Trimmed-Mean Inflation
A statistical antidote to noisy CPI data — trimming outliers from both tails to reveal the central tendency of price movements.
What trimmed-mean inflation is
Trimmed-mean inflation is a statistical measure of underlying price trends that works by removing a fixed percentage of the most extreme price changes from both ends of the distribution in any given month — and then computing the weighted average of what remains. The "trim" excludes both the biggest price increases and the biggest price decreases in that month, leaving the central distribution of price changes that better represents the underlying inflation trend.
The concept was developed by economists at the Federal Reserve Banks of Cleveland and Dallas in the 1990s as an alternative to the fixed-exclusion approach of core inflation (which always removes food and energy). The key insight: there is no single category that is "always" volatile — energy is very volatile on average, but so are hotel rates, used car prices, and airfares in specific months. A method that adaptively removes whichever items are most extreme in a given month is potentially more reliable than always removing the same categories regardless of what is happening.
The two most-used US trimmed-mean measures are: the Cleveland Fed 16% Trimmed-Mean CPI (removes the 8% of the basket with the highest price changes and 8% with the lowest, by weight, each month) and the Dallas Fed Trimmed-Mean PCE (removes 31% from the bottom and 24% from the top of the PCE component distribution). The asymmetric trim in the Dallas measure reflects the empirical finding that PCE has more extreme downside outliers than upside outliers in most months.
Trimmed mean vs core: key differences
The fundamental difference between trimmed-mean and core inflation is adaptive vs fixed exclusions:
Core inflation (fixed exclusion): Always removes the same categories (food and energy), regardless of whether those categories are particularly volatile in a given month. In a month where energy prices are flat, core still excludes energy — even though it is not distorting the headline. The advantage: transparency and simplicity. Everyone knows what core measures and can replicate it easily. The disadvantage: some months the core exclusions are irrelevant, and the volatile items are elsewhere in the basket.
Trimmed-mean inflation (adaptive exclusion): Removes whichever items are most extreme this month. If energy is quiet but used car prices and airfares are both spiking, trimmed mean will remove those — even though core would keep them. The advantage: more responsive to actual outlier patterns; more statistically consistent across different inflation environments. The disadvantage: less intuitive (the excluded items change every month) and harder to communicate to the public.
In practice, trimmed-mean and core measures tell broadly similar stories over medium-term horizons, but can diverge meaningfully in months when unusual price changes hit non-food, non-energy categories. The COVID-era used car price spike of 2021 (+30% year-on-year) was not excluded by core (used cars are not food or energy) but would have been trimmed by trimmed-mean measures in the months of peak volatility.
For a comparison with another alternative measure, see supercore inflation, which uses a conceptual rather than statistical approach to removing noise.
How central banks use trimmed-mean measures
Trimmed-mean inflation measures are primarily used as cross-checks rather than primary policy targets. No major central bank officially targets a trimmed-mean measure — the Fed targets PCE, the Bank of England targets CPI, the ECB targets HICP. But trimmed-mean measures are regularly cited in monetary policy communications as evidence about underlying inflation trends.
The Bank of Canada is the most prominent central bank that formally reports trimmed-mean CPI as one of three "preferred" measures of core inflation (alongside CPI-common and CPI-median). The Bank explicitly uses all three in its inflation assessment framework, giving trimmed mean a more official status in Canada than elsewhere.
In the US, the Dallas Fed Trimmed-Mean PCE is widely cited in academic and research contexts. Fed economists use it in research papers and sometimes in policy discussions, but it is a background indicator rather than a headline measure. The Cleveland Fed's Median CPI and 16% Trimmed-Mean CPI are similarly used as supplementary diagnostics.
The practical value of trimmed-mean measures is greatest in periods of widespread supply disruptions (when many categories are simultaneously distorted) or when unusual sectoral shocks are distorting headline data. During the 2021–22 inflation surge, trimmed-mean measures did a somewhat better job than core of tracking the rapid broadening of price pressures, as the inflationary impulse spread from the originally food/energy categories into a wide range of goods and services. See the inflation persistence entry for more on how breadth of price increases is measured.
Frequently asked questions
Is trimmed-mean inflation more accurate than core inflation?
Neither is definitively more "accurate" — they measure different things and are useful in different contexts. Research comparing trimmed-mean and core measures over long historical samples generally finds that trimmed-mean measures have slightly lower short-run volatility (they are better at filtering noise month-to-month) and somewhat better predictive power for future inflation in some time periods. But core inflation's advantage is transparency and communicability: a central bank can explain to the public exactly what core includes and excludes. Trimmed mean requires a statistical explanation of the trimming methodology, which is harder to communicate. Both measures have good track records as underlying inflation signals over horizons of 6–12 months and beyond.
What is the difference between trimmed-mean and median CPI?
Median CPI (produced by the Cleveland Fed) takes only the price change of the single component at the exact midpoint of the weighted distribution — the item whose cumulative basket weight is exactly 50%. Trimmed-mean CPI takes the average of the middle portion of the distribution (after removing the tails). Median is more extreme in its noise-reduction: it ignores everything except one item each month. Trimmed mean retains more information by averaging the central group. Median CPI tends to be slightly more stable but can miss genuine broad-based inflation if the median item happens to be in a stable category. Most economists prefer trimmed mean over median for this reason, though the Bank of Canada uses both.
Why does the Dallas Fed use an asymmetric trim (31% from the bottom, 24% from the top)?
The asymmetric trimming parameters were derived empirically from historical PCE component data. When the Dallas Fed researchers analysed the distribution of monthly PCE price changes over time, they found that the lower tail (the most negative price changes each month) was systematically fatter than the upper tail — more extreme downside outliers than upside outliers in the average month. A symmetric trim would therefore remove too little from the noisy lower tail and too much from the upper tail. The asymmetric design trims more from the bottom to achieve better noise-reduction symmetry in practice. The parameters are periodically re-evaluated as more data accumulates. This kind of empirical calibration is why different trimmed-mean measures (Cleveland vs Dallas) use different trimming percentages — each is optimised for its own data series.