Time series and forecasting · insight-0220
Check interval coverage by season and segment, not only overall
- Evidence type
- practitioner heuristic
- Difficulty
- medium
- Roles
- Data scientist, Data analyst
- Industries
- aviation, retail, energy
Summary: situation, action, reason, boundary
Situation
Overall coverage of a 90% demand interval is 89%, and the team declares it calibrated.
Action
Break realised coverage down by season, horizon, segment and volume band; fix the worst slices with segment-specific calibration or a model that lets variance change with level.
Reason
Marginal coverage averages over conditions; decisions are taken in specific conditions, and errors are often heteroscedastic.
Boundary
Seasonal demand, peak load, holiday traffic.
Symptom or risk
Coverage can be 97% in off-peak weeks and 70% in peak weeks, so the average hides failures in the periods with the highest stakes.
Exceptions
Slices with few observations give noisy coverage; pool similar slices before acting.
Trade-offs
Segment calibration needs more data per slice and adds parameters to maintain.
Pitfalls
Interpreting a nominal 90% interval as 90% coverage for each departure.
Example
Synthetic airline data: check coverage separately for the seasonality labels off, shoulder and peak.