Hi all

Is anyone familiar with a user-written package or a well documented approach to calculate a reliable confidence interval for a sum of predictions?
Imagine a time series (say monthly) and predictions for a few months. The predict command will provide out of sample predictions for these months and SEs.
Is there a reliable way to combine them into an estimate? Adding up the SEs from predict (assuming zero covariances)? I've tried bootstrapping but because it is a time series it gets messsy and again not supported as standard by Stata commnands (Block bootstrap etc). I tried margins and then mutliplying the estimate by the number of months, and doing the same for the SEs, as I saw in another thread here, but I am not convinced this is entirely appropriate. I contacted Stata but there was no official suggestion - I was advised to use simulate. But surely there must be a straightforward parametric way to obtain the confidence interval for the quantity of interest!

Best

Evan