Time series and forecasting · insight-1247
Use a probability-updating method such as TSB when items can become obsolete
- Evidence type
- documented fact
- Difficulty
- hard
- Roles
- Data scientist, ML engineer
- Industries
- manufacturing, retail, logistics
Summary: situation, action, reason, boundary
Situation
A wholesaler's slow items sometimes stop selling for good, yet a Croston forecast stays at the old demand rate for months because it only updates when a sale occurs.
Action
Add a method that updates the probability of demand every period, such as TSB, to the candidates. Evaluate it on items with long zero runs and on the rest separately, and compare decay speed with bias.
Reason
Croston's method does not update during long runs of zero demand, whereas TSB replaces the interval with a demand probability that is updated every period, so its forecast decays.
Boundary
Spare parts, fashion and technology goods with finite life cycles.
Symptom or risk
Stock is replenished against demand that no longer exists, and obsolescence cost is attributed to the buyer rather than the forecasting rule.
Exceptions
Where non-sales reflect stockouts or seasonal gaps rather than obsolescence, fast decay will under-forecast.
Trade-offs
A faster-decaying probability smoother raises the risk of understocking when demand returns.
Pitfalls
Evaluating on all items together, so the few obsolescing items vanish in the average.
Lessons and concepts
Claims and their sources
Teunter, Syntetos and Babai propose TSB, which updates a demand probability every period, because Croston's method does not update during long runs of zero demand and is positively biased.