RDP 2006-10: The Performance of Trimmed Mean Measures of Underlying Inflation 4. Data

Our analysis uses disaggregated price data for the CPI for Australia, the United States and Japan, and the harmonised index of consumer prices (HICP, hereafter referred to as the CPI) for the euro area. Our focus on the US CPI rather than on the PCE chain price index is mostly for consistency with the other economies in our study.

The disaggregated data used in our empirical work are monthly for the United States, euro area and Japan, and quarterly for Australia, as published by the statistical authorities in each economy.[18] For Australia, we choose to focus on the low-inflation period starting from the March quarter of 1993. For consistency we present results for the United States from January 1993 onwards, though we have calculated results back to January 1987 and obtained results that are broadly similar to those shown here. For the euro area we were constrained by data availability and present results for the period from January 1996. For Japan, we present results from April 1997, so that the effects of the 1997 VAT increase are excluded. The sample period for our assessment of the performance of the underlying measures ends in September 2006, although Figures 2, 3 and 8 show data to October 2006 for the US and the euro area.

Figure 3: Consumer Prices
Monthly percentage change
Figure 3: Consumer Prices

Note: (a) Quarterly percentage change

Our analysis uses expenditure items at a relatively fine level of disaggregation.[19] In the case of Australia, Japan and the euro area, the CPI and its components are published in non-seasonally adjusted terms only. In the case of the United States, the calculation of the overall CPI uses seasonally adjusted data for some components and unadjusted data for the rest (although a comprehensive list of which components are seasonally adjusted over time was not readily available). Given the importance of seasonal adjustment in calculating trimmed mean measures, we have seasonally adjusted the data for all four economies using the Census X12 method in EViews, adjusting all components where a regular seasonal pattern could be identified.

Based on these data, we construct seasonally adjusted CPIs for the four economies. We also calculate seasonally adjusted exclusion measures of core inflation.[20] And we calculate a range of trimmed mean measures using symmetric trims gradually trimming from 1 to 50 per cent of the price distribution from both ends.

We provide a first look at the trimmed mean measures of underlying inflation in Figure 3, along with headline CPI inflation (that is, the 0 per cent trimmed mean constructed from our seasonally adjusted data) for each economy. There are 50 different trims that we could illustrate, so we present the middle of these, the 25 per cent trim, for illustrative purposes. It is immediately obvious that monthly or quarterly inflation as measured by the 25 per cent trimmed mean is much less noisy than as measured by the CPI, suggesting that trimmed means may be promising measures of underlying inflation. We examine this proposition in more detail in the next section.


Namely the ABS for Australia, Bureau of Labor Statistics for the US, Eurostat for the euro area and Statistics Bureau, Ministry of Internal Affairs and Communications for Japan. For Australia, we exclude interest charges up to 1998, and we make one specific adjustment to remove the effect of the tax changes of 1999–2000 from individual expenditure items. In the case of Japan, CPI weights prior to 2000 were not available and so are approximated using data from the Family Income and Expenditure Survey (on which the actual CPI weights are based). Otherwise, the data are as used by the statistical authorities in the construction of CPIs. Nevertheless, there were some minor discrepancies between our replication of CPI inflation and the CPI data published by statistical authorities, presumably reflecting rounding in input data. For consistency, the trimmed mean, exclusion (except in the case of Japan – see Footnote 20) and headline inflation measures used in this paper are all calculated from the same set of disaggregated prices and weights data. [18]

The number of CPI components varies slightly over time as the composition of the CPIs changes, but in the most recent period is 90 for Australia, 70 for the United States, 50 for Japan and 93 for the euro area. [19]

Items excluded are: for Australia, petrol and fruit & vegetables; for the US and Japan, food and energy; and for the euro area, food, energy, alcohol and tobacco. In the case of Australia, the US and euro area, we construct exclusion measures using seasonally adjusted components where appropriate. For Japan, we do not have sufficiently disaggregated data to do this so we seasonally adjust the published exclusion measure. [20]