xtreg won't take aweights with vce(bootstrap)

[paulvonhippel]

xtreg can use aweights, and it can provide boostrapped standard errors with vce(bootstrap). But it can't do both at once.
Is there any command that can? I'd like to get the efficient point estimates that I get using aweights, and still get the accurate standard errors that I get using vce(bootstrap).

I know I can roll my own solution using the bootstrap command, but I'd rather not.

[paulvonhippel]

Related: xtreg will take aweights with cluster() standard errors, but the resulting standard errors appear implausibly large. I'm thinking aweights don't play nicely with cluster(), as cluster() and aweight handle heteroskedasticity in perhaps incompatible ways.

Background: These questions arise from my attempt to use aweights for heteroskedasticity while handling serial correlation in some other way -- either cluster() or bootstrap().

[paulvonhippel]

I've now tried two ways of using bootstrap with aweight, and they both generate the error "weights not allowed." What's the best workaround? Here are the two approaches I've tried:
(1) xtreg ...[aweight=w], fe vce(bootstrap)
(2) bootstrap, cluster(panelid): xtreg...[aweight=w], fe

[Sergio Correia]

I don't think it can be done. You can't even run freq. weights with boostrap in xtreg, although the bootstrap command does allow freq. weights.

Maybe *help bootstrap*, as well as a few online discussions such as http://www.stata.com/meeting/snasug0...v_snasug08.pdf could help.

[paulvonhippel]

It turns out it can be done using the force option, e.g.:
xtreg ...[aweight=w], fe vce(bootstrap, force)