DYNAMIC ASYMPTOTICALLY IDEAL MODELS AND FINITE APPROXIMATION

We extend Barnett and Jonas's asymptotically ideal model (AIM) to mode l for the possibility that the data were generated by a dynamic proces s. Prediction errors for dynamic and static AIM models are compared fo r various simulated datasets. Monetary data are also used to evaluate the AIM specifications. There is substantial evidence that an AR(I) co rrection considerably improves the quality of low-order finite approxi mations of AIM with the cost of estimating only one additional paramet er. Furthermore, restricting a dynamic AIM to approximate only linear homogenous functions often results in severe misspecification.