STOP 18 / 31Different tools. Different assumptions.
Layout
STOP 18 · Models

Different tools. Different assumptions.

EXPLORE~5 min
Back to route map
  • There is no universally best forecasting method.
  • Model choice depends on the data, the horizon, the available predictors, and the decision Northstar needs to make.
Model families: concepts and Northstar examples. Illustrative Northstar examples, not measured business results.
ConceptWhat it meansNorthstar example
ETS describes evolving components.

Exponential smoothing models represent changing level, trend, and seasonality, with an explicit error model.

ETS lets Northstar represent evolving level, trend, and specified seasonal components rather than treating the whole sales line as one pattern.

ARIMA describes temporal dependence.
  • It models relationships among lagged values and errors, often after differencing.
  • Regression with ARIMA errors can add calendar and external drivers.

ARIMA investigates temporal dependence remaining in Northstar's appropriately differenced or adjusted series.

Regression explains relationships to features.
  • Calendar, price, promotion, and Fourier features can be useful when future values are known or supplied as scenarios.
  • Machine learning adds flexibility, not immunity to leakage.
  • A regression can use known calendar dates and planned discounts.
  • It cannot legitimately use next week's realized ad clicks to predict next week.
FIG 18.1

Choose the tool for the pattern

WHAT THIS SHOWS

Choose a model family to inspect which part of Northstar's forecasting problem its assumptions address.

The question it answers: Which candidate is motivated by the pattern you need to explain?

Model family
ETS · MENTAL MODEL

Evolving components

Level, trend, seasonal states

Useful starting point: Simple, stable seasonal patterns

These are rough strengths, not strict boundaries. Hybrid models combine ideas. Compare every candidate with the same baseline, origins, and horizons.

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

What is the strongest reason to choose a more complex model?

Your learning note

Private · just for you