Chapter 14 Trend and seasonal fluctuations

Economic and social time series often exhibit systematic patterns beyond random variation. Two of the most important are long-term trends and seasonal fluctuations. Identifying and separating these components helps with interpretation, comparison, and forecasting.

14.1 Components of a time series

A classical time series \(y_t\) can be decomposed into several components:

  • Trend (\(T_t\)) – long-term direction of change
  • Seasonal component (\(S_t\)) – regular, calendar-related fluctuations
  • Irregular component (\(I_t\)) – random and unsystematic variation

Example

We use monthly death data from Statistics Netherlands. In Figure 14.1, we observe evidence of both a trend and annual seasonality, with mortality levels typically higher in winter than in summer.

Monthly deaths in Netherlands 2009–2024. Source: Statistics Netherlands (https://opendata.cbs.nl/#/CBS/en/dataset/83474ENG/table?dl=24F05)

Figure 14.1: Monthly deaths in Netherlands 2009–2024. Source: Statistics Netherlands (https://opendata.cbs.nl/#/CBS/en/dataset/83474ENG/table?dl=24F05)

We use the stl() function from the stats package in R to attempt to decompose the series into trend, seasonal, and irregular components. Results are presented in figure 14.2.

Monthly deaths in Netherlands 2009–2024. Decomposition attempt with stl() function in R.

Figure 14.2: Monthly deaths in Netherlands 2009–2024. Decomposition attempt with stl() function in R.

14.2 Moving average

14.3 Linear trend model

A simple parametric representation of trend is the linear trend model:

\[y_t = \alpha + \beta t + \varepsilon_t\]

where:

  • \(\alpha\) is the intercept,
  • \(\beta\) measures the average change per period.

Interpretation:

  • \(\beta > 0\): upward trend
  • \(\beta < 0\): downward trend

Linear trends are easy to estimate and interpret, but may oversimplify long time horizons.