Comments on: What about all those zeroes? Measuring performance of models on intermittent demand https://openforecast.org/2020/01/13/what-about-all-those-zeroes-measuring-performance-of-models-on-intermittent-demand/ How to look into the future Fri, 18 Sep 2020 21:00:42 +0000 hourly 1 https://wordpress.org/?v=6.9.4 By: Ivan Svetunkov https://openforecast.org/2020/01/13/what-about-all-those-zeroes-measuring-performance-of-models-on-intermittent-demand/#comment-134 Sun, 16 Feb 2020 11:52:46 +0000 https://openforecast.org/?p=2278#comment-134 Dear Andrey Davydenko,

Yes, trimming can be used indeed. But it is just a solution to the problem appearing in the measure naturally. The point in this post is not about how to fix relative measures, but how to avoid the problem all together: use averages instead of Naive.

Kind regards,
Ivan Svetunkov

]]> By: Turbo Forecasting https://openforecast.org/2020/01/13/what-about-all-those-zeroes-measuring-performance-of-models-on-intermittent-demand/#comment-131 Sun, 16 Feb 2020 06:56:39 +0000 https://openforecast.org/?p=2278#comment-131 If the frequency of obtaining zero MAEs is too high, you can use this approach (page 23):
https://www.researchgate.net/publication/282136084_Measuring_Forecasting_Accuracy_Problems_and_Recommendations_by_the_Example_of_SKU-Level_Judgmental_Adjustments

]]>