st: instrumenting interaction and Hansen J statistic

["Bernini, Michele" <Michele.Bernini@unitn.it>]

Dear statalister,

I am using the command ivreg2 to estimate by 2sls this kind of model:

y= A + B + ( A*X   B*X   X =  A*Z1   A*Z_2    B*Z_1   B*Z_2    Z_1   Z_2) +controls

where X is the endogenous variable that I want to interact with the exogenous regressors A and B, and Z_1 and Z_2 are my two instruments. I had a look on the internet and it seems legitimate to instrument interaction terms this way. However the Hansen statistic is not convincing. When I include all the 6 instruments it gives me a p-value of 0.39, however when I use only 4 of them (in different combinations) the p value gets as smaller as 0.07. First stage statistics look rather good (high F values and K-P stat).
I was wondering whether the failure to reject the validity of the instruments could be caused by high collinearity among them and if someone has a better strategy to instrument the interaction terms.

Thanks a lot,

Michele

p.s. sorry if this email reaches you twice but I am not sure that the first one was sent correctly




*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/