Mean Absolute Percentage Error (MAPE) Has Served Its Duty and Should Now Retire
BlueYonder
AUGUST 4, 2022
This is not dry theory: We simulate a supermarket that relies on a MAPE-optimizing forecast value fed into replenishment. A model with a nice 3% error on the US GDP and an unacceptable 200% error on the Tuvalu GDP would appear to be “better” than a model with 4% error on the US GDP and 10% error on Tuvalu GDP in absolute US$ terms.
Let's personalize your content