[R] HoltWinters in R 2.14.0

[Prof Brian Ripley]

On Fri, 4 Nov 2011, R. Michael Weylandt wrote:

> I believe there were some changes to Holt-Winters, specifically in re
> optimization that probably lead to your problem, but you'll have to
> provide more details. See the NEWS file for citations about the
> change. If you put example code/data others may be able to help you --
> I haven't updated yet so I can't be of much help.
>
> Michael
>
>
> On Fri, Nov 4, 2011 at 2:55 PM, TimothyDalbey <tmdalbey at gmail.com> wrote:
>> Hey All,
>>
>> First time on these forums.  Thanks in advance.
>>
>> Soooo...  I have a process that was functioning well before the 2.14 update.
>> Now the HoltWinters function is throwing an error whereby I get the
>> following:
>>
>> Error in HoltWinters(sales.ts) : optimization failure

Most likely it was incorrect before.  You cannot assume that it was 
actually 'functioning well': all the cases where we have seen this 
message it was giving incorrect answers before and not detecting them. 
And in all those cases the model was a bad fit and using starting 
values for the optimization helped.

>> I've been looking around to determine why this happens (see if I can test
>> the data beforehand) but I haven't come across anything.
>>
>> Any help appreciated!

-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595