Two Different Economies
As of February 2026, the United States had two official inflation rates, and they disagreed by 0.7 percentage points. The Fed's preferred measure, core PCE inflation, showed prices rising at 3.0 percent annually. The Dallas Fed's Trimmed Mean PCE inflation measure showed 2.3 percent. This isn't a rounding error or margin of uncertainty. It's two fundamentally different descriptions of economic reality, constructed from the same underlying price data by choosing which prices to count and which to ignore.
Kevin Warsh, President Donald Trump's nominee for Federal Reserve chair, told lawmakers at his Senate hearing that he wants the central bank to change its strategy for measuring inflation. Specifically, Warsh stated he prefers "trimmed averages" measures of inflation, the lower number. The timing is notable: core PCE and Trimmed Mean PCE moved within a close range from 2023 through September 2025, but diverged after that, just as a new Fed chair needed better numbers to work with.
At stake is more than statistical methodology. The measurement the Fed uses determines whether interest rates rise or fall, whether mortgages become affordable or prohibitive, whether your employer keeps hiring. When Warsh called the current trend in inflation "quite favorable" during his Senate hearing, he was describing a reality that only exists if you accept his preferred metric. The question isn't which measure is correct. The question is how economic truth gets constructed in the first place.
The Architecture of Trimming
The Fed has long favored the core price index for personal consumption expenditures, which excludes volatile food and energy prices. The logic seems straightforward: strip out the noise, focus on the signal. But Warsh wants something different. "What I'm most interested in is: What's the underlying inflation rate?" he said. "Not: What's the one-time change in prices because of a change in geopolitics or change in beef?"
Trimmed Mean PCE inflation takes a different approach to finding that underlying rate. The Dallas Fed's measure uses 177 PCE components and excludes price changes below the 24th percentile and above the 69th percentile. In other words, it trims 24% of the bottom and 31% of the top by PCE expenditure weight. The prices that remain, the middle 45%, become the official measure of inflation.
The method sounds elegantly objective: survey everything, then mathematically isolate the center. Warsh has promised a broader "regime change" for the central bank's approach to inflation measurement, stating he wants to "survey a billion prices." But here's the paradox: trimmed mean inflation surveys a billion prices and then deliberately ignores 55% of them. The question becomes: which 55%?
Disco-Era Patterns, TikTok-Era Economy
The asymmetric trimming thresholds for Trimmed Mean PCE were selected to best match three different measures of medium-term inflation trend from 1977 to 2009, according to the Dallas Fed. Those specific percentiles, 24th and 69th, weren't chosen because of some natural law of economics. They were reverse-engineered to fit historical patterns from an economy that predates the internet, smartphones, and the entire modern service sector.
The asymmetry exists because price-change distributions are, on average, "negatively skewed" during that 1977 to 2009 period. Negative skew means the average of the bottom 10 percent of price changes is much further below the median than the average of the top 10 percent is above the median. So the Dallas Fed trims more from the top than the bottom to compensate. But that skew pattern is itself a historical artifact, a description of how prices behaved during a specific economic era.
We're using the statistical fingerprint of stagflation, the Volcker shock, the dot-com bubble, and the housing crisis to calibrate our measurement of an economy shaped by pandemic supply chains, algorithmic pricing, and geopolitical fragmentation. The measure isn't discovering underlying inflation. It's defining inflation as "whatever looks like the trimmed middle of price changes looked between 1977 and 2009."
The Predictive Power Argument
Proponents of trimmed mean inflation point to its forecasting performance. Headline PCE inflation typically converges toward trimmed mean rather than core PCE over 12-month horizons. The coefficient on the trimmed mean gap is near one after 12 months, indicating strong predictive power, while the coefficient on the core PCE gap is near zero and statistically insignificant across horizons. By its own historical standards, trimmed mean does a better job predicting where headline inflation will land.
Trimmed Mean PCE inflation also has a tighter relationship with labor market slack than core PCE, according to research from the Dallas Fed. These technical advantages matter for central bankers trying to set policy. But they reveal something deeper about how economic measurement works: a metric can be predictively superior within the system that generated it while still being fundamentally constructed rather than discovered.
Bank of America found a 12-month inflation gauge using the trimmed method would have a mean of 2.3% and median of 2.8% as of February 2026. Yet Bank of America economist Aditya Bhave warned that Warsh's proposed reconfiguration might not pan out as hoped. The skepticism isn't about the math. It's about whether a measurement optimized for one economic regime will work in another.
What Gets Left Out
When Warsh says he's not interested in "the one-time change in prices because of a change in geopolitics or change in beef," he's articulating a philosophy: some price changes are real inflation, others are just noise. But to the household buying beef, there's no such distinction. The price is the price. The grocery bill is the grocery bill.
Trimmed mean inflation excludes price changes above and below certain percentile thresholds, which means it systematically ignores the most dramatic price movements in either direction. If your rent spikes, if your childcare costs plummet, if the specific bundle of goods and services you actually purchase happens to fall in the excluded tails of the distribution, your economic reality doesn't count toward the official measure of economic reality.
This isn't a flaw in the methodology. It's the methodology. The entire point of trimming is to exclude outliers. But "outlier" is a statistical category, not an economic one. The prices being trimmed are still prices people pay. They're being excluded not because they're false or erroneous, but because they don't fit the pattern the measure is designed to capture.
Constructed Reality, Real Consequences
The divergence between core PCE and Trimmed Mean PCE after September 2025 creates a natural experiment in how measurement shapes policy. If the Fed adopts Warsh's preferred metric, inflation suddenly looks 0.7 percentage points lower. That's the difference between an inflation problem that requires continued vigilance and an inflation problem that's largely solved. It's the difference between keeping rates higher for longer and cutting rates sooner.
Lower measured inflation justifies looser monetary policy, which makes borrowing cheaper, which stimulates economic activity, which can itself generate inflation. The measurement doesn't just describe the economy. It changes the economy. And because the Fed's inflation target is 2 percent, whether you're at 2.3 percent or 3.0 percent determines whether you've succeeded or failed at your institutional mandate.
This is what makes the choice of inflation metric a choice about reality itself. Economic statistics don't exist in nature waiting to be discovered. They're constructed through layers of decisions: which prices to survey, how to weight them, which components to aggregate, which percentiles to trim, which historical period to optimize against. Each decision is defensible. Each decision shapes the outcome.
Warsh's preference for trimmed mean inflation isn't wrong. But it isn't simply right, either. It's a choice about what kind of inflation we want to see, which price changes we want to count as meaningful, which economic patterns we want to privilege. The incoming Fed chair isn't proposing better measurement. He's proposing different measurement, which will construct a different economic reality, which will justify different policy choices, which will create different actual outcomes for actual people.
The most honest answer to "what's the underlying inflation rate?" might be: there isn't one. There are only the prices people pay, and the various ways we choose to summarize them, and the economic realities those summaries bring into being.