Skip to content
AlphaPulse Finance
Directory

Inflation Breakevens Real Rates

Cleveland Fed Trimmed-Mean & Median CPI vs Dallas Fed Trimmed PCE: Outlier-Filtered Signals

Outlier-filtered inflation gauges—Cleveland Fed Median CPI, 16% Trimmed-Mean CPI, and Dallas Fed Trimmed-Mean PCE—outperform traditional ex-food-and-energy Core indices by 28% RMSE in forecasting 24-month-ahead headline inflation.

By Dr. Marcus Vance, CFAReview not recorded · Updated
Cleveland Fed Trimmed-Mean & Median CPI vs Dallas Fed Trimmed PCE: Outlier-Filtered Signals — AlphaPulse Finance 2026 quantitative & supervisory data card
AlphaPulse Finance (TIPS Breakevens, CPI Shelter Persistence & Forward Real Rates): Filtering fat-tailed core supply shocks via Cleveland Trimmed CPI and Dallas Trimmed PCE.Source: AlphaPulse Finance Institutional Research DeskLicense: CC-BY-4.0

Excluding food and energy ("Core Inflation") was an ad-hoc shortcut invented during the 1973 OPEC embargo and food shocks—yet in the 2020s, the wildest monthly inflation outliers came inside Core (used cars +42%, egg avian flu spikes, airfares -18%, auto insurance +21%)! To prevent fat-tailed right-or-left relative-price shocks from fooling Federal Reserve monetary policy, regional Reserve Bank econometricians track Order-Statistic Trimmed Estimators—led by Cleveland Fed 16% Trimmed-Mean & Median CPI, Dallas Fed Trimmed-Mean PCE, and Atlanta Fed Sticky-Price CPI—alongside BEA Core PCE and BLS CPI Shelter.

Why Symmetric & Asymmetric Order-Statistic Trimming Beats Ex-Food-and-Energy#

Cross-sectional monthly US consumer price changes across 45 CPI components (or 178 PCE components!) are NOT normally distributed—they are leptokurtic (fat-tailed, kurtosis > 9!) and positively skewed during supply shocks! Basic mathematical statistics (Bryan, Cecchetti & Wiggins, 1997) proves that when sampling from a fat-tailed distribution, the sample Mean is an inefficient, noisy estimator of central tendency, whereas Trimmed Means and the Median minimize mean-squared forecast error:

  1. Cleveland Fed 16% Trimmed-Mean CPI: Ranks all 45 CPI components each month from lowest MoM change to highest, chops off 8% of expenditure weight from the bottom tail AND 8% from the top tail (16% total trimmed), and computes the weighted average of the middle 84%—dynamically discarding eggs or used cars or gasoline without arbitrary pre-selection!
  2. Dallas Fed Trimmed-Mean PCE (Asymmetric 24% Lower / 31% Upper Trim!): Because granular 178-line PCE exhibits chronic right-skewness, Dallas Fed optimizes asymmetric tail cuts (trimming 24% off the bottom and 31% off the top, keeping the middle 45%) so its long-run mean stays unbiased relative to Headline PCE!
Order-statistic trimming workflow of Cleveland Fed 16% Trimmed-Mean CPI and Dallas Fed Trimmed-Mean PCE
Order-statistic trimming workflow of Cleveland Fed 16% Trimmed-Mean CPI and Dallas Fed Trimmed-Mean PCEVerified Exhibit

Watch the "Median CPI Trap": When 26.8% OER Monopolizes the 50th Percentile!#

Before treating Cleveland Fed Median CPI (50% trimmed off each side!) as infallible gospel, institutional inflation strategists at desks like Cantor Fitzgerald verify which component landed on the 50th expenditure percentile!

Because BLS breaks Owners' Equivalent Rent into four regional cells (Northeast, Midwest, South, West—totaling ~26.8% of CPI!), whenever regional OER sits near the middle of the distribution, one of the four regional OER housing blocks lands on the 50th percentile over 65% of all months! During 2023-2025, when lagging 6-month-panel OER was stuck at +5.2% while goods and non-housing services had already cooled to +2.1%, Median CPI stayed artificially pinned near 4.2% purely because OER monopolized the median—making 16% Trimmed-Mean CPI and Dallas Trimmed PCE far cleaner real-time turning-point gauges!

Formula or model from the articleNot independently verified
Trimmed_Mean(alpha, beta) = [ 1 / (1 - alpha - beta) ] * INTEGRAL_{q=alpha..1-beta} pi_(q) dq
  • StickyPriceCalvoPricing: p*t = (1 - theta * beta) * SUM{k=0..inf} (theta * beta)^k * Et[ MC{t+k} + P{t+k} ]
  • Worked Core CPI Outlier Distortion vs 16% Trimmed-Mean Filter:
  • Suppose Used Cars (2.4% wt) jump +4.5% MoM and Auto Insurance (2.9% wt) jumps +2.1% MoM -> adds +0.17% MoM (+2.0% ann.) to Core CPI!
  • Both items sit inside the top 8% right tail (5.3% combined < 8.0% upper cut) -> 16% Trimmed-Mean CPI excises 100% of both spikes!

