So I have a lot of time series with a hierarchical structure, and need to produce forecast for each base series and its aggregates by the hierarchical structure.
I have decided to produce forecast using the R hts package by Rob J Hyndman. I have produced reasonable point forecast for each series using the auto.arima function, and the forecast.gts function as in hts-viggnete page9. But prediction intervals are not yet implented directly in the package so I have followed the formulations in ref1 to get the variance matrix sigma.
But now Im lost in how to produce prediction intervals, should I use the diagonal of this matrix as the variance of each series to produce naive intervals? as in:
c(-1,1)*diag(sigma)+aggts(myhts.forecast)
Is using the combinef() function on the intervals given by the auto.arima a correct way to produce conficence intervals?
thank you.
