Clustering Standard Errors with Pooled-Mean-Group Estimation (xtpmg, pmg)

[Bea K]

Dear Stata users,

I am using the pooled mean group estimator (“pmg”) provided in the “xtpmg” command by Blackburne and Frank (2007). I would like to cluster my standard errors. However, when I enter the option “cluster(id)”, I get the following message: “ Cluster option only meaningful with DFE. Ignoring option and continuing.” Why is clustering not possible with the pmg-estimator? Has it just not been programmed yet or is it technically impossible (i.e. not meaningful)?

This is my command: xtpmg d.c d.pi d.y, lr(l.c pi c) pmg cluster(id)

Thank you for your help.

Bea

[Jeff Wooldridge]

This command is intended to estimate models for nonstationary panels when T is large (where "nonstationary" also means "not weakly dependent). Clustering, as far as I know, has not been justified in this setting. That's probably why it's not allowed. Now, it could be valid, but I think that hasn't been shown.

If you have large N, small T, then clustering would be valid -- but then you should probably be using a different approach.

[Bea K]

Thank you very much for your quick reply. Is there a specific reference that I could use in order to make this point?

[JoeSchmidt]

Dear Stata users,

I am using the pooled mean group estimator (“pmg”) provided in the “xtpmg” command by Blackburne and Frank (2007). I would like to cluster my standard errors. However, when I enter the option “cluster(id)”, I get the following message: “ Cluster option only meaningful with DFE. Ignoring option and continuing.” Why is clustering not possible with the pmg-estimator? Has it just not been programmed yet or is it technically impossible (i.e. not meaningful)?

This is my command: xtpmg d.c d.pi d.y, lr(l.c pi c) pmg cluster(id)

Thank you for your help.

Bea

It is possible to cluster, but perhaps not in the way you might intend. See the help file and the article by Blackburne and Frank in Stata Journal. http://www.stata-journal.com/sjpdf.h...iclenum=st0125

Quoting from that article:

"The dynamic FE estimator, like the PMG estimator, restricts the coecients of the cointegrating
vector to be equal across all panels. The FE model further restricts the speed
of adjustment coecient and the short-run coecients to be equal. xtpmg with the dfe
option ts the model in (8) while allowing panel-speci c intercepts.8 An allowance for
intragroup correlation in the calculation of standard errors is made with the cluster()
option."

Try adding dfe to your code, assuming that is meaningful for your research question. Without knowing the context the code below may be garbage but it may allow you to cluster and avoid the error message.

xtpmg d.c d.pi d.y, lr(l.c pi c) pmg dfe cluster(id)

The model estimated by the dfe/cluster option is quite different to the core PMG model. The article confirms that using this option means that the xtpmg command becomes a wrapper for Stata's xtreg, fe command. So it seems this command doesn't allow you to cluster using PMG itself.

[Brahim KHOUILED]

I have the same problem; is there a solution, please?