For our alternative estimates, we construct a data set of monthly price data from February 1977 to September 2026 and a distribution of 201 mutually exclusive PCE categories that are available across the sample period. We construct each of our alternative measures by modeling its performance against trend inflation. We define trend inflation as the 36-month centered moving average of headline PCE inflation (which includes food and energy prices); that is, for any point in time, we estimate trend inflation as the average of all monthly headline inflation rates from 18 months before through 18 months after that point in time.3 Since trend inflation requires 18 months of data before and after, we model the performance of each measure using inflation data from August 1978 to March 2025, but we are able to apply each model to the latest data and obtain estimates up to September 2026.
We compute our bias-adjusted estimate by using business-cycle indicators and percentiles of the price-change distribution to model the gap between the Dallas Fed’s trimmed mean PCE inflation measure and trend inflation, then we subtract the estimated gap from the Dallas Fed’s measure.4 This adjustment corrects for the divergence of the Dallas Fed’s measure from trend inflation without relying on forward-looking data, allowing us to apply the bias correction to the latest PCE release. This builds on the insight in Rich, Verbrugge, and Zaman (2022), who show that when the price-change distribution is positively skewed—that is, the data include a larger than usual share of sizable price increases—the Dallas Fed method yields an estimate of trimmed mean PCE inflation that falls below trend inflation.
For our extended measure, we construct a model to determine the optimal trim points for the September 2026 inflation estimate using data from August 1978 to March 2025. We compute what inflation would be when measured with every possible set of trim points and select the set that minimizes the divergence from trend inflation. The Dallas Fed method is similar, but it always uses inflation data from 1977 to 2009.5
To compute the 10-year measure for September 2026, we follow an approach similar to our extended-measure method, except that the model determining the optimal trim points uses data from January 2016 to March 2025.