while the latter receives a lower weight in the CPI and has been accelerating.

The Federal Reserve prefers the PCE, which reflects a stable core rate and a higher supercore, a persistent increase in prices of services that reflect labor costs. It also reflects a rise in saving rather than spending following a rise in disposable income.

None of these facts indicate an emergency, but all of them are relevant to a conversation about whether the current monetary policy is sufficient three weeks from now.