Time series and forecasting · insight-0209
Compute WAPE at the level where actuals cannot sum to zero
- Evidence type
- practitioner heuristic
- Difficulty
- easy
- Roles
- Data analyst, Data scientist
- Industries
- retail, aviation, logistics
Summary: situation, action, reason, boundary
Situation
WAPE is reported for every store-SKU-week cell and many cells are blank or infinite.
Action
Compute WAPE at the aggregation the business manages (category-week, route-month), report undefined slices as a count with their absolute error, and never average cell-level WAPE values.
Reason
WAPE is a volume-weighted ratio; it is well behaved only when the denominator is large and positive.
Boundary
Retail and supply-chain reporting, airline route dashboards, any sparse panel.
Symptom or risk
WAPE divides total absolute error by total actuals, so any slice whose actuals sum to zero is undefined, and tiny slices produce extreme values that alarm stakeholders.
Exceptions
If every slice has substantial volume, cell-level WAPE is fine.
Trade-offs
Aggregating hides poor accuracy on small items; pair it with an item-level absolute error table.
Pitfalls
Treating 'WAPE is undefined' as 'error is zero' in a dashboard.
Example
Hypothetical: a route-week with zero actual passengers due to a cancelled schedule and two forecast passengers has undefined WAPE but an absolute error of 2.