Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
xtdpd n L.n x z, div(x z) dgmmiv(n , lag(2.)) lgmm(n, lag(1)) hascons
where n is the dependent variable, x is a set of time-varyng exogenous
regressors and z is a set of time-invariant exogenous regressors.
When I estimate this type of model on data the coefficients on z are
counter-intuitive and do not resemble what I would expect to see (for
example, they diverge greatly from OLS on the above model, or a static
version of the above). I find this very puzzling. Surely estimation in
levels (which identifies the coefs on z) should not have a dramatic
impact on the coefs of z.
I am also puzzled why one doesn't specify something like the following:
xtdpd n L.n x z, div(x z) liv(x z) dgmmiv(n , lag(2.)) lgmm(n, lag(1)) hascons
xtdpd n L.n x z, iv(x z) dgmmiv(n , lag(2.)) lgmm(n, lag(1)) hascons
xtdpd n L.n x z, div(x z) liv(z) dgmmiv(n , lag(2.)) lgmm(n, lag(1)) hascons
The output of the latter lists first-differenced xs as standard
instruments for estimation in 1st-differences, and levels of z as
standard instruments for the levels equation. Isn't this as required?
The coefficients on z look far more convincing with this version of
the command. If these are not correct, then why - i.e. what are they
doing that they shouldn't be doing?
The manual fails to talk about the estimation of time-invariant
regressors within xtdpd.
Any help/explanation gratefully received.
* For searches and help try: