11-05-2012 12:57 AM
Let y1ij and y2ij denote the ith observation in group j. I want to estimate the 2 simultaneous equation system
y1ij = a1j + b*xij + c*zij + error1ij
y2ij = a2j + b*xij + d*wij + error2ij
It is easy for me to do this for a pooled 3SLS regression using PROC SYSLIN, but doing a pooled regression forces a1j to be the same for all j. I need to account for heterogeneity in some way (not necessarily with a fixed effect model, but that's the example I give above). The problem is, creating dummy variables for each j is not feasible because there are about 13,000 different j.
Does anyone have a suggestion how to go about this? I'm open to a random effects model, or GMM, or something else, so long as I can account for both simultaneity and heterogeneity.