xtgls or xtpcse

[Rosalia Fatmawati]

I have a data panel, and after all the test for classic assumption my dataset has a heteroskedasticity , multicolineary , and autocorrelation issues.
My dataset is T > N
i've tried to do something, and everything ruined
so i browse and get that xtgls / xtpcse command can fix it.
anyone can help?

[Joro Kolev]

-xtpcse- resolves all of the problem that you mention. -xtgls- is better (more efficient estimator) but to make it take care of heteroskedasticity you need to work a lot.

If your test for heteroskedasticity shows you that you have it, just proceed with -xtpcse-.

If you want us to show you how to do it exactly, check -dataex- and post some data with which we can work

[Phil Bromiley]

Let me add to Joro's helpful comment that I don't think either of these estimators estimates a fixed or random effects model. So, while one is getting a much more sophisticated model of the error term, this is that the cost of forgoing the fixed or random effects. As the documentation notes, both are consistent if the conditional mean is correctly specified, but if there should be random or fixed effects and one doesn't have them than the conditional mean may not be correctly specified.

[Muhammad Ahmad]

Hi,
In my panel data below tests are showing that there is heteroskedasticity and autocorrelation problem

xtserial variables

hettest

I am using xtpcse, which command (from the list below) should i use in this case ?

xtpcse depvar indepvar timedummies sectordummies, correlation(psar1)

xtpcse depvar indepvar timedummies sectordummies, hetonly correlation(psar1)

xtpcse depvar indepvar timedummies sectordummies, independent correlation(psar1)