Chow-type Tests Under Heteroscedasticity

Abstract

Inference for structural parameters that are common to several regressions is a frequent problem of statistical practice. In the unlikely case that the residual error variances are equal for all regressions, the inference problem usually has a closed–form solution within the standard regression framework. In the case of unequal and unknown residual error variances, the problem falls into the category of Behrens-Fisher problems. This article describes a comprehensive solution to this inference problem using the concept of a generalized p value introduced by Tsui and Weerahandi in 1989. An example investigating the effect on stock return due to a firm's inclusion in the Standard and Poor's 500 illustrates the method.