confidence intervals for linear regression

#1
Hi, newbie here, so please forgive the bad terminology. I started with ten years of historical data. I used Excel to give me a linear trendline and equation. I used that equation to get the forecasted data points for the following ten years. So far so good.

Now I want to get the data points for a 95% confidence interval on this forecast. For the one-step forecast, I know that I just need to go +-2 SDs to get this interval. However, how many SDs should I use for steps 2-10? Is there some table somewhere that will tell me how many SDs to use for a given step forecast and confidence interval?

I know that I could just use 2 SDs for each step of the forecast, but then the interval does not diverge, it just runs parallel to the point estimates. Since the cone of uncertainty should expand over time, the number of SDs should get larger, right?
 
Last edited:

Dason

Ambassador to the humans
#2
Are you sure you want a confidence interval or are you looking for a credible interval? Either way just using +- 2 SDs isn't the way to go.
 
#3
I did some quick research on what a credible interval is, and from what I can tell, yes that is actually what I'm looking for.

Either way though, I'm trying to put this all together for a non-statistics crowd, so I don't think they care about confidence vs. credible. Frankly it probably doesn't have to be terribly accurate. It just needs to make sense. So are there any rules of thumb out there for this kind of estimating?
 

Dason

Ambassador to the humans
#4
I actually meant to ask about prediction intervals. I'm sorry for the confusion. Credible intervals are nicer (I'm more of a Bayesian myself) but what I really meant to ask was whether you want to put an interval that will contain the observed value with 95% certainty or one that will contain the true value of the regression line at that point with 95% certainty? My guess is that you want the former.