st: xtabond2: which cross-sections should be used with instruments lagged3+ periods

[Marcello Pagano <pagano@hsph.harvard.edu>]

For Alessandra:

From:  alessandra guariglia

Dear All;

I have the following questions:

I have an unbalanced panel and need to estimate a static investment 
Equation. I need to use endogenous variables lagged 3 and 4
times as  instruments. I therefore use the following command:

xtabond2 ik cfk y11-y118 , gmm(ik cfk, laglimits (3 4) ) iv(y11-y118 ) 
noleveleq robust small  (ik is investment; cfk, cash flow; y11-y118, 
time dummies)

I notice that when I run this command, Stata uses all the observations  
in the sample in estimation, except the first one for each firm  (which 
is obviously lost due to first-differencing).

My question is the following: when instruments lagged 3 and 4 times  
are used as instruments shouldn't the first three cross-sections for  
each firm be automatically dropped? The reason why I am asking is that 
for those equations corresponding to the first-three cross-sections only 
time dummies would be used as instruments.

I also noticed that if I use a balanced panel and run the 2  commands:

xtabond2 ik cfk y11-y118 , gmm(ik cfk, laglimits (3 4) ) iv(y11-y118) 
noleveleq robust small xtabond2 ik cfk y11-y118 if index>3 , gmm(ik cfk, 
laglimits (3 4) ) iv(y11-y118 ) noleveleq robust small

(where index takes values 1, 2, 3, 4 etc. within each firm), I get the 
same one-step estimates, whereas I get very different estimates when I 
run the same commands on an unbalanced panel. Why do you think this is 
the case?

Any help would be greatly appreciated

Alessandra
*
*   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/