L. Mosca says
I'm testing for panel-level heteroskedasticity ..
One form that you might consider is that tested by xttest3 (findit
xttest3). This tests for groupwise H, that is, each panel run
separately would have homo errors, but the sigma^2 are not the same
across panels, so if you run a FE model over the panel you will have
a problem of hetero. I believe this is what the xtgls panels(hetero)
is taking account of. But if you just want to test for it rather than
estimating such a model, xttest3 should do the trick.
