1
$\begingroup$

The question title doesn't help, but i'll try to explain it:

I have historical data (weekly points) from 2001/01/01 to yesterday and my variable has a strong seasonal pattern of 12 months (weather data).

I'm using the stlf() function (from the R 'forecast' package) to make weekly forecasts - and getting very good results by the way.

The problem is that for some models i don't have the 2 last years of data, so for these models my data goes from 2001/01/01 to 2010/12/31. What can i do to bring these forecasts 'to the future'? In other words, i need to extrapolate these forecasts to the current week.

What i have: data from 2001/01/01 - 2010/12/31 What i need: make forecasts for the next weeks.

To add confusion, i do have some points from the last month, but i don't know how to take advantage of this.

Thanks for any help!

|cite|improve this question
$\endgroup$
  • $\begingroup$ Why not set h = 365 + 292 + 7? $\endgroup$ – Jim Oct 18 '12 at 23:43
  • $\begingroup$ You mean h = 2*52 + weeks_ahead. I'm avoiding doing that because i do have access to some 'present tense' points, so i need to consider those. $\endgroup$ – Fernando Oct 19 '12 at 12:13
1
$\begingroup$

Try the h=2*52 and see how that works vs the 'present tense" points you do have. See how far (or not far) off the error is? Might even try taking h=(however many weeks till your present data) then add in the present data then do h=(however long till the next present data point) so on and so forth till you have added all of the present data. Then compare that vs just the h=2*52+weeks_ahead to find the better of the two models?

|cite|improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.