2.2 Seasonal component and patterns

Seasonality in Time Series

Seasonality in Time Series

Seasonality refers to regular, predictable patterns that repeat over a fixed period in your data. Think of it this way: if you plotted monthly ice cream sales over several years, you'd see the same spike every summer and the same dip every winter. That repeating shape is seasonality.

These patterns are driven by factors that consistently influence data at specific times:

• Weather patterns cause predictable swings in energy usage, agriculture, and retail (e.g., winter heating costs rise every December through February)
• Holidays create reliable spikes in spending and travel (e.g., Christmas retail sales, Thanksgiving airline bookings)
• School schedules shape demand cycles (e.g., back-to-school shopping in August, lower activity during summer breaks)
• Business cycles impose structure on financial data (e.g., quarterly earnings reports, year-end budget spending)

Seasonality is one of the most common features in real-world time series. If you ignore it, you risk drawing wrong conclusions. For example, you might mistake a normal summer sales increase for evidence of a growing trend, or you might underestimate demand heading into a holiday season.

Recognition of Seasonal Patterns

The simplest way to spot seasonality is to look at a time series plot. If you see regularly repeating peaks and troughs, that's a strong signal. The distance between consecutive peaks tells you the periodicity, which is the length of one full seasonal cycle.

Common periodicities include:

• Yearly (period = 12 months): Holiday retail sales peak every December; allergy medication sales spike every spring
• Quarterly (period = 3 months): Corporate earnings follow a quarterly reporting cycle; some sports viewership repeats every quarter
• Monthly (period within a month): Utility bills may follow a consistent monthly pattern tied to billing cycles
• Weekly (period = 7 days): Restaurant traffic peaks on weekends; streaming TV viewership dips on weekday mornings

Getting the periodicity right matters. If your data has a yearly cycle but you model it as quarterly, your seasonal adjustments won't line up with the actual pattern, and your forecasts will suffer.

Additive vs. Multiplicative Seasonality

Not all seasonal patterns behave the same way. The two main types differ in how the seasonal effect relates to the overall level of the series.

Additive seasonality means the seasonal swings stay roughly the same size regardless of the series level. If holiday sales always add about 500 units above the trend, whether the baseline is 2,000 or 5,000 units, that's additive. The decomposition model is:

$Y_t = T_t + S_t + I_t$

• $Y_t$: observed value at time $t$
• $T_t$: trend component at time $t$
• $S_t$: seasonal component at time $t$
• $I_t$: irregular (random) component at time $t$

Multiplicative seasonality means the seasonal swings grow or shrink in proportion to the series level. If holiday sales are always about 20% above the trend, the absolute increase gets larger as the trend rises. The model is:

$Y_t = T_t \times S_t \times I_t$

Here, $S_t$ is expressed as a ratio (e.g., 1.20 for 20% above trend, 0.85 for 15% below).

Impact of Seasonality on Forecasting

Seasonality can mask what's really going on in your data. A genuine upward trend might be hard to see when large seasonal swings dominate the plot. Conversely, a flat trend might look like growth if you only compare seasonal peaks.

Seasonal adjustment techniques strip out the seasonal component so you can analyze the underlying trend and irregular movements more clearly:

• Seasonal decomposition methods like STL (Seasonal and Trend decomposition using Loess) and Census X-13 split the series into its trend, seasonal, and irregular parts
• The resulting seasonally adjusted data makes it much easier to spot real changes in direction or level

Forecasting models that explicitly incorporate seasonality produce more accurate predictions:

• Seasonal ARIMA (SARIMA) extends the standard ARIMA framework by adding seasonal autoregressive and moving average terms
• Holt-Winters exponential smoothing captures both trend and seasonality, with separate smoothing parameters for each component

Failing to account for seasonality in your forecasts can lead to systematic errors. You might consistently overpredict during off-peak periods and underpredict during peak ones, which translates directly into poor resource allocation and planning decisions.