Dear all,
I estimated a panel with N=20 and T=20 using both -xtreg- fe. In order to
include the LDP as regressor and to test the suspected endogeneity of
another one I tried -xtabond2-.The problem is that the latter should be used
for large panel with short T, that is when N is very large and T is
relatively small. The well known problem here is the number of instruments
which is very big compared with the panel size (more than 200 instruments)
and can overfit the results. After carefully reading David Roodman paper
How to do xtabond2" and the related note on the "Theme of too many
instruments" problem, I tried the different options (lags limit and collapse
options) in order to reduce the number of instruments. As expected the
significance of my regressors change as the number of instruments decreases,
making them not significant at all when I reduce considerably the number of
instruments (ten with all the restrictions).
Now the point is that using -xtreg- with LDP the results are fine nd close
to those obtained without the LDP, but my wish was to use the -xtabond2- for
the sake of robustness. Moreover, considering that without limiting the
number of instruments I get the same results as using -xtreg- fe, does it
mean that also the Within Group FE estimation is not reliable?
Any comment would be greatly appreciated
Ivan
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
