Pooled FMOLS

[ylijohtaja]

Hi,
I wonder whether there is a procedure to implement POOLED FMOLS with RATS? It appears that the @panelfm always estimates mean-group FMOLS.
Cheers,
Y

[TomDoan]

@PANELFM offers three choices for weighting the individual estimates only one of which is actually "mean-group". What exactly do you mean by Pooled FMOLS? What's heterogeneous and what's homogeneous? You can apply @FM to panel data since all the instructions will respect panel data boundaries; however, that would only be appropriate if the model were completely homogeneous, down to the intercept.

[ylijohtaja]

Many thanks Tom,
Sorry for not being clear enough: What I would need to do is to estimate a fixed-effects FMOLS, where the slope coefficients are homogenous across cross-sections but intercepts are allowed to vary between cross-sections. If I understood correctly, @FM does not too that, since it does not allow for heterogeneous intercepts?
Cheers,
Y

[TomDoan]

OK, but there's a third "player" in FM, which is the correction for serial correlation/simultaneity. What's the assumption you're making regarding that? @PANELFM assumes that the intercepts and the corrections are heterogeneous---any other assumption than those two seems far-fetched. You seem to be concerned that the mean group estimate averages separately estimated cointegrating vectors, but in fact, if you work out the math, a pooled panel regression (for least squares) is itself a precision-weighted average of the individual member estimates. So if you don't want to do equal weights, do the precision weighted average.