Accuracy & evaluation

WAPE

Also called wMAPE, weighted absolute percentage error

WAPE is the sum of absolute errors divided by the sum of actuals, usually shown as a percent. It weights misses by volume, so a tiny SKU cannot dominate the score the way MAPE allows.

Write it once: Σ|ŷ − y| / Σy. That is the whole trick. Zeros in actuals do not explode the denominator unless the whole window is zero — in which case you have a different problem.

WAPE is not sacred. It hides bias (you can over and under and look fine). Pair it with a signed bias metric. And never compare WAPE across grains without saying so.

When actuals can be negative (some cash series), WAPE gets weird. Use MAE or a custom scale. Metrics are not a personality test; they are a cost function in disguise.