Assumptions of HausmanTest variance calc. inconsistent?

[reto2410]

Hello Everybody

I got a nice warning for a performed Hausman Test in EViews 6.

** Warning: robust standard errors may not be consistent with
assumptions of Hausman test variance calculation.

Does anybody know why this can happen and where to get information about the Hausman test assumptions and how EViews does calculate the Hausman statistics? I searched the EViews 6 User's Guide and some Books (e.g. Baltagi(2001)) but didn't find anything.

I know that it can happen, that the test variance is invalid when the convarianc estimate {V[Beta(FE)]-V[Beta(RE)]} to test {Beta(FE)-Beta(RE)} is negativ definite. But then I get the EViews answer "* Cross-section test variance is invalid. Hausman statistic set to zero" which is not the same as the one mentioned above.

Thanks for your help
Reto

[EViews Glenn]

Any textbook that describes the Hausman test should help you out with the assumptions. The computation of the variance in the Hausman statistic requires that the random effects estimator be fully efficient under the null hypothesis. Specification of robust standard errors suggests, but does not impose, that you believe that the random effects estimator is not fully efficient. Hence the warning.

The reason you probably won't find any textbooks that describe this particular issue is that textbooks focus on the application of the Hausman statistic in the simplest setting...straightforward RE and FE using the conventional estimators of the coefficient covariances. It's only when you start generalizing that issues of this sort come up. I should point out that EViews will not always warn you in cases like this, it's really up to the user to understand what they are doing, but where we can, we try to help out.

[reto2410]

Thank you Glenn