Some forecast methods such as forecast.ETS() use the analytical forecast mean for the point forecast, and the simulated paths for the uncertainty. Should we create a new distribution modifier for distributions with known mean, or change the forecast behaviour to use sample means for point forecasts?
The analytical means are probably a little more accurate, but I'm in favour of simplicity. And since you have to do the simulation anyway, I think we could just use the sample means.
cc @robjhyndman
Some forecast methods such as
forecast.ETS()
use the analytical forecast mean for the point forecast, and the simulated paths for the uncertainty. Should we create a new distribution modifier for distributions with known mean, or change the forecast behaviour to use sample means for point forecasts?