Time series and forecasting · insight-0244
Prefer simple models on very short series and say how little the backtest can tell you
- Evidence type
- documented fact
- Difficulty
- easy
- Roles
- Data scientist, Data analyst
- Industries
- aviation, retail
Summary: situation, action, reason, boundary
Situation
A new route has 30 weeks of history, and someone proposes a seasonal ARIMA with many parameters.
Action
Use simple models (naive, damped trend, ETS with few components), borrow strength from similar series or a hierarchy, and report the small number of backtest origins as a limitation.
Reason
With very short series, simple models are usually preferred and information criteria help balance fit against parameter count when holdouts are too small.
Boundary
New products, new routes, new branches, recently launched services.
Symptom or risk
Parameters outnumber the information in the data, so the fit looks good and forecasts are unstable; there is not even one full annual cycle to estimate seasonality.
Exceptions
If a global model trained on many related series is available, it may transfer seasonality to the new series.
Trade-offs
Simple models miss real structure that will appear later; plan to revisit after a full cycle.
Pitfalls
Claiming seasonality from less than one full cycle.
Lessons and concepts
Claims and their sources
For very short series, simple models are usually preferred, and AICc helps balance the number of parameters against the noise.