Forecasting semi-annual data #430
Description
When forecasting with ARIMA for semiannual data, model function works, even fitting seasonal ARIMA model with correct seasonal period. However piping the output through to forecast throws a bug due to periods having noninteger values. There may be a workaround, however it would be nice if it worked by default. Minimum working example below:
# Load packages
library(tsibble)
library(fable)
#Set date variable as yearquarter (but only observed every 2 quarters)
dates<-yearquarter(seq.Date('1970-02-15','2019-08-15',length.out = 100))
#Randomly generate data in a way that has seasonality
y=(-1)^(1:100)+0.01*rnorm(100)
#Put into a tsibble
data<-tsibble(dates,y)
#Model (this part works as expected)
data%>%
model(arima=ARIMA(y))->mod
mod
#Forecast(this next line throws an error)
forecast(mod)