Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

st: xtdpd without LDV, and with excluded instruments

From   Hewan Belay <>
To   Stata List <>
Subject   st: xtdpd without LDV, and with excluded instruments
Date   Wed, 23 Feb 2011 09:30:30 -0800 (PST)

Dear statalist,

I have two questions regarding xtdpd:

1) I would like to consider incorporating excluded instruments in a dynamic panel data model. To see whether or how this can be done, I carefully perused all discussion on xtdpd (which want to continue using because of its flexibility, rather than xtabond or xtdpdsys) in the Stata-11 manual, and checked prior statalist questions, to no avail. Mechanically speaking, xtdpd does run with the incorporation of excluded instruments in the model, e.g. using the structure
xtdpd L(0/1).Y X, div(Z, nodiff) div(X) dgmmiv(Y)  
which uses the excluded instrument in the differenced-equation, or
xtdpd L(0/1).Y X,  iv(Z, nodiff) div(X) dgmmiv(Y)  
which uses it in both the differenced- and level-equation, or 
xtdpd L(0/1).Y X, liv(Z) div(X) dgmmiv(Y)  
which uses it only in the level equation.
Assuming I do the usual checks eg overidentification tests etc, is this at all a legitimate thing to do? Is there anywhere a discussion of using excluded instruments in xtdpd (dynamic models)?

2) My second question is: I would like to also run my regressions using the GMM estimation method, with lagged variables of RHS endogenous vars as GMM style instruments, etc….. but want a specification that does not incorporate dynamics. In other words, I want to use xtdpd the standard way, but without any lagged dependent variable. Once again, simply dropping the LDV from the command doesn’t stop it from running, e.g. both
xtdpd L(0/1).Y X1 X2, div(X1) dgmmiv(Y X2)  
as well as
xtdpd Y X1 X2, div(X1) dgmmiv(X2)  
run fine. But is this legitimate to do and can I interpret the results for the RHS endogenous and exogenous vars in the usual way that I would if I used the conventional specification with a LDV?

Many thanks for any thoughts on these two questions.


*   For searches and help try:

© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index