What is mean absolute percent error?
Mean absolute percent error (MAPE) is the average size of a model’s errors expressed as a percentage of the actual values. For each case, divide the absolute difference between forecast and actual by the actual value, then average the results. A miss of 10 units on actual sales of 100 is a 10 percent error, and so is a miss of 1,000 units on sales of 10,000. That independence from scale lets a planner compare accuracy across stores and product lines with one figure, which is why MAPE is common in demand forecasting.
Why mean absolute percent error matters
MAPE is easy to communicate but has three known weaknesses. It is undefined when the actual value is zero. It explodes when actuals are small: a forecast of 3 units against actual sales of 1 is a 200 percent error. It also favors forecasts that run low, because an under-forecast can never miss by more than 100 percent while an over-forecast can miss by any amount.
A regional grocery chain saw a store-by-item MAPE of 48 percent for its new replenishment forecast and nearly rejected it. Most of the error came from slow-moving items that sell one or two units a week. Weighted absolute percent error (WAPE), which divides total absolute error by total sales so that high-volume items count more, put the same forecasts at 18 percent. The chain now uses WAPE for replenishment decisions and reports mean absolute error in units for its slowest sellers.