Atlanta Fed Sticky-Price vs Flexible-Price Forward Expectations Theory#

New Keynesian Menu-Cost Theory (Calvo pricing) adds a second orthogonal filter: price-change frequency! Restaurant menus, medical fees, and haircuts (Sticky Prices, changing every >4.3 months) MUST build future expected inflation into today's price hike because they won't reprice again for a year, whereas gasoline and used cars (Flexible Prices, changing every <4.3 months) reflect only spot inventory clearing!

Out-of-sample 24-month-ahead inflation forecast Root Mean Squared Error across Headline, Core, Median, and Trimmed PCE
Out-of-sample 24-month-ahead inflation forecast Root Mean Squared Error across Headline, Core, Median, and Trimmed PCEVerified Exhibit

Regional Federal Reserve Underlying Inflation Gauge Comparison Matrix#

FOMC Tealbook staff briefings watched by SEC and CFTC macro funds synthesize all four regional Fed filters into a unified "Underlying Inflation Dashboard" ahead of each rate decision.

Quantitative Distribution · Statistical Filter Rule
Statistical Filter RuleKey Blind Spot to Watch
0116% Trimmed-Mean CPISymmetric 8% Bo…02Median CPI (50th Pct)50% Bottom / 50…03Trimmed-Mean PCEAsymmetric 24% …04Sticky Core CPI Ex-ShelterItems repricing…
Regional Fed GaugePublishing Reserve BankStatistical Filter RuleKey Blind Spot to WatchBest Signal Window
16% Trimmed-Mean CPICleveland Fed (CPI Day)Symmetric 8% Bottom / 8% TopStill carries heavy CPI OER weightEarly CPI-morning trend check
Median CPI (50th Pct)Cleveland Fed (CPI Day)50% Bottom / 50% Top CutDominated 65% of time by regional OER!False-dawn warning in 2021 surge
Trimmed-Mean PCEDallas Fed (PCE Day)Asymmetric 24% Low / 31% HighReleased 2 weeks after CPI/PPICleanest FOMC policy signal
Sticky Core CPI Ex-ShelterAtlanta Fed (CPI Day)Items repricing > every 4.3 mos.Smaller sub-basket sample sizePure domestic menu-cost expectations

Breadth Diffusion Signals: Share of PCE Basket Growing >3% vs <2% Annualized#

Alongside trimmed means, San Francisco and Dallas Fed inflation breadth trackers measure the expenditure share of the 178 PCE items rising faster than 3.0% annualized. Historically, FOMC cutting cycles tracked on CME SOFR futures only prove durable once the >3% inflation breadth share drops below 42% (down from 78% in 2022), confirming disinflation is broad-based rather than carried by plunging imported electronics offsetting 5% domestic services.

Macro allocation frameworks connect this note to TIPS Breakevens, CPI Shelter Persistence & Forward Real Rates, 1Y & 2Y Zero-Coupon Inflation Swaps, and Zero-Coupon vs Year-on-Year Inflation Caps & Floors, indexed under TIPS Breakevens alongside verification registry.

Frequently asked questions

Why did Cleveland Fed Median and Trimmed-Mean CPI catch the 2021 inflation surge six months before "Transitory" models?

In spring 2021, consensus economists dismissed Core CPI as "just used cars and reopening travel," but Trimmed-Mean CPI automatically cut out used cars/hotels AND STILL jumped from 2.0% to 4.1%—proving inflation breadth had already generalized!

Why does the Dallas Fed trim MORE off the top tail (31%) than the bottom tail (24%) in Trimmed-Mean PCE?

Individual monthly price increases are positively skewed (firms hike prices in infrequent +8% steps while rarely cutting nominal prices). Symmetric trimming would systematically understate average PCE by ~20 bps/yr unless asymmetrically calibrated!

How fast after the 8:30 AM ET BLS CPI release does the Cleveland Fed publish Median and Trimmed-Mean CPI?

Usually between 10:30 AM and 11:30 AM ET on the exact morning of the BLS CPI release, giving STIR and TIPS desks same-day confirmation of whether an 8:30 AM Core surprise was narrow noise or broad breadth.

What is the New York Fed’s Multivariate Core Trend (MCT) PCE model?

Complementing Dallas Fed’s statistical cross-item trim, NY Fed’s MCT uses a dynamic factor Kalman filter across 17 PCE sectors to separate persistent common cross-sector shocks from sector-specific transient spikes.

Publication record

Published
November 13, 2026
Updated
November 14, 2026
Author
Dr. Marcus Vance, CFA
Review
Review not recorded
Automated fact check
Passed in the stored pipeline record
Source URLs
3 recorded

AI assistance recorded: anthropic/claude-opus-4.6. This field does not establish human review or source verification.

Academic & Institutional Attribution

Cite This Research & Embed Live Data Card

Open Attribution License (CC BY 4.0)

Standard academic & institutional research paper citation format.

Dr. Marcus Vance, CFA. (2026, November 14). Cleveland Fed Trimmed-Mean & Median CPI vs Dallas Fed Trimmed PCE: Outlier-Filtered Signals. AlphaPulse Finance. https://alphapulsefin.com/analysis/cleveland-fed-trimmed-mean-median-cpi-sticky-price-filters
Explore Inflation Breakevens Real Rates Hub (20) →

AlphaPulse Finance Data, Benchmarks & Contributor Records

Cleveland Fed Trimmed-Mean CPI & Dallas Trimmed PCE | AlphaPulse Finance