Time series and forecasting · insight-0208
Guard MASE against zero denominators and state which naive scaling you used
- Evidence type
- documented fact
- Difficulty
- medium
- Roles
- Data scientist, Data analyst
- Industries
- retail, manufacturing
Summary: situation, action, reason, boundary
Situation
MASE is computed across thousands of SKUs and some values come back infinite or enormous.
Action
Detect series with zero or near-zero naive MAE, report them separately with an absolute metric, and document whether the scale uses naive or seasonal-naive errors and which training window.
Reason
MASE scales errors by the in-sample naive MAE; it is only undefined when that scale is zero, which flat series produce.
Boundary
Large retail or spare-parts panels, new products, intermittent series.
Symptom or risk
Series that were constant or all zero in the training window have zero naive in-sample error, so MASE divides by zero and a few items dominate the average.
Exceptions
Series with stable non-zero variation give a well-defined, comparable MASE.
Trade-offs
Excluding problem series makes the headline cleaner but hides part of the portfolio; always show the excluded count and volume.
Pitfalls
Comparing MASE values computed with different seasonal periods or training windows.
Lessons and concepts
Claims and their sources
MASE scales absolute errors by the in-sample MAE of a naive (or seasonal naive) forecast; values below one mean better than the average naive forecast on training data.
Hyndman and Koehler proposed MASE as a scale-free accuracy measure that avoids the problems of percentage and relative errors.