[Gretl-users] Cochrane Orcutt: help!

[Sven Schreiber]

Giovanni Inglisa schrieb:
> Hi all! I'm not sure this is the most appropriate way to ask for
> support...but...here are my doubts
> 
> I'm estimating AR parameters + a constant with a Cochrane-Orcutt
> alghorithm, which is the same used by the AR estimate procedure in gretl.
> 
> Unfortunately I get really different values with respect to the
> constant, while the coefficients for the lagged variables are almost
> equal (and these coefficients are the same as those which result form
> the OLS regression of Y over const+lags) . Is it possible that gretl is
> not just using the cochrane alghoritm but something more sophisticated?
> Or maybe it doesnt'build the regressosr matrix as I do: the constante
> vector + the three arrays of lagged variables.
> 
> Does any of you have an idea? I spent all the day writing this cochrane
> code and I didn't gain any improvement :-((
> 

Giovanni,
I'm not sure exactly what you are doing, but depending on how you write
the model, the constant may need to be transformed using the
autoregressive coefficient. If you really care about the estimate of the
constant (as opposed to the residuals or the other coefficients) you
should make sure that you understand exactly which incarnation of the
constant you care about before estimation. From what you are saying it
really sounds like the models are equivalent; whether or not that
corresponds to some original definition of Cochrane-Orcutt I can't tell,
however.

good luck,
sven