The use of prior regressions in the estimation of error correction models

Abstract

This note outlines a simple approach to the estimation of error correction models which allows the long-run properties of the data to be isolated and examined separately from the main error correction model. The approach used here is closely related to the proposals of Granger and Weiss (1983), who suggest the use of simple prior regressions as a test for co-integration between two or more variables to be used in a general error correction model. The suggestion here is to set the prior regression up in a slightly different form from that suggested by Granger and Weiss, so that structural misspecification can be more easily detected and corrected.