I have added a small but I think important feature to xtabond2 (version 2.7.0 or later). It has effect in system GMM only. The gmmstyle() option now accepts a -split- suboption. The purpose is to break, for difference-in-Sargan/Hansen testing, the instruments generated by a gmmstyle() option into two groups--on for the levels equation, one for the transformed (differenced) equation. The value I see in this is that it makes it much easier to test the validity of the instruments for the levels equation that are based on first differences of the lagged dependent variable. These are in general the instruments that are most suspect--the ones Blundell and Bond focus on in their 1998 paper that defines the "initial conditions" under which these instruments are valid. Think of it: should we expect the lagged change in the dependent variable to be uncorrelated with the errors in a regression of the current change (which is what we model in a dynamic regression)? I stress the importan!
ce of testing these instruments in my "Short Note on the Theme of Too Many Instruments" paper.
For example, instead of
. webuse abdata
. xtabond2 n L.n L(0/1).(w k) yr1978-yr1984, gmm(L.(w k n)) iv(yr1978-yr1984, eq(level)) h(2) robust twostep
I would do:
. webuse abdata
. xtabond2 n L.n L(0/1).(w k) yr1978-yr1984, gmm(L.n, split) gmm(L.(w k)) iv(yr1978-yr1984, eq(level)) h(2) robust twostep
The two commands generate the same results except that the latter does narrower difference-in-Hansen tests, focusing on instruments based on D.n.
--David
David Roodman
Research Fellow
1776 Massachusetts Avenue NW
Washington, DC 20036
Center for Global Development
Independent Research and Practical Ideas for Global Prosperity
�
Phone:�1 202 416-0723
*
* 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/
