Is there a general and completely automated solution to modelling and forecasting time series data?

I think this question is extremely important.

If it is not possible, please provide an explanation why it isn't.

I'm interested to know whether anyone has completely automated the process of model building and forecasting in a way that works and allows a machine or programme do so without intervention. I'm not so interested in whether a human can improve on what a machine can do. I'm more interested in finding out whether a computer / programme can do this kind of work for us in a way that is "reasonable" - I don't have any metric in mind for what "reasonable" means.

  • 3
    $\begingroup$ autobox.com/cms $\endgroup$
    – Sycorax
    Aug 2, 2016 at 0:57
  • 2
    $\begingroup$ What does "general" mean in this context? What (goodness) properties does the output have to satisfy in order to qualify? What if a human can apply expertise and do better for some time series than the completely automated solution? Etc. $\endgroup$ Aug 2, 2016 at 0:59
  • 2
    $\begingroup$ I'm voting to re-open this question because I think the generality is actually what makes it interesting. There's much that could be said specifically about the performance of algorithms over the set of all possible problems, the relevance of priors, etc. In general, I think the answer is: no, such a method is not possible $\endgroup$
    – user20160
    Aug 2, 2016 at 5:18
  • 3
    $\begingroup$ I think this is too broad, but if there are people who want to try to answer this, I'm inclined to let them. $\endgroup$ Aug 2, 2016 at 11:09
  • 4
    $\begingroup$ Various "No Free Lunch" theorems may be of interest in this regard. $\endgroup$
    – whuber
    Aug 2, 2016 at 15:36


Your Answer

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