MerQuR+Statistics Glossary

Statistics Glossary → Seasonality and seasonal period

Seasonality and seasonal periodconcept

A pattern in a time series that repeats regularly with a calendar cycle of fixed length (day, week, year); the length of the cycle in number of observations is the seasonal period (m).

The period depends on the frequency of the data: m = 12 for a yearly cycle in monthly data, 4 in quarterly data, 7 for a weekly cycle in daily data, and 24 for a daily cycle in hourly data. A series can have several seasonalities (e.g. both daily and weekly cycles in hourly electricity demand). With additive seasonality the amplitude of the seasonal swing does not depend on the level of the series; with multiplicative seasonality the swing grows in proportion as the level grows. A multiplicative pattern can usually be made additive with a log transformation.

Seasonality is handled by decomposition (STL), seasonal dummies, seasonal differencing (yₜ - yₜ₋ₘ), or seasonal ARIMA and exponential smoothing models. Seasonal strength summarizes, on a scale from 0 to 1, how dominant the seasonal component is relative to the remainder (the Wang, Smith and Hyndman measure). Seasonality should not be confused with cyclical fluctuations that occur at irregular intervals (e.g. business cycles), whose length is not fixed. If the seasonal structure is ignored, trend tests and forecasts mislead; a series that rises every summer, for example, might be taken as 'trending upward' by looking only at summer months.

An everyday example

An ice-cream seller's monthly sales peak every July and hit their lowest in January; the period is 12 months. As the business grows, the gap between summer and winter also grows from 2,000 to 6,000 tubs; this is multiplicative seasonality. Taking the log of sales keeps the summer-winter gap roughly constant over the years and makes the analysis easier.

In MerQur

Available as 'Period' (0 = automatic) and 'MSTL periods' in STL seasonal decomposition, 'Seasonal type' (additive / multiplicative / none) and 'Seasonal period' in exponential smoothing, 'Seasonal' (dummy, trigonometric, none) in the unobserved components model and 'Seasonal dummies (Johansen)' in cointegration. It also appears in the Time Series, ARIMA, Mann-Kendall (seasonal Kendall test) and anomaly detection (STL remainder) analyses.

Analyses where this concept appears

Time SeriesSTL Seasonal DecompositionARIMA ForecastExponential Smoothing (Holt-Winters)Mann-Kendall + Sen's SlopeAnomaly DetectionCointegration (Johansen / Engle-Granger)Unobserved Components Model (Structural Time Series)

Related concepts

Seasonal Kendall testDifferencingPeriodogram and spectral density