Issues with xtabond2 and xtdpdsys

[Karun Advani]

Dear All,

I am currently using a panel dataset to investigate the relationship between the growth of poverty and economic growth along with other covariates. I have some issues with my baseline model when using xtabond2 in comparison to xtdpdsys. After estimating these baseline models, I would add in more covariates that may also be endogenous (such as corruption) but following the same methodology as the baseline regressions.

When I am using xtdpdsys with the following command:

xtdpdsys grlhcountglob grlminc, vce(robust) twostep

The result works fine. However, I would like to compute the Hansen over-id test in addition to the Sargan test as the Sargan test may not be asymptotically valid for a system-GMM estimator (https://www.statalist.org/forums/for...n-test-results).




The xtabond2 command would allow me to do this. However, when I run the following:


xtabond2 grlhcountglob grlminc, robust twostep





I get the following error:

"Equation not identified. Regessors outnumber instruments."




I am unsure why I get this error with xtabond2 but not with xtdpdsys.




It seems to work if I run:

xtabond2 grlhcountglob grlminc, gmmstyle(lhcountglob) twostep





I am not sure whether this would use the same instrument set with differenced and levels instrument and automatic lag selection like in xtdpdsys. Do you think there is a way to work around this problem?





Thank you very much

[Sebastian Kripfganz]

With xtabond2, you need to specify all instruments manually. That means, in your first xtabond2 example there are zero instruments. xtdpdsys instead automatically creates the instruments for you. Please see the following topic on how to specify equivalent command lines with xtabond2 and xtdpdsys: how to make xtdpdsys and xtabond2 equivalent?

[Karun Advani]

Thank you for your reply Sebastian.