[R] Holt-Winters code in R or S?
M. Edward (Ed) Borasky
znmeb at aracnet.com
Tue Feb 20 18:20:09 CET 2001
On Tue, 20 Feb 2001, John Aitchison wrote:
> dunno if there is any specifically as such, and a lot of the ad-hoc
> forecasting procedures can be implemented as special cases of AR/MA
> approaches which are variously implemented in ts, tseries, dse etc .. but
> fwiw I just implemented ewma as a programming exercise and will do
> discounted least squares/Holt Winters in the next day or so .. I am
> following the notation and procedure of Harvey "Forecasting, structural
> time series models and the Kalman filter".. anyone who wants this code,
> rudimentary as it is,,just email me.
Do you have a web page? You could post it there. If not, sure, I'll take it.
> I have an interest in the "BATS" approach (Bayesian Analysis of Time
> Series.. as in Harrison, West, Pole ) and somewhere I have some skeleton S
> code that I think I could probably port to R
> I am also working through "Non-linear time series models in empirical
> finance" (Franses & van Dijk) and slowly implementing some bits and pieces
> .. I'd be glad to contribute these as and when.
Now how did you guess I was interested in computational finance? Did you peek
at my web page or something? :-)
> I think this raises the issue , attention BR and the core development
> team, as to whether there is a role for a repository of "slop code" .. use
> at your own risk, probably not a lot of documentation, not a full fledged
> package, perhaps slightly "disreputable", a sort of relaxed and
> comfortable approach and hey, why not?... if there is no right answer to
> an ill-specified question, then maybe it is sensible to admit some other
> For example, I have followed with interest the process of admitting
> factanal() into the fold. got varimax and promax rotations and caveats
> galore, as there should be. Personally, I have always liked procedures
> along the lines of the 'iterative powered vector factor analysis'
> approach in Overall and Klett ~1972 and would be happy to implement such
> .. without having done it in R , I suspect that it is just a few tens of
> lines of code.
> So, is there a role for //RLIB ?.. just R code, not interfaces to DLL's ..
> where people can easily upload simple R functions that can be used
> caveat emptor.
Well, I am planning a (more or less) finance-specific R code repository on *my*
web page, although it was intended only for my *own* creations in this area. I
only have 50 megabytes to work with, so it might not be feasible for me to
serve up *everyone's* R finance code. I was also planning R code and small
demo-sized datasets *only*; no attempt made to provide S compatibility (although
I do intend to test on both Windows and Linux), no packages, C or Fortran, etc.
The working title is "R foR tRadeRs" :-).
I guess my belief about more general repositories is that there are quite a few
already -- the UCLA XLISP-STAT archive, "statlib" at CMU, and many more -- that
we in the statistics game know about. My intent, as implied in the title, was
to provide a bridge to computational finance for traders, including pointers to
downloads of R for Windows, interesting statistical analyses on sample financial
datasets and references to the more readable literature. There are *some*
traders who use Holt-Winters, Kalman filters, ARIMA, neural nets and such, but
very few. A typical trader simulates trading systems over varying parameters,
using rather expensive ($3000 US is a typical price) trading-specific software,
then bases his or her trading decisions on very minimal and often quite non-
robust statistical analyses. I think R presents an opportunity to change this.
znmeb at aracnet.com (M. Edward Borasky) http://www.aracnet.com/~znmeb
Stand-up comedy -- because man does not live by dread alone.
r-help 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-help-request at stat.math.ethz.ch
More information about the R-help