Re: [Gretl-users] cosmetics for 1.6.0
On Mon, September 4, 2006 22:25, Sven Schreiber wrote:
Thanks for your answers...
Allin Cottrell schrieb:
>> - in spectrum estimation: I can't see any difference between selecting
>> the sub-menu entries "periodogram" or "Bartlett lag window",
the same
>> things appear to happen (namely table and graph).
>
> Well yes, but take a look at the details. One gives the "raw" sample
> periodogram and the other gives a smoothed version (along with
> fractional integration tests).
>
>
ok, but doesn't the term "periodogram" only refer to the non-smoothed
raw estimate? I'm asking because even the smoothed values in the table
have that heading. And vice-versa, isn't the term "spectrum" a little
misleading as a graph title if it's the unsmoothed periodogram?
(I'm not 100% sure though what the most widespread conventions are in
the frequency domain.)
cheers,
sven
As far as I know, the spectrum is a theoretical construct, whereas the
periodogram is its sample counterpart. As a consequence, the periodogram can
be viewed as an estimator of the spectrum. The trouble is that, unlike
autocorrelations, the sample statistic is not a consistent estimator, so you
normally want to use a smoothed version (which _is_ consistent under mild
assumptions); it's no surprise people sometimes use the word "spectrum"
when
they really mean "estimate of the spectrum", be it consistent or not, the same
way we often call "autocorrelations" what really are "sample
autocorrelations".
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"
Ancona