EViews.com

Hello All,

I have run a cross-section regression using 2SLS that is free from all specification problems except heteroskedasticity.

To correct for this I have used the White method and, as expcted, the coefficients have remained the same, and the standard errors and t-statistics vary slightly.

I then re-ran the misspecification tests and, as when using 2SLS, the model passed all tests (in fact all the test statistics were the same) except the test for heteroskedasticity, implying that although I corrected for the problem, heteroskedasticity was still present. Could somebody just tell me if running the misspecification tests is appropriate after applying the White remedy, or do I just accept that the coefficients are no longer fully efficient and carry on with interpretation as normal?

Thank you in advance,

cyclechic.

cyclechic wrote:Hello All,

I have run a cross-section regression using 2SLS that is free from all specification problems except heteroskedasticity.

To correct for this I have used the White method and, as expcted, the coefficients have remained the same, and the standard errors and t-statistics vary slightly.

I then re-ran the misspecification tests and, as when using 2SLS, the model passed all tests (in fact all the test statistics were the same) except the test for heteroskedasticity, implying that although I corrected for the problem, heteroskedasticity was still present. Could somebody just tell me if running the misspecification tests is appropriate after applying the White remedy, or do I just accept that the coefficients are no longer fully efficient and carry on with interpretation as normal?

Thank you in advance,

cyclechic.

White's method doesn't remove the heteroskedasticity. After all, as you point out, the coefficients are unchanged. White's method only corrects the standard errors.

Ah, ok. Thanks for your help.

To remove the heteroscedasticity.. Weighted Least Squares?

cyclechic

Weighted TSLS is certainly fine, that's essentially GMM. Or you might just stick with 2SLS with White standard errors.