On Aggregation of Linear Dynamic Models
This paper provides a general framework for aggregating linear dynamic models by deriving the aggregate model as an optimal prediction of the aggregate variable of interest with respect to an aggregate information set generated by current and past values of available aggregate observations. The paper shows how the results in the literature can be readily obtained using the proposed forecasting approach, and provides a number of important extensions and generalisations. In particular, it does not require the assumption of independence of the micro distributed lag coefficients, and establishes that in general the long-run coefficients obtained from the optimal aggregate relation are equal to the averages of the long-run coefficients from the micro relations. Finally, the approach advocated in the paper is applied to aggregation of life-cycle decision rules under habit formation, and the implications of the heterogeneity in habit formation coefficients across individuals for the analysis of aggregate consumption are investigated.