Re: [Gretl-users] Dealing with heteroskedasticity
On Mon, 21 Mar 2011, Md. Mohan Uddin wrote:
(1) In Gretl there is an option "Robust standard error" for
correcting for
heteroskedasticity.
(2) I can see that there is another option from: Model>other linear
model>heteroskedasticity corrected... in GUI.
My question is when can I use (1) ** "Robust standard error" and when** (2)
**Model>other linear model>heteroskedasticity corrected... for correcting
for heteroskedasticity.*
Robust standard errors give you a means of inference that is
robust with respect to heteroskedasticity, but the point estimates
are not altered: if the model is estimated via OLS you still get
the OLS coefficients.
The "heteroskedasticity corrected" ("hsk") routine not only
revises the standard errors, but also the point estimates
(coefficients); it does so via weighted least squares.
IF the model is correctly specified and the heteroskedasticity is
of a form that can be well approximated as a quadratic function of
the regressors, the hsk estimator is more efficient than OLS.
Allin Cottrell