In a recent remark, Kevin Warsh indicated a preference for alternative measures of inflation, such as the trimmed mean, suggesting that they may have advantages over traditional core inflation measures that exclude food and energy. The natural question, of course, is what superior properties these alternative measures might have relative to standard indicators. As it happens, Riccardo Trezzi addressed this exact issue in a Fed note (“Comparing Two Measures of Core Inflation: PCE Excluding Food & Energy vs. the Trimmed Mean PCE Index”) published in 2019, which was discussed at the FOMC in 2018 (see here the FOMC memo) together with Matteo Luciani. In this note, we revisit the main findings of that research.
The short, summary conclusion is that neither measure—the exclusion-based core inflation index (excluding food and energy) nor the trimmed mean—can be considered superior a priori. Both have appealing properties, but neither clearly dominates the other. From a communication standpoint, however, the trimmed mean is considerably more complex to explain and interpret. For this reason, Riccardo and Matteo Luciani ultimately argued in their Fed note in favor of continuing to use the traditional core inflation measure, as it is simpler to communicate and does not exhibit inferior statistical properties relative to the trimmed mean.
We believe this remains the likely position of the Fed staff, even in light of Warsh’s remarks. That said, we will continue to monitor developments closely and assess how the new Chair’s views evolve once formally in office.
What Warsh said
“What I’m most interested in is: What’s the underlying inflation rate? Not: What’s the one-time change in prices because of a change in geopolitics or change in beef?” “The measures I prefer are looking at things that are called trimmed averages” “We take out all of the tail-risks, all of the one-off items, and we ask ourselves whether the generalized change in prices is having second-order effects on the economy.”
Review of “Comparing Two Measures of Core Inflation” (Luciani and Trezzi, 2019)
The Fed note (“Comparing Two Measures of Core Inflation: PCE Excluding Food & Energy vs. the Trimmed Mean PCE Index”) provides a concise comparison of two widely used measures of core inflation within the PCE framework: the exclusion-based measure (excluding food and energy, hereafter ex-FE) and the Dallas Fed’s trimmed-mean estimator. The objective is to assess their relative performance as indicators of underlying inflation, based on three standard criteria: reduced volatility relative to headline inflation, consistency with the long-run mean of headline inflation, and informational content regarding future inflation.
From a technical point of view, the analysis shows that both measures successfully attenuate the high-frequency volatility of headline inflation, reflecting the substantial cross-sectional dispersion of price changes across consumption categories. However, the two approaches differ fundamentally in construction. The ex-FE measure excludes a fixed subset of components, while the trimmed mean removes extreme price movements in each period based on the cross-sectional distribution. As a result, the trimmed-mean estimator is better equipped to address time-varying distributions of price changes and is more robust to idiosyncratic shocks affecting individual sectors. One possible criticism of the trimmed mean is that the trimming is generally calibrated ex-ante (that is, on past shocks).
Despite these differences, the note finds no clear dominance of one measure over the other. Over long samples, both exhibit similar average inflation rates and track the central tendency of price changes. Moreover, neither measure consistently outperforms the other in forecasting future inflation. The implication is that ex-FE and trimmed-mean inflation capture distinct but complementary aspects of underlying price dynamics and should therefore be considered jointly.
At the same time, from a central banking perspective, the use of trimmed-mean measures raises several concerns. First, the procedure of systematically discarding a non-trivial share of the cross-sectional price distribution—often in a state-dependent and asymmetric manner—lacks a clear economic foundation. While statistically appealing, this approach risks eliminating information that may be relevant for policy, particularly when large relative price changes reflect persistent structural adjustments rather than transitory noise. In such cases, trimming may mechanically suppress signals of emerging inflationary pressures, thereby biasing real-time assessment. Second, trimmed-mean measures pose significant challenges in terms of communication, transparency, and policy accountability. Unlike exclusion-based measures, which rely on easily interpretable categories, trimmed estimators depend on evolving statistical rules that are less accessible to the public and more difficult to justify in policy deliberations. This opacity may complicate the transmission of monetary policy and weaken the credibility of the central bank, especially if different trimming specifications yield materially different signals. Moreover, the reliance on distributional filtering introduces an element of discretion that may be perceived as ex post adjustment of the data, raising concerns about robustness and consistency over time.
Riccardo’s view
My sense is that my 2019 note is still the prevailing Fed staff view. My expectation is that the Fed staff will try to push back Warsh’s approach. We will see who will win the fight. Interesting times ahead.