09-23-2011 12:02 PM
I am running a GMM with proc model to correct for heteroscedasticity, but I am having some trouble understanding the SAS diagnostics. My code looks something like that,
1. proc model data=test;
2. parms const a b c;
3. instruments vara varb varc ;
4. y= const+ a*vara+b*varb+c*varc;
5. fit y / white breusch =(1 vara varb varc );
6. fit y / GMM white breusch =(1 vara varb varc );
My understanding is that step 5 gives simple ols results and breush pegan test will show heteroscedasticity if there is any.
but step 6 corrects for heteroscedasticity and hence will give same parameter estimates as step 5 but different standard errors, but breusch-pegan step at this point should show no heteroscedastictiy.
But in my application I see exact same breusch-pegan test value from step 5 and step 6. Am I wrong in interpreting it or am I doing something wrong?
Thanks for any help.
09-23-2011 04:38 PM
As it turns out, I was infact mis-interpreting the results. The data is still heteroscedastic, GMM is only giving heteroscedastic-consistent variance covarince matrix.