Interpretation and Use of Generalized Chow Tests.

The authors examine tests of structural change in m linear models of arbitrary rank. They find that the null hypothesis implicit in these tests is broader than the hypothesis of coefficient equality and that it depends on the ranks of the design matrices constituting the models. As a consequence, these tests have no power against a region of the parameter space that includes nonequality of coefficient vectors. However, these tests possess a UMP property and imposing those restrictions implicit when the null hypothesis is not rejected does produce the conventional pooled estimator. The authors also provide various interpretations of the tests. Copyright 1991 by Economics Department of the University of Pennsylvania and the Osaka University Institute of Social and Economic Research Association.