Closed pstamolampros closed 1 year ago
By subsetting the data in this way, it loses its time series attributes including the frequency, so ets()
does not know it is seasonal data. The correct way to subset a time series without losing information is to use either subset(AP, end=144)
, or window(AP, end=c(1960,12))
.
Thank you for your reply and for sharing your great work
I do have another question about the qualified model from the automatic ets(). From the output below I understand that the MMM model is better based on the information criteria but the "ZZZ" selects the MAM. What Am I missing?
`> data(AirPassengers)
AP <- AirPassengers AP.train <- window(AP, end=c(1958,3)) AP.test <- window(AP, start=c(1958,4)) ets(AP.train , model="ZZZ") ETS(M,Ad,M)
Call: ets(y = AP.train, model = "ZZZ")
Smoothing parameters: alpha = 0.75 beta = 0.0233 gamma = 1e-04 phi = 0.98
Initial states: l = 120.3561 b = 1.7593 s = 0.9013 0.799 0.9181 1.06 1.2002 1.2147 1.1032 0.9775 0.9877 1.0297 0.8985 0.9101
sigma: 0.0371
AIC AICc BIC
1004.257 1011.692 1053.028
ets(AP.train , model="MMM") ETS(M,Md,M)
Call: ets(y = AP.train, model = "MMM")
Smoothing parameters: alpha = 0.7209 beta = 0.0134 gamma = 1e-04 phi = 0.9789
Initial states: l = 120.3501 b = 1.0134 s = 0.903 0.7993 0.919 1.0591 1.2002 1.2139 1.1036 0.9754 0.9858 1.0296 0.8998 0.9113
sigma: 0.037
AIC AICc BIC
1003.631 1011.065 1052.402 `
The automatic search does not consider multiplicative trend models. See the allow.multiplicative.trend
argument.
install.packages("forecast") library("forecast")
data(AirPassengers) AP <- AirPassengers
This will return an (M,Ad,M) model fit<- ets(AP)
same observations but will return (M,N,N) fit2<- ets(AP[1:144])
this return an error no seasonal data fit3 <- ets(AP[1:144], model="MAM")