Calculating demand forecast accuracy: Difference between revisions

Revision as of 19:38, 13 October 2015

Revision as of 15:08, 30 November 2015

Line 11:

A simpler and more elegant method to calculate MAPE across all the products forecasted is to divide the sum of the absolute deviations by the total sales of all products.

This calculation <math>\sum{(|A - F|)}\over\sum{A}</math>, where <math>A</math> is the actual value and <math>F</math> the forecast, is also known as WAPEMAPE, WeightedMean Absolute Percent Error. Another interesting option is the weighted

<math>MAPE (WAPE) = \frac{\sum(w\cdot|A-F|)}{\sum(w\cdot A)}</math>. The advantage of this measure is that could weight errors, so you can define how to weight for your relevant business, ex gross profit or ABC. The only problem is that for seasonal products you will create an undefined result when sales = 0 and that is not symmetrical, that means that you can be much more inaccurate if sales are higher than if they are lower than the forecast. So sMAPE is also used to correct this, it is known as symmetric Mean Absolute Percentage Error.

Last but not least, for intermittent demand patterns none of the above are really useful. So you can consider MASE (Mean Absolute Scaled Error) as a good KPI to use in those situations, the problem is that is not as intuitive as the ones mentioned before. You can find an interesting discussion here: http://datascienceassn.org/sites/default/files/Another%20Look%20at%20Measures%20of%20Forecast%20Accuracy.pdf