Weighted mean absolute percentage error (WAPE, also written wMAPE) measures total forecast error relative to total actual demand. Calculate it as the sum of absolute errors divided by the sum of actual values, then express the result as a percentage. Weighting by actual volume makes the metric useful when comparing forecasts across products, periods, or business units with different scales. section 12.1.2 section 5.3.1
Use WAPE carefully when interpreting forecast quality:
- Pair WAPE with bias. Two models can have identical WAPE while one consistently underpredicts and the other overpredicts. section 12.1.2
- Tie the metric to the business decision rather than treating it as the decision itself. Ticket forecasts affect overbooking, unsold capacity, pricing, and revenue. section 12.1.2
- For low-volume or intermittent count data, compare WAPE with alternative metrics and consider the full predictive distribution rather than point forecasts alone. section 5.3.1
- Do not substitute ordinary MAPE without checking for zeros: MAPE is undefined when any future realization is zero, while symmetric MAPE can distort zero-actual periods. section 5.3.1
- Track WAPE as an offline metric alongside operational outcomes such as revenue, profit, sales volume, and model update frequency. section 12.4.1
Airline ticket forecasts
Across four days, actual ticket sales are [120, 90, 110, 80]. Model A predicts [90, 90, 90, 90], while Model B predicts [110, 110, 110, 110]. Each model has WAPE of 15%, but Model A has −10% bias and Model B has +10% bias. Using the day’s ticket prices and the lower of forecast and realized sales, Model B produces $83,900 in revenue versus $75,300 for Model A. Equal WAPE therefore does not imply equal business value. section 12.1.2
