Use of the Chow Test under Heteroscedasticity

Abstract

the assumption of equality of variances of error terms between two separate sample regimes. It is shown that the test is well behaved when there are variations of variances if at least one of two sample sizes is very large. However, if two samples are of small size, even moderate heteroscedasticity has considerable effect on the level of significance of the test. In what follows, it is assumed that T1 samples belong to the first regime and T2 samples to the second regime; subcripts 1 and 2 denote the first and the second regimes, respectively. Consider the regression model