Skip to content
Tutorials Center
My learning

Topics

Lesson 6 of 40 ยท Components and Decomposition

Trend, Seasonality and Noise

Understand the components of a time series: trend, seasonality, cycles and noise. Learn additive versus multiplicative structure and build one by simulation to see how they combine.

Almost every series you meet can be described as a combination of a few building blocks. Naming them is the key to both understanding and forecasting: once you know which blocks a series has, you know which methods to reach for. This lesson defines the components and shows how they combine by building a series from scratch.

The components

Four building blocks

Trend Slow long-term direction.
Seasonality Fixed-length repeating pattern.
Cycle Irregular, longer swings.
Noise Random leftover.

Trend plus season plus noise covers most business data.

1 / 2
  • Trend: the long-run movement of the level, up, down or flat. It may be a straight line or a curve.
  • Seasonality: a pattern that repeats with a fixed, known period tied to the calendar: holiday sales, daily temperature, weekday traffic.
  • Cycle: rises and falls with no fixed period, such as business cycles lasting several years. Cycles are harder to forecast than seasons because their length varies.
  • Noise (also called the irregular or remainder component): what is left after the rest. Ideally it is unpredictable.

A common mistake is to call any wave a season. If the length changes from one swing to the next, it is a cycle.

Additive structure

In an additive model the components are added:

y(t) = trend(t) + season(t) + noise(t)

The size of the seasonal swing stays the same whether the level is high or low. Use it when the swings look like constant-width ribbons.

Multiplicative structure

In a multiplicative model the components are multiplied:

y(t) = trend(t) ร— season(t) ร— noise(t)

Here the seasonal swing grows with the level: a shop that doubles its sales also doubles its holiday spike. Many economic series behave like this. Taking logarithms turns a product into a sum, which is why log is such a common transformation (later).

flowchart TD
    A["Do seasonal swings grow with the level?"] -->|"no, constant size"| B["Additive"]
    A -->|"yes, proportional"| C["Multiplicative"]
    C --> D["Take log to make it additive"]

A quick way to choose.

Build both and compare

We build the same trend and season, but combine them two ways. The measure we print is the size of the seasonal swing (max minus min) within each year.

Python Additive versus multiplicative

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The additive swing is roughly constant, near 40 each year. The multiplicative swing keeps growing as the level rises. That growth is the signature of a multiplicative series.

Looking at both on a plot

Python Plot the two structures

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

In the right panel the waves visibly fan out, like a megaphone.

Adding structure you might miss

Real series carry more than a plain trend and sine wave:

  • A changing trend: the slope may bend or break (a pandemic, a price change).
  • Several seasons at once: hourly data can have a daily and a weekly pattern.
  • Level shifts and outliers: a sudden jump or a single wild value.
  • Calendar effects: holidays that move around (see calendar effects).
Python Two seasons in hourly data

Press Run (or Ctrl + Enter) to see the result here.

Program input ยท what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

Both a 24-step and a 7-day pattern are visible in one series. Methods that allow only one seasonal period need extra work here, and later lessons show how.

Why it matters

Identify the components before you pick a model:

You see Think
Rising line, no wave trend only: differencing, drift, Holt
Wave around a flat level seasonality only: seasonal naive, seasonal dummies
Both Holt-Winters, SARIMA, decomposition-based models
Swings grow with level multiplicative: log transform or multiplicative seasonality
Wave with changing length cycle: ARMA-type or machine learning models

Check your understanding

Quick check ยท 4 questions

of 4 answered You scored / 4 ยท perfect!

  1. 1. What distinguishes seasonality from a cycle?

    Seasonal patterns repeat at a calendar-driven fixed period; cycles rise and fall with varying length.
  2. 2. Sales double over five years and the December spike also roughly doubles. Which structure fits?

    When the seasonal swing scales with the level, the components multiply.
  3. 3. Which transformation turns a multiplicative series into an additive one?

    log(a ร— b) = log(a) + log(b), so the product of components becomes a sum.
  4. 4. Hourly electricity load with a daily and a weekly pattern has...

    A 24-hour and a 168-hour pattern coexist, so the series has two seasonal periods.

Summary

  • A series is commonly trend + seasonality (+ cycle) + noise.
  • Seasonality has a fixed period; cycles do not.
  • Additive when swing size is constant, multiplicative when it grows with the level; log converts one to the other.
  • Real data may have several seasons, shifts and calendar effects.
  • Identifying components guides model choice.

Next: Classical Decomposition