Median CPI

Median [inflation] is the price change at the exact middle of the distribution of all component price increases—the rate at which half the things you buy got cheaper and half got more expensive. Unlike headline [inflation], which sums all price changes, or [core-inflation], which excludes food and energy, the median approach filters noise by construction, leaving only the central trend. It is one of several [inflation] gauges the [Federal Reserve] and international central banks monitor to distinguish transitory shocks from persistent wage-driven pressure.

How the median differs from headline and core

The headline [consumer-price-index] is the arithmetic mean of all price changes, weighted by household spending. If energy prices spike 40% one month while haircuts rise 1%, the average reflects both shocks. This makes headline [inflation] volatile and backward-looking—useful for consumer budgeting but noisy for policy.

Core [inflation] strips out food and energy, assuming those are temporary. Yet rent, medical care, or new cars can be equally transitory. The median approach inverts the logic: instead of deciding which items to exclude, it ranks all 300-plus components by price change and picks the middle one. If 150 components rise and 150 fall, the median is the inflation rate of the 151st cheapest component. By definition, it captures the central tendency and is resistant to outlier shocks.

The median CPI has historically been more stable and predictive of future [inflation] than either headline or core measures alone. During supply-shock episodes—oil crises, semiconductor shortages—the median often signals whether underlying demand inflation is present or whether the shock is genuinely transitory.

Trimmed inflation and percentiles

Median [inflation] is a special case of trimmed inflation—a family of measures that exclude the top and bottom percentiles of price changes. The Federal Reserve computes a “16% trimmed mean” [inflation] rate, which drops the largest and smallest 8% of price changes and averages the rest. Other central banks use 10%, 15%, or 20% trimming thresholds.

The trimming idea is straightforward: very large price increases (luxury goods, housing) and very large decreases (some services, import-driven goods) are often supply-driven or temporary. The middle bulk of the distribution reflects genuine demand pressure and wage growth. Percentile-based approaches allow policymakers to trade off between robustness (more trimming = more stable, but ignores real outliers) and comprehensiveness (less trimming = captures all components, but noisier).

A median measure is, in effect, infinite trimming—only the middle 1% (the 50th percentile) survives. This makes it extremely robust but also somewhat disconnected from household experience, since few people buy exactly the median basket.

Why the Fed watches median inflation

After the 2008 financial crisis, when headline [inflation] crashed but wage growth stalled, the Federal Reserve began publishing trimmed-mean and median [inflation] measures as part of its favoured inflation gauges. The logic was clear: if underlying inflation—the kind driven by labour costs and demand—is low, aggressive rate cuts are justified even if headline figures are volatile.

In 2021–2022, this distinction proved crucial again. Headline [inflation] surged to 9%, but trimmed-mean and median inflation rose more slowly, suggesting supply constraints rather than runaway demand. The Fed initially maintained an accommodative stance, partly because these “core” measures (using trimming instead of energy exclusion) implied less pressure than headlines suggested. When median [inflation] eventually accelerated alongside wages in mid-2022, it signaled a shift to demand-driven inflation and justified the rapid rate-hiking cycle.

The median’s lag and smoothness mean it is a follower, not a leader. It reaches its peaks and troughs after headline [inflation], making it unreliable for near-term forecasting but useful for confirming whether a spike is structural or noise.

Computing the median from disaggregated data

The median is computed using the full disaggregated basket of items—roughly 350 categories in the US CPI. The statistical agency calculates the month-over-month price change for each category (or the annual rate), sorts them, and picks the middle one. Because the basket is large and diverse, the median is usually somewhere between headline and core, closer to core in stable times but potentially above it during sectoral shocks.

The granularity matters: if you aggregate items too coarsely (e.g., “transportation” instead of “new cars,” “gasoline,” “car insurance”), the median becomes less meaningful. Central banks that publish median figures use fine categories to ensure the 50th percentile represents a real, purchasable item or service.

Different countries weight their baskets differently—housing, transportation, and food shares vary widely. The median CPI in Australia, where housing costs dominate, may tell a different story about underlying inflation pressure than the median in Switzerland, with a small housing weight. This makes cross-country comparisons of median inflation tricky but also shows that the measure is sensitive to real differences in consumption patterns.

The debate over which measure to use

Economists and policymakers disagree on which filter—trimming, median, or core—best reveals “true” underlying inflation. Some argue that the median throws away information: if luxury goods and housing are surging, that’s real inflation affecting real people, and the median ignores it. Others counter that monetary policy should target the stable centre of the distribution, not chase every shock.

In practice, the best approach uses multiple gauges in dialogue. If headline, core, trimmed-mean, and median [inflation] all rising together, the signal is clear: demand-driven inflation is broad-based. If they diverge—headline soaring while median barely budges—the story is shock-driven, and policy might wait. The median’s virtue is its mechanical robustness; its vice is that it offers no guidance on which real items are getting expensive fastest.

See also