Improving The Measurement Of Core Inflation In Colombia Using Asymmetric Trimmed Means

Variations in the Consumer Price Index (CPI) are a key source of information about inflationary trends for monetary authorities. However, month to month movements may often be misleading. Monthly CPI changes are notorious for their volatility and frequent short-term swings. The noise in high-frequency CPI data is compounded by flaws in the standard measure of overall inflation— i.e., the weighted mean of price changes. This measure is a least variance estimator of economy-wide price changes only if the price change distribution is normal. Empirical studies have rejected this assumption for a large spectrum of countries.2 A popular method to deal with the noise in price data is to eliminate the food and energy components from the CPI. By excluding sectors that are presumed to exhibit a high noise-to-signal ratio, the remaining index is expected to be dominated by items that contain more information about persistent trends in prices. Successful application of this technique requires that food and energy sectors have little influence in long-term inflationary trends, a valid assumption when they represent a relatively minor share of the consumer basket. This requirement is often not met in low and middle income economies, where food expenditures, in particular, may account for a large share of consumer outlays. Recent work on improved indicators of underlying inflation suggest that statistics such as the median price change and trimmed averages (i.e., means of subsamples of component CPI price change information) may have desirable characteristics (Bryan and Ceccheti, 1993a).3 Broadly, these indicators reduce the influence of highly volatile prices, irrespective of their sector of origin. Unfortunately, little effort has been made to evaluate how much improvement can be expected from these estimators. A pioneering effort in this area is the recent evaluation of symmetric trimmed means for the U.S. by Bryan et al. (1997).