R help - Jun 2009 - Where can I find information on how to subsample a time series?

I suspect I'm looking in the wrong places, so guidance to the relevant
documentation would be as welcome as a little code snippet.

I have time series data stored in a MySQL database.  There is the usual DATE
field, along with a double precision number: there are daily values
(including only normal working days: Monday through Friday).  I actually
have to do a couple things here.  Because of how the result is to be used, I
need to first create two time series.  The first is the delta between 22
working days, and the second is the delta between 66 working days.  I have
hundreds of these datasets, and some go back 30 years.  I need to estimate
the correlation between 22 day deltas (i.e. is the delta for one month
correlated with that of the previous month) and between the 22 day delta and
the 66 day delta that ends the day before the the first day of the 22 day
delta.  However, I KNOW the statistical properties of the time series are
not constant (so the usual assumptions do not apply to the entire series).
Therefore, I want to subsample finely enough to get a reasonably sensible
correlation and examine how that changes through time.  (There are no tests
of significance here: I just want to explore just how much the properties of
these series change through time).

I have C++ code, admittedly not written particularly efficiently, that does
this.  The question is, is it possible to do this reasonably efficiently
using R?

Thanks

Ted

	[[alternative HTML version deleted]]

assuming you pull the data you want into x and y:

whit at ubuntu:~$ R
> library(fts)
> x <- fts()
> y <- fts()
> xy.cor.200 <- moving.cor(x,y,200)
> tail(xy.cor.200)
                 [,1]
2012-03-12 -0.3009635
2012-03-13 -0.2923489
2012-03-14 -0.2824015
2012-03-15 -0.2662689
2012-03-16 -0.2566354
2012-03-17 -0.2537089
2012-03-18 -0.2490421
2012-03-19 -0.2391911
2012-03-20 -0.2263381
2012-03-21 -0.2113029
>
which is just using c++ to do the calculation.

here is the template function for correlation that fts uses:
http://github.com/armstrtw/tslib/blob/5b0fe2fc5ecb393d1dca097c2c19008227eb6c7e/tslib/vector.summary/cor.hpp

-Whit

On Fri, Jun 26, 2009 at 4:57 PM, Ted Byers<r.ted.byers at gmail.com>
wrote:
> I suspect I'm looking in the wrong places, so guidance to the relevant
> documentation would be as welcome as a little code snippet.
>
> I have time series data stored in a MySQL database. ?There is the usual
DATE
> field, along with a double precision number: there are daily values
> (including only normal working days: Monday through Friday). ?I actually
> have to do a couple things here. ?Because of how the result is to be used,
I
> need to first create two time series. ?The first is the delta between 22
> working days, and the second is the delta between 66 working days. ?I have
> hundreds of these datasets, and some go back 30 years. ?I need to estimate
> the correlation between 22 day deltas (i.e. is the delta for one month
> correlated with that of the previous month) and between the 22 day delta
and
> the 66 day delta that ends the day before the the first day of the 22 day
> delta. ?However, I KNOW the statistical properties of the time series are
> not constant (so the usual assumptions do not apply to the entire series).
> Therefore, I want to subsample finely enough to get a reasonably sensible
> correlation and examine how that changes through time. ?(There are no tests
> of significance here: I just want to explore just how much the properties
of
> these series change through time).
>
> I have C++ code, admittedly not written particularly efficiently, that does
> this. ?The question is, is it possible to do this reasonably efficiently
> using R?
>
> Thanks
>
> Ted
>
> ? ? ? ?[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.
>