I have to forecast some monthly timeseries 1.5 years into the future and then sum them up to get 3 semi-annual counts.

The forecasts give forecast intervals, but how do I get from the 6 monthly forcast intervals to one semi-annual one?

Is there some clear formula for that or a good estimate?

Or would it be better to sum the data into semi-annual counts first and then forecast those 3 periods into the future?

I'm using R and the forecast package.

Thanks in advance.

EDIT: My question has been marked as a duplicate of another question that is not actually similar.

First of all, the other questions deals with annual counts and I'm interested in semi-annual counts. It is not quite clear form the answer to the other question how I can get from the annual counts to semi-annual counts since the mechanism behind the trick mentioned isn't really explained.

Second, I'd like to know the theory behind it, not just a trick that works in R (we're using other time series platforms as well, so the R trick is of absolutely no help in those cases, even if it solves my problem with this particular aggregated forecast).

The Otexts book "Forecasting: Principles and Practice 2" seems to have a planned chapter about this problem, but there's only a headline and no text (unless it's my browser that's malfunctioning):


  • $\begingroup$ A question like yours was answered in this topic: stats.stackexchange.com/questions/59065/… The author created a trick that allowed to forecast total counts. $\endgroup$ – Alexey Burnakov Oct 5 '17 at 13:50
  • $\begingroup$ Yes, I think I can use that answer, only I'm interested in 3 semi-annual totals of the forecasts. Would I get that by using the following code? y <- filter(x,rep(1,6), sides=1) fit <- auto.arima(y) forecast(fit,h=3) $\endgroup$ – SiKiHe Oct 6 '17 at 6:42