st: Interaction Variables with regards to xtabond2

[john ebireri <jebireri@yahoo.com>]

Hello,

I am estimating the following panel-data model by system GMM:

y(i, j, t) = a0*y(i, j, t-1) + a1*x(j, t) + a2*x(j, t)*z(i, t) + ...

where the dependent variable y is the growth of value added in manufacturing sector i in country j; x is a country-level explanatory variable of interest; z is a variable measuring a characteristic of manufacturing sectors (which may be common across countries). x is a weakly endogenous variable; z is exogenous. Therefore, x is included in the gmm statement, while z is in the iv statement.

My query is whether the interaction variable x*z should appear in the gmm or the iv statement. I have tried both and it makes a difference. Specifically, the significance of control variables is unaffected by the location of variable x*z; the significance of a1 - i.e., the coefficient of x - is also almost unchanged (it sometimes shows a small drop when x*z is in the gmm statement) but a2 - i.e., the coefficient of x*z - is always insignificant when x*z is in the gmm statement and always significant when x*z is in the iv statement. No misspecification is detected in either case.

I would appreciate any advice. Thank you.

John.    

*
*   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/