STOP 19 / 31Put smoothing on a statistical footing.
Layout
STOP 19 · More models

Put smoothing on a statistical footing.

EXPLORE~6 min
Back to route map
  • ETS names the Error, Trend, and Seasonal components of a state-space model.
  • It turns smoothing recursions into probabilistic models with likelihood-based estimation and prediction intervals.
ETS: the framework: concepts and Northstar examples. Illustrative Northstar examples, not measured business results.
ConceptWhat it meansNorthstar example
Read the three letters.
  • ETS(A,A,A) has additive errors, additive trend, and additive seasonality.
  • N means none; M means multiplicative where allowed.
  • Damping is a separate trend choice.

Northstar reads ETS(A,N,A) as additive error, no trend, and additive seasonality, rather than treating ETS as a single fixed model.

Same points, different uncertainty.

Some additive and multiplicative error models can have the same point forecasts but different prediction intervals.

Two ETS choices can have similar point forecasts for Northstar but imply different uncertainty, so compare their interval behavior too.

Select responsibly.
  • Information criteria such as AICc help compare suitable fitted candidates.
  • Still check residuals and holdout performance at your actual horizon.

A lower training AICc can motivate a candidate; Northstar still checks whether its chosen model forecasts future event periods usefully.

FIG 19.1

How quickly should a model react?

WHAT THIS SHOWS

Move alpha to change the weight given to Northstar's latest observation. The level update is computed recursively; additional supplied trend illustrations are labeled separately.

The question it answers: How does a faster reaction trade off against following noise?

Interactive time series chart$23k$27k$31k$35kTodayHistoryLevel shift+20 days
Observed salesSmoothed levelIllustrative forecast

Higher α follows the level shift faster but also follows noise. This playground computes SES level updates, not a full seasonal ETS fit. SES point forecasts are flat beyond the last observation.

Fine print & assumptions

Constructed teaching data and illustrative calculations, not a production forecasting service. Read the diagram's method notes and distinguish supplied simulator components from fitted estimates.

CHECK YOURSELF · CHOOSE AN ANSWER

In ETS(A,N,A), what is absent?

Your learning note

Private · just for you