WMAPE: Weighted Forecast Accuracy for Uneven Portfolios

WMAPE, the weighted mean absolute percentage error, is the forecast accuracy measure that most consumer product companies should use as their headline number, because it fixes the biggest flaw of plain MAPE: the tendency of tiny items to dominate the result. The idea is simple. Instead of averaging percentage errors item by item, you add up all the absolute errors and divide by the total actual demand. A 50-unit miss on a 1,000-unit item then counts for exactly what it is worth, and a 10-unit miss on a 10-unit item no longer swamps the report. This guide gives the formula, an Excel version, a worked example with checked arithmetic that compares WMAPE with MAPE and with a cost-weighted variant, explains how to read the number, how to set targets by segment and how to avoid common traps. It is part of demand planning and inventory optimization, follows MAPE and forecast accuracy metrics, and leads to forecast bias and forecast value added. The same logic applies to financial forecasts, as in measuring financial forecast accuracy, and accuracy affects the cash described in cash flow management.

The formula

WMAPE = sum of |Forecast - Actual| across items and periods, divided by the sum of Actual across the same items and periods.

It is also called WAPE (weighted absolute percentage error) or, in some textbooks, the MAD/mean ratio. All are the same calculation. Many teams report accuracy as 100 percent minus WMAPE, though this convention hides the fact that WMAPE can exceed 100 percent for very poor forecasts, so be clear about which you are quoting.

In Excel, with actuals in column B and forecasts in column C for rows 2 to 100, the formula is:

=SUMPRODUCT(ABS(C2:C100-B2:B100))/SUM(B2:B100)

Format the result as a percentage. If you have a table with a row per item per month, the same formula measures accuracy across all of them, and a filter or a pivot table breaks it down by category, customer or month.

A worked example

Five items, one month (illustrative):

WMAPE: 180 / 1,760 = 10.2 percent. Check: 50 + 30 + 30 + 60 + 10 = 180, and 500 + 200 + 50 + 1,000 + 10 = 1,760.

MAPE for comparison: item percentages are A 10 percent, B 15 percent, C 60 percent, D 6 percent and E 100 percent. The average is (10 + 15 + 60 + 6 + 100) / 5 = 191 / 5 = 38.2 percent. The slow item E, which sold 10 units, drives the result, even though an error of 10 units is tiny in volume terms.

Bias: (1,820 - 1,760) / 1,760 = +3.4 percent. The forecast was slightly high overall.

The gap between 10.2 percent and 38.2 percent is not a rounding difference. It is a different message. WMAPE says the forecast is decent by volume. MAPE says it is poor, because a few small items were badly missed. Which is right depends on what you care about, and for most inventory decisions volume or value is what matters.

The cost-weighted variant

Units are not all alike. A miss of 10 units on a $50 item matters more than a miss of 60 units on a $2 item. A cost-weighted WMAPE multiplies each item's error and actual by its unit cost (or margin) before summing. Absolute error in dollars: A 50 × 4 = 200; B 30 × 10 = 300; C 30 × 20 = 600; D 60 × 2 = 120; E 10 × 50 = 500. The total is $1,720. Actual demand in dollars: A 500 × 4 = 2,000; B 200 × 10 = 2,000; C 50 × 20 = 1,000; D 1,000 × 2 = 2,000; E 10 × 50 = 500, a total of $7,500. The cost-weighted WMAPE is 1,720 / 7,500 = 22.9 percent.

This is more than double the unit-weighted figure, because the large unit-volume item D, which is cheap and accurate, dominated the unit-based number, and the expensive items C and E, which were poorly forecast, count for more in dollars. If your concern is inventory investment and cash, the dollar-weighted version is the more honest measure. Many companies track both: units for operations and dollars for finance.

How to read WMAPE

WMAPE answers the question "What share of demand did the forecast miss, in total, regardless of direction?" A value of 10 percent means that the absolute errors added up to a tenth of demand. It does not say whether the misses were high or low, which is why bias goes beside it. It also depends strongly on the level of aggregation: item-by-location-by-week WMAPE will be much higher than category-by-month WMAPE, because errors offset when you aggregate. Always state the level and lag when you quote a figure.

Setting targets by segment

A single target for all items is misleading, because accuracy varies naturally with the demand pattern. Stable, high-volume items can be forecast well. Promoted, seasonal or new items cannot. Set targets by segment, based on your own history. A reasonable approach is to measure WMAPE for each segment over the last 12 months, take the median as the baseline and set a modest improvement goal, such as a reduction of one or two percentage points a year. Compare with a naive forecast in each segment, so that you know whether the process is adding value. An unreachable target invites gaming, and an easy one wastes the exercise.

Using WMAPE to find where to work

Because WMAPE sums errors, you can see where the errors come from. Rank items by their share of the total absolute error. In the example, item D contributes 60 of 180 units, or 33 percent, A contributes 28 percent (50 / 180 = 0.278), B and C each contribute 17 percent (30 / 180 = 0.167) and E contributes 6 percent. Working on D and A will reduce WMAPE the most in unit terms. In dollar terms the ranking changes: C contributes 600 / 1,720 = 35 percent and E 500 / 1,720 = 29 percent. The question of which items to prioritize depends on which weighting reflects your economics, and the answer is usually dollars.

Traps to avoid

Aggregating before you compute. Computing WMAPE on a monthly total of all items is not the same as the sum of item errors. The former cancels over- and under-forecasts and hides problems. Compute errors at the level of the decision first, then sum.

Ignoring zero-demand periods. Unlike MAPE, WMAPE handles zeros in individual items, but it treats forecast units on a zero-demand item as pure error, which is right. Do not drop them.

Comparing across different mixes. A company with mostly stable products will have a lower WMAPE than one with many promotions. Compare with yourself and with a naive benchmark.

Mixing lags. Measure at the lead time of your supply chain.

Quoting accuracy without bias. A low WMAPE with a high bias still creates inventory problems.

Reporting it well

Show a trend over at least 12 months, with the segment breakdown, the contribution of the top items and the bias beside it. Include the naive forecast WMAPE as a benchmark. Put the definition in a footnote: level, lag, units or dollars. Review the same page every month so that people learn to read it.

Common mistakes

1. Using MAPE and calling it WMAPE.
2. Computing on aggregates, which hides item errors.
3. One target for everything.
4. No bias beside it.
5. Ignoring dollars when cost per unit varies.
6. Changing the definition without telling anyone.

FAQs

What is WMAPE?

The weighted mean absolute percentage error: total absolute forecast error divided by total actual demand, so larger items carry more weight.

How do you calculate WMAPE in Excel?

With actuals in one range and forecasts in another, use SUMPRODUCT of the absolute differences divided by the sum of actuals.

Is WMAPE better than MAPE?

For most portfolios, yes. It handles zero-demand items and does not let small items dominate.

What is the difference between WMAPE and WAPE?

None. They are the same calculation under different names.

Should I weight by units or dollars?

Units suit operations, dollars suit inventory investment and finance. Many teams report both.