MAPE Calculator (Mean Absolute Percentage Error)

MAPE (Mean Absolute Percentage Error) is the average percentage gap between forecasts and actual values. The formula is MAPE = (1 / n) * sum of |actual - forecast| / |actual| * 100. For each pair, take the absolute residual, divide by the absolute actual, average across all pairs, and multiply by 100. The result is a single percent: a MAPE of 8 means your forecasts are off by about 8 percent on average. The metric is unitless, which is why it is the default forecast-accuracy headline in retail, supply chain, and demand planning.

MAPE divides by the actual value, so any row with actual = 0 produces a division by zero and is mathematically undefined. This calculator skips zero-actual rows from the MAPE average (and tells you how many it skipped), but they still count toward MAE, RMSE, and the bias check so a real forecast miss against zero is not hidden. If your series has frequent zeros (intermittent demand, sparse counts, new product launches) MAPE is the wrong tool. Use SMAPE (Symmetric MAPE, divides by the average of |actual| and |forecast|) or MAAPE (Mean Arctangent APE, bounded between 0 and pi/2) instead. Both stay defined when actuals hit zero.

Rough rules of thumb from forecasting practice: MAPE under 10 percent is highly accurate and rare outside very stable demand; 10 to 20 percent is a good forecast typical of mature retail or manufacturing planning; 20 to 50 percent is reasonable and common in volatile categories, promotions, or new SKUs; over 50 percent is inaccurate and usually means you need a different model, more granular features, or to forecast at a higher aggregation. These thresholds also drive the verdict label in this calculator. Always benchmark MAPE against a naive baseline (last value, seasonal naive) before claiming a model is good: beating MAPE of 30 percent only matters if the naive baseline scored 45 percent.

MAE (Mean Absolute Error) averages the raw absolute residuals in the same units as your data: easy to read but not comparable across series of different scales. RMSE (Root Mean Squared Error) squares residuals before averaging, so it punishes large misses far more than small ones. Use RMSE when one big error is much worse than several small ones (capacity planning, safety stock). MAPE is the percentage analog of MAE: scale-free and easy to communicate, but it explodes near zero actuals and treats overforecasting and underforecasting asymmetrically. A solid practice is to report MAPE for the headline, MAE for the per-unit cost of an error, and RMSE to see whether a few outliers are dominating the picture. This calculator returns all three from the same dataset.

MAPE has a hard floor at 0 percent on the underforecasting side and no upper bound on the overforecasting side. If actual is 100 and you forecast 0, the absolute percentage error is 100 percent. If actual is 100 and you forecast 300, the absolute percentage error is 200 percent. Forecasting twice the truth costs more than forecasting zero, even though both miss by the same absolute amount in opposite directions. This asymmetry creates a quiet bias: a model that minimizes MAPE will systematically underforecast to stay safe. The bias indicator in this calculator flags exactly this pattern (overforecast or underforecast on average), and SMAPE was designed specifically to fix the asymmetry by dividing by the average of |actual| and |forecast| instead of |actual| alone.