Feasible GLS biased??

[strypste]

Hi

In textbook of wooldridge "Introductory Econometrics A modern approach", I read that Feasible GLS is biased because we have to estimate the variance (the "h" function) using the same data as the mean model. I do not see why fGLS is biased because of that. Any thoughts???

Help is much appreciated!
S

[startz]

One simple explanation is that the estimated weights are themselves random variables. As a result the usual proofs of unbiasedness don't go through. However, feasible GLS is generally consistent.