A point optimal test for heteroscedastic disturbances

A new test for additive heteroscedasticity in the disturbances of the linear regression model is proposed. Power functions of various forms of the new test are compared empirically with those of currently favoured tests for a range of heteroscedastic models. The results highlight the power advantage of a test which is MP1 at a central point in the alternative hypothesis parameter space. The main conclusion is that the recommended version of the new test is generally more powerful than existing tests against medium and severe heteroscedasticity, whereas the King and Szroeter tests perform better against weak heteroscedasticity.