The Cost of Heterogeneity: Can Density Peaks Clustering Improve CES Aggregation?

Abstract

Listen

Translate article

This paper applies recent advances in machine learning to a long-standing macroeconomics question by using density peaks clustering to improve estimation of aggregate price indexes. To measure a price index that properly accounts for consumer welfare, it is important to consider the effect of changes in product variety. Standard methods for estimating this effect using CES demand systems implicitly rely on the assumption that consumer tastes can be accurately represented by a single taste parameter. However, if consumers have heterogeneous unobserved tastes across goods, the estimated aggregate elasticity will tend to be lower than that of the groups. By clustering consumers into groups that share similar tastes, we can accurately estimate elasticity and get more representative measurements for each consumer’s cost of living. Applying the method to a panel of consumer retail purchases, I show that methods that ignore consumer heterogeneity imply elasticities on average 20% below my method and inflation rates about half a percentage point per year lower than a weighted average of the heterogeneous groups.