Course contentsShow
← Data Science
Lesson 766 of 2,14520. Time Series ModelsFree lesson

Additive vs Multiplicative Seasonality

Understanding when to use additive seasonality versus multiplicative seasonality in Holt-Winters.

Additive vs Multiplicative Seasonality

What you'll learn: When to choose additive versus multiplicative seasonality in the Holt-Winters forecasting method.

The Core Distinction

In Holt-Winters forecasting, you need to decide how seasonality interacts with your trend. Additive seasonality means seasonal fluctuations stay roughly constant in size regardless of the data's level. Multiplicative seasonality means seasonal swings grow or shrink proportionally with the trend level.

Additive Seasonality: Fixed Swings

Imagine monthly sales that swing ±200 units every summer and winter, whether your baseline is 1,000 units or 5,000 units. The seasonal effect adds the same absolute amount each period.

Use additive when:

  • Seasonal variations are roughly constant over time
  • The data doesn't show seasonal amplitude changing with the trend
  • Your data includes values near or below zero

Multiplicative Seasonality: Proportional Swings

Now imagine sales that increase by 20% every December, regardless of base level. When baseline sales are 1,000, December adds 200 units. When baseline grows to 5,000, December adds 1,000 units. The seasonal effect multiplies the current level.

Use multiplicative when:

  • Seasonal variations grow larger as the trend increases
  • You see "wider" seasonal peaks as time progresses
  • Your data is strictly positive (multiplication with negatives causes issues)

Visual Clue

Plot your time series decomposition. If the seasonal component's range stays flat over time, choose additive. If it expands or contracts with the trend, choose multiplicative.

Key Takeaway: Choose additive seasonality when seasonal fluctuations have constant amplitude, and multiplicative seasonality when they scale proportionally with the trend level—this choice directly impacts Holt-Winters forecast accuracy.