Dear Christopher,
Thanks for the reply. I too was suspicious of using SURE.
As I have different number of obs in every sample it
makes it stronger the argument of not using it. I guess I
have to pool the sample to test for coefficient differences.
It will however introduce endogeneity in the estimation,
exactly the reason why I had to split the sample.
Sincerely,
Bersant
It is not clear why pooling the data, and using a fully interacted
specification (the dummy variable approach), using an instrumental
variables (or GMM) technique would not give you what you want. That
would handle the non-independence, and would also allow you to test for
coefficient stability across subsamples.