[Gretl-users] Re: Error using Holt Function
Am 28.08.2026 um 14:10 schrieb Ramki S:
I am trying to replicate the text book example given in fpp2 book
using fcModels package. https://otexts.com/fpp2/holt.html. I added 5 observations and
reduced the windows size to 1990 to 2016. If I don't reduce the sample, there is no
error. Any help?
https://www.dropbox.com/scl/fi/5ngkokbozi6cbach92o63/fpp2_ausair.xlsx?rlk...
? List02 = HoltDES(Y, 1, null, 1, 0.3, 0.1, 0)
>>>> ***sm>>>>>>> EOS=47; EOB=47; predh=5;
error in new starting obs
*** error in function HoltDES, line 54
> smpl firstobs(y_) -predh
First, I can confirm this, which is a (another) bug in the fcModels
contributed package.
Second, from the function signature of HoltDES, I believe that if you
added 5 obs at the end of your dataset, then you might want to call the
function with a value 5 as the last parameter. (Not 100% sure, since the
function isn't very well documented.)
Third, but the hard error is with the starting obs. I think the package
uses the firstobs() hansl function incorrectly - the hansl function help
clearly states that this only takes subsampling into account with the
optional second arg set to 1/TRUE, which doesn't happen here. So I
believe the package is broken with such subsampling.
Fourth, as a workaround you could save your subsample to a new dataset
file, starting at 1990, and work with that. (But I haven't tested it
myself.)
cheers
sven