[R] C/C++/Fortran Rolling Window Regressions

[Achim Zeileis]

Jeremiah,

for this purpose there are the "roll" and "RcppRoll" packages. Both use 
Rcpp and the former also provides rolling lm models. The latter has a 
generic interface that let's you define your own function.

One thing to pay attention to, though, is the numerical reliability. 
Especially on large time series with relatively short windows there is a 
good chance of encountering numerically challenging situations. The QR 
decomposition used by lm is fairly robust while other more straightforward 
matrix multiplications may not be. This should be kept in mind when 
writing your own Rcpp code for plugging it into RcppRoll.

But I haven't check what the roll package does and how reliable that is...

hth,
Z

On Thu, 21 Jul 2016, jeremiah rounds wrote:

> Hi,
>
> A not unusual task is performing a multiple regression in a rolling window
> on a time-series.    A standard piece of advice for doing in R is something
> like the code that follows at the end of the email.  I am currently using
> an "embed" variant of that code and that piece of advice is out there too.
>
> But, it occurs to me that for such an easily specified matrix operation
> standard R code is really slow.   rollapply constantly returns to R
> interpreter at each window step for a new lm.   All lm is at its heart is
> (X^t X)^(-1) * Xy,  and if you think about doing that with Rcpp in rolling
> window you are just incrementing a counter and peeling off rows (or columns
> of X and y) of a particular window size, and following that up with some
> matrix multiplication in a loop.   The psuedo-code for that Rcpp
> practically writes itself and you might want a wrapper of something like:
> rolling_lm (y=y, x=x, width=4).
>
> My question is this: has any of the thousands of R packages out there
> published anything like that.  Rolling window multiple regressions that
> stay in C/C++ until the rolling window completes?  No sense and writing it
> if it exist.
>
>
> Thanks,
> Jeremiah
>
> Standard (slow) advice for "rolling window regression" follows:
>
>
> set.seed(1)
> z <- zoo(matrix(rnorm(10), ncol = 2))
> colnames(z) <- c("y", "x")
>
> ## rolling regression of width 4
> rollapply(z, width = 4,
>   function(x) coef(lm(y ~ x, data = as.data.frame(x))),
>   by.column = FALSE, align = "right")
>
> ## result is identical to
> coef(lm(y ~ x, data = z[1:4,]))
> coef(lm(y ~ x, data = z[2:5,]))
>
> 	[[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list -- To UNSUBSCRIBE and more, see
> 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.
>