In this blog post, @rob-hyndman says:

If you only want point forecasts, that is the best approach available in the forecast package. It is also better than any of the commercial software (at least as far as they have been prepared to subject their algorithms to independent testing).

If I follow this same approach and take the average of the prediction intervals produced by both models, will they be reliable?

Update: I've come across with this nice blog post that shows how to create prediction intervals from hybrid models.

  • $\begingroup$ How do you "average" two prediction intervals such that you do not affect its nominal accuracy (e.g., if you make the interval wider, then of course I'd expect it to be more reliable, especially if it contains both of the smaller intervals.) $\endgroup$
    – user75138
    Jul 6, 2016 at 4:20

1 Answer 1


Combining prediction intervals is not straightforward. It is conceptually easier to work directly with the underlying predictive distributions and combine them, e.g., through mixtures. This also allows you to weight different forecasted densities differently.

Literature includes Wallis (2005, Oxford Bulletin of Economics and Statistics) and Hall & Mitchell (2007, International Journal of Forecasting).

  • $\begingroup$ Thanks! By the way, I've been following your responses to Time Series questions. Could you take a look at another thread of mine that has not gotten replies? $\endgroup$
    – iatowks
    Jul 6, 2016 at 21:58

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.