In forecasting (ets) annual data, I notice that when I use the full data set of 10 years, the prediction intervals are much wider than when using an abridged version of the data set (5 years).

I assume this occurs because more variability is being modelled when the full range of data is provided.

Is there any way to determine whether it is more appropriate to use the full data set (10 years) or abridged data set (5 years)?

  • 3
    $\begingroup$ More data means more information. That information may, however, not always give you the answer you want to hear. $\endgroup$ – TrynnaDoStat Aug 15 '19 at 20:55
  • $\begingroup$ Agreed. Is there a way to show that the predictions using the larger data set are more valid than those based on the smaller data set? $\endgroup$ – SaHa Aug 15 '19 at 21:01
  • 3
    $\begingroup$ Is the future going to be more like the last five years or like the last ten years? The answer to that question determines which data are more relevant. Unfortunately, we cannot provide an answer because we just don't know. Sometimes you have to make a choice and defend it. $\endgroup$ – whuber Aug 15 '19 at 21:20
  • 3
    $\begingroup$ I agree with @whuber's comment. This is not a question of technicalities of statistical modeling. It's a question of validity -- external validity -- the representativeness of certain time periods vis-a-vis the one for which you are trying to predict. Non-statisticians often see through the models stat. people put together because the former are not so captivated by the intricacies of modeling that they fail to see basic validity issues. $\endgroup$ – rolando2 Aug 16 '19 at 0:51
  • 1
    $\begingroup$ Your first question in the above comment is already answered by whuber's comment. As for your second, you can show this by trying out rolling forecasts with models fit to the different data ranges, if you have sufficient data to do this. $\endgroup$ – mkt - Reinstate Monica Aug 16 '19 at 9:44

Your Answer

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

Browse other questions tagged or ask your own question.