The Effect of Component Disaggregation on Measures of the Median and Trimmed-Mean CPI

Abstract

For decades, the Federal Reserve Bank of Cleveland (FRBC) has produced median and trimmed-mean consumer price index (CPI) measures. These have proven useful in various contexts, such as forecasting and understanding post-COVID inflation dynamics. Revisions to the FRBC methodology have historically involved increasing the level of disaggregation in the CPI components, which has improved accuracy. Thus, it may seem logical that further disaggregation would continue to enhance its accuracy. However, we theoretically demonstrate that this may not necessarily be the case. We then explore the empirical impact of further disaggregation along two dimensions: shelter and non-shelter components. We find that significantly increasing the disaggregation in the shelter indexes, when combined with only a slight increase in non-shelter disaggregation, improves the ability of the median and trimmed-mean CPI to track the medium-term trend in CPI inflation and marginally increases predictive power over future movements in CPI inflation. Finally, we examine the practical implications of our preferred degree of disaggregation. Our preferred measure of the median CPI suggests that trend inflation was lower pre-pandemic, while both our preferred median and trimmed-mean measures suggest a faster acceleration in trend inflation in 2021. We also find that higher disaggregation marginally weakens the Phillips curve relationship between median CPI inflation and the unemployment gap, though it remains statistically significant.