st: Re: on using xtabond2

["Zou Hong" <hzou@ln.edu.hk>]

Dear listers,

I have some questions (probably they are navie) about xtabond2. I tried to 
look for answers in the posted questions but find very little relevant. I 
will be appreciated if anyone can help me clarify the following points:

1. If a regressor does not appear in gmm( ) and iv( ), how is it treated in 
xtabond2? Is it a good practice to have a regressor included in neither 
gmm() or iv()??

2. In using xtabond2 to estimate the following system of two-equation 
simultaneous model. Will xtabond2 simply treat each equation as a separate 
equation in estimation?

y(i,t)=a1+b1*y(i,t-1)+c1*x(i,t)+d1*z1(i,t)+u(i,t)
x(i,t)=a2+b2*x(i,t-1)+c2*y(i,t)+d2*z2(i,t)+v(i,t)

3. Since in xtabond2, a predetermined variable (e.g., w) can be dealt with 
in two ways: gmm(w) or iv(w, equation(level)) in system GMM.
I wonder when to use either of the two ways.

4. In Roodman (2006) "how to use xtabond2", p.32:
The Stata code contains "iv(yr*, equation(level)), here yr* are treated as 
predetermined and are only used as instruments of the level equation. Why 
year dummies can be not exogenous?

In p.21, there is a code:
Ivreg D.n (D.nL1=nL2) D.(nL2 w wL1 k kL1 kL2 ys ysL1 ysL2 yr1979-yr1983)

Is it meaningful to difference year dummies?

5. I have a data set (T=8, N=550) and have the one-period lag of Y and some 
variables that are not strictly exogenous in the model. I have tried many 
different specifications of gmm() and iv(), but the Hansen test of 
over-identification still cannot pass (always significant).

I carefully spcified and split the gmm() option into several gmm() 
sub-options. Judging from the C-statistics, the main reason for the failure 
of the Hansen test is that the instruments of Yi,t-1 are not good enough 
despite my use of gmm(y, lag(3 .)).

Can this happen simply because my dataset is not suitable for GMM analysis 
(no valid instruments from the existing variables) or it is most likely due 
to my wrong specification of the gmm() and iv() options?

Can it arise because the model may have missed some important variables?

Many thanks
Joe 

*
*   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/