Michael Reiter, Universitat Pompeu Fabra, Barcelona

Evaluation of Discrete and Continuous Time Dynamic Models by Spectral Methods with an Application to Automobile Demand

November 1994

Abstract

The paper argues that structural economic models should be evaluated by comparing the multivariate spectral densities implied by the model with those directly estimated from the data. The application of spectral methods presupposes a stationary transformation of the model as well as of the data. The paper describes in detail how this can be done in a consistent way for a wide class of linear dynamic models, formulated as state space models in discrete and continuous time. Two simple examples demonstrate the use of the method with nonlinear models. Finally, the method is illustrated for a model of demand for consumer durables, applied to annual data of German automobile purchases. A discrete time and a continuous time version of the model are estimated and evaluated.