[R] alpha in Holt-Winters method

Thomas Lumley tlumley at u.washington.edu
Thu Aug 31 21:41:10 CEST 2006


On Thu, 31 Aug 2006, MARK LEEDS wrote:

> you would have to take your series, y_1, ... y_t and run it over various
> values of alpha ( say zero to 1 insteps of .01 ) and see which one gives say
> the least MSE.
> but that will be the optimal alpha for the window you are looking at. it
> doesn't mean it will be the optimal apha for the data you deal with
> at y_t+1,..... y_t+n.
>
> it's been a while, but , in holt winters, i think there are 3 smoothing
> constants so i would think you would want to optimize over all 3 of them ?
>
> someone else can give the following  in more detail there is probably a way
> to use nlmin or one of those optimization functions to find the optimal
> smoothing constants but i
> don't think it's worth the hassle because they are in a small enough range
> that you can do it by brute force as above.
>

Or you could read the documentation for HoltWinters

      The function tries to find the optimal values of alpha and/or beta
      and/or gamma by minimizing the squared one-step prediction error
      if they are omitted.

If you want to know how the optimization is done, looking at the code will 
tell you.


 	-thomas



More information about the R-help mailing list