Anyways:

I'm making some estimates for 12 years using a regression model. The model gives me monthly predictions, with both confidence and prediction intervals.

Now I wan't to accumulate these estimates into a yearly estimate. To estimate the point prediction (I think thats the right word?) for the year I just sum the months. The difficulties occur when I wan't to measure the uncertainty by making confidence and prediction intervals.

I have different standard deviations for each month, so I think I'm able to use the stds to use in both the calculation of the prediction and the confidence interval - but how many degrees of freedom should I use when calculating the t-value?

The intervals calculated for each months were made based on ca 5 years, so I had about 60 observations. Therefore I used ~60 degrees of freedom for these calculations. Should I use the same number when calculating the t-value for the yearly intervals?

Don't know if this is an easy question, but hopefully someone knows the answer.