Page 1 of 1
two-way panel model
Posted: Sat Nov 29, 2008 1:06 am
by haru
:? EViews can estimate a two-way panel model by choosing the effects from the Fixed and Random Effects boxes.
My question is the following: Suppose that I choose the fixed effect for cross-section and the random effect for period.Then which estimation methods does EViews actually apply? SGLS , LSDV or mixed?
Re: two-way panel model
Posted: Mon Dec 01, 2008 12:49 am
by EViews Glenn
It estimates using the mixed model. LSDV for the fixed effect, and GLS on the remainder (though I'm not certain what the "S" in your "SGLS" refers to).
Re: two-way panel model
Posted: Tue Dec 02, 2008 6:42 pm
by haru
Thank you for your quick reply.
Sorry for my misspelling. "SGLS" should be read as FGLS (Feasible GLS).
However, I am still confused. How does Eviews accommodate the two different estimation methods?
I cannot find any standard textbooks, say Baltagi, Wooldridge, or Arellano, which explain the estimation
for cross-fix and time -random model.
Do you know such a textbook?
Regards,
Re: two-way panel model
Posted: Wed Dec 03, 2008 5:33 am
by EViews Glenn
I don't have a reference handy, but from my recollection the idea is pretty straightforward. For balanced designs, the order that the GLS and fixed effects transformations are applied is not relevant since the Kronecker product can be factored into separate period and cross-section transformations. Thus, you can simply orthogonalize the data with respect to the fixed effect, then apply the standard one-way estimator for random effects to the orthogonalized data.