Trimmed-Mean CPI

Trimmed-mean CPI is calculated by removing items with the most extreme price changes (highest and lowest 10–20%) each month, then averaging the remainder. This approach aims to capture underlying inflation trends better than core inflation by filtering out both temporary commodity shocks and unusual demand spikes.

How trimmed-mean works

Each month, statisticians:

1. Collect price changes for all ~80,000 items in the CPI basket.
2. Remove the top 10–20% of largest price increases.
3. Remove the bottom 10–20% of largest price decreases.
4. Average the remaining ~60–70% of items.

The items trimmed vary month-to-month. In an oil-shock month, gasoline (big price increase) is trimmed out. In a deflationary month, items with steep discounts are trimmed. This rotation means trimmed-mean is less distorted by any single shock.

Why trim?

Trimming serves multiple purposes:

1. Removes outliers. A sudden 50% surge in used-car prices (2021-22) is removed, preventing outsized headline impact.
2. Captures underlying demand. If half of items are raising prices 2–3% due to demand, trimmed-mean will show this even if gasoline is spiking 15%.
3. Smooths month-to-month noise. Reduces the statistical noise that makes month-to-month CPI hard to interpret.

Comparison with core inflation

Trimmed-mean is more sophisticated but less familiar to the public. Core inflation is simpler but sometimes includes items that should be excluded.

Trimmed-mean as a leading indicator

Research shows trimmed-mean CPI predicts core CPI inflation 1–2 months ahead:

• When trimmed-mean spikes, core CPI usually follows.
• When trimmed-mean moderates, core CPI usually moderates.

This makes trimmed-mean valuable for real-time assessment of whether inflation pressures are building or fading.

2021-22 example

June 2021:

• Headline inflation: 5.4%
• Core inflation: 4.5%
• Trimmed-mean: 3.8%
• Interpretation: Gasoline and food were spiking, but underlying demand was more moderate.

November 2021:

• Headline inflation: 6.8%
• Core inflation: 4.9%
• Trimmed-mean: 4.2%
• Interpretation: Commodity shock was masking a rising underlying trend.

June 2022:

• Headline inflation: 9.1%
• Core inflation: 5.9%
• Trimmed-mean: 5.0%
• Interpretation: Persistent underlying demand-driven inflation was becoming clear.

By fall 2022, core CPI had risen to 6.5%, confirming what trimmed-mean had signaled months earlier.

Fed’s adoption of trimmed-mean

The Federal Reserve has increasingly cited trimmed-mean CPI in statements:

• 2021-22: Emphasized trimmed-mean as evidence of persistent inflation.
• 2023-24: Tracked trimmed-mean deceleration to assess progress toward 2% target.

This reflects the Fed’s recognition that both headline and core measures have limitations.

Variants of trimming

The Cleveland Federal Reserve calculates trimmed-mean removing the top and bottom 16.4% of items (roughly one standard deviation). Other institutions use different percentages (10%, 20%). The choice matters:

• Aggressive trimming (20%): Smooths more; very stable but lags by several months.
• Light trimming (10%): More responsive but retains some noise.

Most analysts use the Cleveland Fed’s 16.4% version for consistency.

Relationship to sticky-price CPI

Trimmed-mean and sticky-price CPI are conceptually different:

• Trimmed-mean: Adaptive trimming based on price movements; fluid set of items.
• Sticky-price: Fixed set of items with infrequent price changes; structural approach.

They often move together but can diverge. In 2023, sticky-price CPI remained elevated while trimmed-mean fell faster, suggesting sticky-price items (rent, healthcare) were where inflation persistence lay.

Limitations

Trimmed-mean is less intuitive than core inflation:

• What is being trimmed varies month-to-month, making narrative storytelling harder.
• The methodology is complex; few people outside academia understand it.
• It is not an official Fed target (though watched closely).

For these reasons, core inflation remains the primary policy metric despite trimmed-mean’s statistical merits.

See also