Use of the RESET test and AR(1) disturbances

[Bigbrotherjx]

I have a theoretical model which gives me a relationship between two variables in levels.

My model fails the Ramsey RESET test...but does it make any sense to adding squared or cubed explanatory variables since this would violate economic logic?

Secondly, the model has some serial correlation problems. I know that imposing AR(1) disturbances implies the imposition of some untested coefficient restrictions. Is it better just to use Newey-West?

Thirdly, in the presence of serial correlation/heteroscedasticity, is SUR still more efficient than equation by equation OLS provided that there is cross-dependence in the disturbances (the original motivation for SUR)?