Working Paper: CEPR ID: DP17485
Authors: Sergio Ocampo; Raphael Schoenle; Dominic Smith
Abstract: Time series data for robust inflation measures, such as median and trimmed mean inflation, only start in 1977. We extend these series back to 1960 for Personal Consumption Expenditure (PCE) inflation, providing additional episodes of high and rising inflation. We evaluate the robustness of the series along multiple dimensions: First, we find that robust inflation measures tend to diverge in periods of low inflation, but agree when headline inflation is high. The range between the robust measures averages 0.76 percentage points. Second, using yearly instead of monthly inflation when trimming or computing median inflation produces markedly different time series. Third, by contrast, variation in the number of PCE categories used in calculation and trim points for trimmed means do not have significant effects. Finally, we compare the performance of 61 robust inflation measures in predicting (current and future) trend inflation. Trimmed mean measures that trim based on yearly inflation perform best overall, while core inflation performs well when inflation is low, and median inflation consistently underperforms.
Keywords: inflation; robust measures
JEL Codes: E3; E5; E6
Edges that are evidenced by causal inference methods are in orange, and the rest are in light blue.
| Cause | Effect |
|---|---|
| inflation levels (E31) | variability of robust measures (C20) |
| high inflation (E31) | robust measures converge (C52) |
| low inflation (E31) | robust measures diverge (C46) |
| frequency of data used (C80) | resulting inflation measures (E31) |
| set of expenditure categories (H50) | robustness of measures (C52) |
| construction methodology of inflation measures (E31) | effectiveness in predicting trend inflation (E31) |