ACCOUNTING FOR AGGREGATION BIAS IN ALMOST IDEAL DEMAND SYSTEMS

This study revisits the consistent aggregation (over households) property of almost ideal demand system (AIDS) models and presents a method to explicitly account for expenditure aggregation bias when estimating the aggregate AIDS model with time-series data. Ignoring aggregation bias can lead to biased and inconsistent parameter estimates and can cause aggregate demand functions to be inconsistent with the demand functions at the individual household level. Recognizing the general limited information contained in aggregate time-series data for explicitly modeling aggregation bias, we present a new method of constructing an aggregation bias term that is derived from the proportions of household in different income groups. This information is generally available in developed economies. We use this framework to estimate aggregate meat demand within a complete demand system based on U.S. annual expenditure data.