Mean Absolute Percentage Error

Forecast error as a percentage, and where it misleads.

Also known as: MAPE, mean absolute percentage error, percentage error

MAPE expresses forecast error as a percentage: the average of |actual − forecast| / |actual|, times 100. “Off by 8% on average” is immediately meaningful to stakeholders in a way that raw error units never are, which is why MAPE dominates business forecasting conversations.

actual 100, forecast 110  →  |100-110|/100 = 10%
MAPE = average of those percentages (e.g. 8% across the period)

Its convenience hides two asymmetries. Dividing by the actual punishes errors unevenly: over-forecasts are capped at 100% per period only in the extreme, while under-forecasts are bounded — and when actuals are near zero, tiny absolute misses become enormous percentages. MAPE also treats a 10% miss on a huge month the same as on a tiny one, even though the money differs wildly.

The classic mistakes:

MAPE on intermittent or near-zero series. Spare-parts demand, new products, thin days — percentages explode and the metric becomes noise. Use a scaled or absolute measure there instead.
Averaging percentages across different scales. A 5% miss on the flagship product dwarfs a 20% miss on a minor one. Weight by volume, or report RMSE alongside.
Optimising MAPE blindly. Minimising percentage error favours under-forecasting (a forecast of zero can never exceed 100% error on one side). Check the bias, not just the average.
No baseline. “MAPE 8%” means nothing without the naive forecast’s MAPE beside it.

When to use it: communicating accuracy on stable, non-zero series where percentages read naturally. Pair it with a scale-aware measure and always show the baseline — see forecast accuracy.