This paper considers the point optimal tests for AR(l) errors in the linear regression model. It is shown that these tests have the same limiting power characteristics as the Durbin-Watson test. The limiting power is zero or one when the regression has no intercept, but lies strictly between these values when an intercept is included.
