STOP 11 / 31Make the pattern stable enough to learn.
Layout
STOP 11 · Signal

Make the pattern stable enough to learn.

EXPLORE~7 min
Back to route map
  • Many ARIMA ideas work with a stationary series: its statistical behavior does not depend on where you are in time.
  • Raw growing sales with a changing seasonal pattern usually do not meet that ideal.
Stationarity & differencing: concepts and Northstar examples. Illustrative Northstar examples, not measured business results.
ConceptWhat it meansNorthstar example
Transform the scale when needed.
  • Logs or Box–Cox transformations can stabilize proportional variation.
  • Logs require positive values.
  • Transforming and differencing solve different problems.
  • If Northstar's fluctuations grow proportionally with sales, a suitable transformation may stabilize their scale.
  • Check positivity before using logs.
Difference to remove changing levels.
  • First differencing uses yₜ − yₜ₋₁.
  • Seasonal differencing uses yₜ − yₜ₋ₘ.
  • Choose differencing using plots, domain knowledge, and tests—not an automatic rule.

A weekly difference compares today's sales with seven days earlier, helping remove a stable weekday pattern before modeling remaining dependence.

Don’t over-difference.
  • Unnecessary differences can add noise and create artificial correlation.
  • A one-off structural change or deterministic trend may require other treatment.
  • A permanent pricing change is not a reason to difference Northstar's sales repeatedly until a chart looks flat.
  • Diagnose the structural change first.
Δyₜ = yₜ − yₜ₋₁ · seasonal Δ₇yₜ = yₜ − yₜ₋₇
FIG 11.1

Levels versus changes

WHAT THIS SHOWS

Switch between Northstar's sales levels, daily changes, and same-weekday changes.

The question it answers: Which transformation removes the changing level or repeating pattern, and what still remains?

Working series
Interactive time series chart$20k$23k$26k$29kJanMarMayJun
Daily sales

First differences expose day-to-day changes; seasonal differences compare the same weekday. These transformations do not automatically guarantee stationarity or remove a moving event.

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 does the ‘I’ in ARIMA represent?

Your learning note

Private · just for you