What is Mean Absolute Percentage Error (MAPE) in Forecasting?

Definition: MAPE (Mean Absolute Percentage Error) measures forecast accuracy as a percentage of actual values: MAPE = (1/n) × Σ|actual - forecast| / |actual| × 100%. A MAPE of 10% means forecasts are off by 10% on average. The percentage scale makes MAPE interpretable across different magnitude series.

Why MAPE is Popular

MAPE provides intuitive interpretation: "forecasts are 15% off on average" is meaningful to business stakeholders without statistical background. It normalizes error by actual values, allowing comparison across series with different scales (units sold vs revenue dollars). MAPE is the most commonly reported forecast accuracy metric in industry.

MAPE Limitations

MAPE has critical flaws. Division by actual causes problems: when actual is zero, MAPE is undefined; when actual is small, MAPE explodes. A forecast of 5 vs actual of 1 produces 400% error, while forecast of 105 vs actual of 100 produces only 5%—same absolute error, vastly different MAPE. MAPE penalizes over-forecasts more than under-forecasts of equal magnitude.

Warning: MAPE is asymmetric: over-forecast of 100% (forecast 200, actual 100) contributes 100%, but under-forecast of 100% (forecast 0, actual 100) contributes 100%—yet these feel very different. Use symmetric MAPE (sMAPE) for balanced treatment.

When to Use MAPE

Use MAPE when: values are always positive and never near zero, stakeholders need intuitive percentage interpretation, comparing accuracy across different-scale series. Avoid MAPE when: data contains zeros or near-zeros, symmetric treatment of errors matters, or you need mathematically sound optimization (MAPE is not convex).

Industry Benchmarks

Typical MAPE ranges: 5-10% is excellent for stable demand, 15-25% is acceptable for variable demand, 30%+ indicates poor forecasting or inherently unpredictable series. Compare against naive forecasts (use last period value) to assess model value-add.