st: too small clusters: bootstrapping as solution? (3 dimensional panel, ivreg2)

[Christopher Parker <christopher.parker101@gmail.com>]

Dear Statalisters,

should one bootstrap if the the clusters are too small for normal
clustering? Somewhere on Statalists I have read that if the number of
clusters is too small, bootstrapping would also be not of any help.

For my masther thesis I use panel data where each observation is
uniquley identified by time (30 periods), industry (7), and country
(24). I would like to cluster on industries and countries, at least
this seems most reasonable to me. To my knowledge the number of
clusters is too small in each dimension. Furthermore I obtain (while
clustering for industry and country) the following error message
"Warning: estimated covariance matrix of moment conditions not of full
rank. model tests should be interpreted with caution", which should be
due to the fact that the number of regressors exceeds the number of
clusters.

I run a regression like the following:

ivreg2 D.y l.D.x1 l.D.x2 l.D.x3 (l.D.x4=l2.x4 l3.5), first
cluster(country industry)

My emergency solution would be assuming that correlations over time do
not exist due to differencing, and clustering on country-year and
industry-year level. Even if that is a bit optimistic...

To sumup: Will bootstrapping be of any help in this scenario? If yes,
how do I correctly implement it?

Thanks,

Chris

note: If somebody is wondering why Im using ivreg2 and not xtivreg2,
fd: I prefer ivreg because I can see directly whats going on (are
instruments first differenced etc.)
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/