🎯 Forecast Accuracy

A forecast is a promise about the future; accuracy metrics grade how well it was kept. Three questions matter: how big were the misses (error / MAPE), and were they lopsided — always too high or too low (bias)? Enter forecasts and what actually happened, and watch every number get calculated, period by period.

The three metrics

For each period t, the forecast error is simply what happened minus what you predicted:

Error_t = Actual_t − Forecast_t Positive = you under-forecast (sold more than planned). Negative = you over-forecast.

How big? Average the size of the misses, ignoring direction:

MAD (mean abs. deviation) = mean of |Error_t| · MAPE = mean of |Error_t| ÷ Actual_t MAPE is a percentage, so it compares products of any size. “Forecast accuracy” is usually quoted as 100% − MAPE.

Lopsided? Bias keeps the sign, so over- and under-shoots cancel — what's left is the systematic lean:

Bias (mean error) = mean of Error_t · Bias% = Σ Error_t ÷ Σ Actual_t Near zero is healthy. A persistent + or − means the forecast is consistently wrong one way and should be re-centred.

Enter your data

Forecast (one number per period)

Actual demand

Overall accuracy

MAPE

–

Accuracy (100−MAPE)

–

MAD

–

Bias (mean error)

–

Bias %

–

Period-by-period calculation

Spotlight period

Highlights one row and shows its full working below

Rolling accuracy

One overall number hides trends — a forecast can be drifting out of control while its lifetime average still looks fine. A rolling window recomputes the metrics over just the last few periods, so you see accuracy as it moves.

Rolling window

How many recent periods each rolling figure averages

Rolling MAPE Rolling bias % zero bias