I am not sure what the appropriate title for my question is, so I am open to suggestion/correction.

Now, the question:

  • I observed a daily series $X_t, t=1,...,T$ also series $Y_t, t=1,...,T$.
  • My aim is to use $X$ to predict $Y$ in the future. Specifically, I will observe $X_t, t=T+k,...,T+K$ where $k$ can be large. I will not observe any more $Y_t$. But I would like to estimate it.
  • So I fitted a regression with ARIMA error using $(X,Y)_t,t=1,...,T$. (this is done using the forecast package.)
  • The question is, (a) since new data won't come until some time later ($k$ being large), is any forecast still valid? (b) Even if it is valid, how do I forecast anyway? Do I start from $t=T+1$ and work my away towards $t=T+k$? This seems inefficient because I don't care about $t=T+1,...,T+k$. Furthermore, this isn't really possible because $X$ is not observed during that gap period.(I suppose one can argue you can forecast $X$ using some other model to fill the gap, but this still doesn't solve the problem that it is forecasting things I don't care about, let alone part (a) of my question)

So, what should I do?

P.S. Extra info. We can assume $T$ is large enough to cover any periods multiple times.


1 Answer 1


Of course you can, you're just not going to have the confidence you would if you had those 'in between' values. If I asked you to predict the next 100 values, 101-200, how is that any different from asking you to predict 121-200 when you don't have 101-120?

(Unfortunately, error bounds get big pretty fast on arima predictions.)

  • $\begingroup$ He has X, pretty important info. $\endgroup$ Feb 26, 2014 at 0:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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