Some properties of the durbin-watson test after a preliminary t-test

Abstract

We consider the commonly encountered situation in which the application of the Durbin-Watson test for serial correlation in the errors of a linear regression model is preceded by a preliminary t-test for the significance of one of the regressors. The effect of such pre-testing on the size and power of the Durbin-Watson test is examined by means of a Monte Carlo experiment.

Keywords:

• Pre-testing
• conditional specification
• autocorrelation
• regression model