Evaluating core inflation measures: A statistical inference approach

Abstract

We propose a framework for consistently evaluating core inflation measures via a straightforward application of sound statistical inference principles. Under this framework, inflation measures (both headline and core) are regarded as estimators tracking the economy’s true, unobserved inflation rate. We depart from the arbitrary convention in the literature of approximating true (or trend) inflation as some moving average of the observed headline inflation. Instead, we regard trend inflation as the unobserved inflation rate that corresponds to the whole population of consumer price changes while the observed inflation measures are estimators of trend inflation based on particular samples of consumer price changes. Hence, the evaluation of inflation measures is rigorously derived from the sampling distribution properties of the corresponding estimators, in contrast to the use of ad hoc criteria for evaluating core inflation measures, prevalent both in the academic literature and in most central banks’ practices. We implement our evaluation approach for the Guatemalan Consumer Price Index (CPI) data by applying a computational bootstrapping technique. Finally, we showcase the evaluation results for the Guatemalan data regarding the performance of some widely used core inflation measures.