[Gretl-users] Re: Function package fcModels (Sven Schreiber)
On Thu, 24 Oct 2019, George Matysiak wrote:
Sven,
Yes, I should have provided more details.
Based on two different data sets, both the AutoSES and HoltDES provide
different RMSE figures from those obtained in R's forecast package. The
fcModel figs are consistently lower, and the differences seem to be roughly
of the order magnitude that the fcModel RMSE squared is approx. equal to
the forecast package RMSE.
Ah, it appears that Yi-Nung, the author of fcModels, didn't notice
that back in 2016 we changed the behaviour of the built-in gretl
function fcstats(), as stated at
http://gretl.sourceforge.net/Backward.html#v2016d
to provide the RMSE rather than MSE as the second element in the
vector returned. So George is right and this package needs to be
updated.
Allin