A statistical measure of core inflation

This paper examines alternative statistically-based measures of core inflation in Ireland over the period 1976-1999. A highly disaggregated (approximately 500 price series) dataset from the Irish HICP is used. The distribution of quarterly price changes is shown, in common with other international studies, to be highly kurtotic (i.e., fat-tailed) and right skewed. This would suggest there is considerable ‘statistical noise’ in the measured inflation rate, motivating the use of ‘limited influence’ estimators of central tendency over the mean measure on the grounds of statistical efficiency. It is found that even a relatively small amount of trim from both ends of the distribution of price changes results in considerable improvement in root mean square error (RMSE) relative to a benchmark measure of core inflation. This improvement is even larger when monthly data are examined.