Should mape (tsmape) be calculated using datapoints where a forecast is not approved?
Hello Experts,
The RPAS Mape (tsmape) function in RDF compares Approved Forecasts against Sales (where sales are not zero). By itself this is debateable (a known mape "issue") because we can have an approved forecast of some units and have had zero (real) sales. But, considering the mape formula, it has to be like this or we would get a Division-by-Zero.
Things get messier when:
1) We include history where there are no approved forecasts
2) There are some forecasts not approved in the historical horizon.
In the first case it will be worse in the beginning of an implementation but it is, to an extent, manageable by the end-user (they must know what they're "asking", for instance, in a Forecast Scorecard).