Pick and interpret accuracy metrics that fit the demand pattern and business risk. See if you can navigate scale effects, zeros, seasonality, and cost asymmetry.
RMSE differs from MAE primarily because RMSE ______.
Squares errors and penalizes large misses more heavily
Is scale-free across series
Ignores large errors entirely
Can never exceed the standard deviation
A common limitation of MAPE in retail demand forecasting is that it ______.
Requires log-transformed data
Becomes unstable or undefined when actuals approach zero
Is measured in the same units as the series
Penalizes over- and under-forecasting equally
MASE makes errors comparable across series because it ______.
Uses the maximum observed value as a divisor
Normalizes by the mean of predictions
Scales by the in-sample naive forecast error
Divides by sample size only
For strongly seasonal weekly data, an appropriate MASE baseline uses ______.
A linear trend with no seasonality
Random walk with drift
A moving average with window 3 only
A seasonal naive forecast with a 52-week lag
Compared to RMSE, MAPE is often preferred by business users because it ______.
Expresses error as a percentage that’s easy to interpret across price points
Is unaffected by zeros in the data
Is always more statistically efficient
Guarantees symmetry between over- and under-forecasting
sMAPE differs from MAPE by dividing by ______.
The geometric mean of actuals
The average of absolute actuals and forecasts
The maximum actual value
The standard deviation of the residuals
When selecting a single-number score for model selection across different SKUs with different scales, a good choice is ______.
Mean RMSE across series
Maximum sMAPE across series
Median MASE across series
Sum of MAE across series
Weighted MAPE (wMAPE) helps when SKUs vary in volume because it ______.
Uses squared residuals like RMSE
Penalizes over-forecasting only
Removes the need for seasonality modeling
Weights errors by actual quantities or revenue
If the cost of stockouts is much higher than overstock, the best metric choice is one that ______.
Reflects asymmetrical costs, such as a custom weighted loss
Always uses MAE because it is robust
Defaults to MAPE for interpretability
Uses RMSE because it squares errors
Before computing MAPE for intermittent demand, a common practice is to ______.
Switch to classification metrics
Aggregate or smooth to reduce zeros or use alternative metrics like sMAPE/MASE
Drop high-demand periods only
Replace zeros with ones universally
Starter
Beginner level—review when percentage metrics break and why scale-free scores help.
Solid
Good read on pros and cons—start aligning metrics with item volume and risk.
Expert!
Expert: your metric choices match data quirks and business costs across portfolios.
