ORIGINAL COMPANY-CONTEXT PRACTICE
Compare fuel-demand forecasts at a real ordering horizon
Original practice scenario: A hypothetical distribution planner orders fuel several days ahead. A model beats a baseline on next-day forecasts but performs poorly on ordering decisions. Explain the mismatch.
Written for independent practice using company context; this is not a verified question from a company interview.
A REASONED APPROACH
Model answer.
Evaluate at the decision horizon and available-information cutoff. A model can be accurate one day ahead while providing little value when replenishment commitments must be made earlier.
Work through it
- Define location-product demand, order lead time and decision timestamps; reconstruct only the information available when each historical order would have been placed.
- Backtest rolling origins at the actual horizons, comparing seasonal-naive and simple calendar baselines with the candidate under identical training cutoffs.
- Report horizon-specific error and uncertainty alongside stockout and surplus simulations using stated assumptions about delivery limits and inventory.
Tradeoffs to discuss
- Optimizing forecast error is simpler than evaluating inventory decisions, but operational value also depends on capacity and asymmetric shortage costs.
Common pitfalls
- Do not use revised historical demand that was unknown at the order time without marking the experiment retrospective.
- Separate lost sales from observed demand when stockouts censor what customers could purchase.