Re: st: xtivreg and clustered bootstrap

[Stas Kolenikov <skolenik@gmail.com>]

I would venture a guess that if firm and year uniquely identify the
observations in the sense that

bysort year firm : assert _N == 1

or

bysort firm year : assert _N == 1

you are bootstrapping the individual observations, which is very far
from what you want to do (twoway panel). Frankly, I cannot think off
the top of my head of a good bootstrap scheme here, since all
observations in your data set are correlated. I would probably
bootstrap by the -cluster(firm)- and run the two-way panel data
analysis if such a routine (and an estimation method, in principle)
exists. I am pretty sure that this is not what -xtivreg- estimates
though. May be you could -gmm- this somehow.