Extracting Standard Deviation from Combination Forecast in Fable

I am using fable to create numerous forecasts and measure the accuracy of their point forecasts and distributions against test data. I've noticed the following message from the fabletools package:


I am wondering if the combination of non-normal distributions is inherently wrong, or if the package just does not support doing so yet.

Referred here by Forecasting: Principles and Practice, by Rob J Hyndman and George Athanasopoulos

1 Like

There is nothing wrong with combining forecasts from different distributions, but it is more challenging to implement. We do plan to eventually support a wide range of distributional combinations. It is always possible to do it by simulating from each model and combining the samples to get an empirical mixture distribution. However, even that must be done carefully because the errors that drive each model will be correlated, so independent simulations will not give the right distribution.


This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.

If you have a query related to it or one of the replies, start a new topic and refer back with a link.