On Mon, 19 Jul 1999, Mark Lehr wrote:
> Help
> For those of you who use the Splus Function
> ARIMA.MLE
> there are many outputs from this routine but
> one seems to be lacking. The time series model
> returned provides the AR and MA coefficient estimates
> with their covariance matrix.
> Also returned is the innovation variance but is
> there a variance of this term calculated. In other
> words the variance of the variance. It is an
> estimated parameter just like the AR and MA
> coefficients and I would expect this as well. Maybe
When you estimate a regression, you get a residual
variance, but no variance for (the estimator of) that term.
This is exactly the same (just an autoregression, not a regression).
As far as I know the same reasoning applies (it certainly does
in the AR(p) case):
n s^2/sigma^2 ~ chisq_n
where n is approximately the length of the series, and this gives
you an approximate variance for s^2. As in regression, the
estimators of the coefficients and s^2 are approximately uncorrelated.
If this matters, please check the asymptotic theory in, for example,
Brockwell & Davis.
--
Brian D. Ripley, ripley@stats.ox.ac.uk
Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
University of Oxford, Tel: +44 1865 272861 (self)
1 South Parks Road, +44 1865 272860 (secr)
Oxford OX1 3TG, UK Fax: +44 1865 272595
-----------------------------------------------------------------------
This message was distributed by s-news@wubios.wustl.edu. To unsubscribe
send e-mail to s-news-request@wubios.wustl.edu with the BODY of the
message: unsubscribe s-news
|