Files
Abstract
We study the exact power of the Goldfeld-Quandt test in a linear regression model with errors which are both heteroscedastic and autocorrelated. The test is not robust to this form of mis-specification, but is less sensitive to autocorrelation in smaller samples.