time series in R

Paul Gilbert pgilbert@bank-banque-canada.ca
Mon, 19 Jul 1999 16:42:09 -0400

>As the feature-freeze for 0.65 is probably about 2-3 weeks away, we
>ought to concentrate on getting the basic stuff (bats-like) in first.
>How does that sound as a plan?

I think this is a good idea.

>ar -- at least univariate by Yule-Walker

I'd really like the multivariate one if possible. I'm quite sure ar.yw
in bats is multivariate.

Following is not important for the immediate goal, but I am surprize and
would also like to correct a couple of mis-conceptions.

>As far as I can see (it is very complex) this handles state-space
>and can handle ARMA models by converting them (approximately)
>to state-space form.

No. It handles both ARMA models and state-space models without any
conversion. It can also convert between the representations. In theory
this is algebraic and not approximate, but it is of course subject to
the usual numerical approximations of computers. The tolerances I get
are typically as good as or better than the differences between Solaris
and Linux calculation. (for example, 1e-15 for differences in residuals,
with data having order of magnitude around 1.) Model roots also compare
to very tight tolerances.

>dse is rather un-R-like, and did not compile for me without fixing 20
>or so Fortran errors, and did not load for me even after an hour's
>work fixing many more in the R code.

This surprizes me very much as it has compiled for me in Solaris and
Linux since R 0.16 (although I had to compile directly into R in those
days). At one point there was a bug in the g77 compiler which caused a
problem, but that seems to be fixed now. In earlier versions of R I used
f2c, and at  one point there were some unresolved references which I
fixed (by something simple like changing ^2 to ^2.0). When I used the
Sun Fortran compiler it always worked, but I haven't used it in a long
time now.

>I think the version in devel is
>currently unusable, and has far, far too many conflicts.
The version on CRAN is fairly old (by Internet time). Although the code
has not changed much, R's installation procedures have. I am surprized
that there are conflicts, but am not surprized that there are install
problems with recent versions of R, given all the changes in the way R
installs packages.  I intend to update the CRAN version as soon as R's
installation procedures stabalize. In most respects I believe that has
happened now, but there is a pending change in the building of help
files which I have been waiting to use. I have tried to advertize that
the most recent version at <www.bank-banque-canada/pgilbert> should
install with the most recent versions of R. I guess it would be useful
to have some indication of this on CRAN. Perhaps it would be better not
to leave an old version on CRAN and instead just point to were the most
recent versions are located?

If others are having problems compiling this code please let me know.

Paul Gilbert

r-devel mailing list -- Read http://www.ci.tuwien.ac.at/~hornik/R/R-FAQ.html
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: r-devel-request@stat.math.ethz.ch