library smooth


Fortran must die
tsdata=ts(mydata$Spend,start=c(2014,12),frequency=12,end=(c(2020,11))) # a training data set to choose the best model
esmtr<-es(tsdatatr, model = "ZZZ")
esmtr # it will show the chosen model
Unfortunately I can not provide the data, I am not sure if this is permitted. This is the R exponential smoothing module. The ES function is supposed to chose among 30 different models. But without exception in the six months I have run it, it chooses the same exponential smoothing model. That is very different than my experiences with exponential smoothing in SAS where commonly a different model is better at different times.

I was curious if anyone knew how reasonable it is as the data changes over time for the same model to always be chosen this way.

Of course since no one does time series here probably no one uses this... :p

Well on the bright side I am slowly learning R.


Fortran must die
I am having the same problem with autoarima. The best model never changes. That does not seem likely to me although it is not impossible.


Less is more. Stay pure. Stay poor.
I would wonder what criteria it is using for selection and how sensitive it is to small changes.

If you think there is an issue, insert some outlier or synthetic data into your dataset to prove that it can select a different model.

Also, do you need to set seeds with these functions, just curious.