Accuracy & evaluation

Forecast bias

Bias is the average signed error: systematically high or systematically low. A forecast can have a handsome WAPE and still drain cash if it is always 6% heavy.

Over-forecasting perishable demand is waste. Under-forecasting is empty shelves and a quieter kind of waste. Bias tells you which way you lean.

Percent bias (mean error / mean actual) is easy to put next to WAPE. Tracking it by category often shows the model is fine in aggregate and drunk on one brand.

You can debias with a simple multiplier if the miss is stable. If bias flips every month, you have a regime problem, not a constant to twiddle.