Bookmark and Share

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]

Re: st: GMM estimation.


From   "JVerkuilen (Gmail)" <jvverkuilen@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: GMM estimation.
Date   Thu, 3 Jan 2013 13:36:50 -0500

On Thu, Jan 3, 2013 at 6:30 AM, Usman Gilani <u.gilani@icloud.com> wrote:

> my question, is that the correct way to input the equation ? because STATA
> giving me the error message "could not calculate numerical derivatives --
> flat or discontinuous region encountered"
>
> please suggests me how can I perform this estimation

As I said previously, this model is so complex it's next to impossible
to diagnose what's going wrong. You have so many failure points
there's no telling where it's breaking. Just a guess but I suspect
that it's unidentified, but tracking that down will not be easy. It
could also be that you are giving it bad starting values and it's
diverging.

Simplify dramatically and build up. Find an example of the kind of
model you want to run that has only one or two variables in it using
data that you know work. Never mind that it's not the model you want
to run. Gain experience with specifying it in -gmm- syntax and seeing
situations of what a working model looks like and then translate that
over to a one or two variable subset of your data. Then build up to
the model you want to run. If it crashes on only one or two variables
then it's not identified. If it starts diverging or performing badly
as you add variables, chances are good that you have an empirical
identification issue (i.e., insufficient data). You can explore this
by using simulated datasets.

If you look at how Stata estimates a model such as -xtmelogit- it does
exactly this strategy to generate good starting values.

Yes, this seems like a pain in the zxx, but it's the only thing that
works. These kinds of nonlinear models are simply not easy to work
with and require a lot from the user.



-- 
JVVerkuilen, PhD
jvverkuilen@gmail.com

http://lesswrong.com/
*
*   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/


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