Seasonal Decomposition

Seasonal decomposition is a time series technique that splits a series into trend, seasonal and residual components; the classical additive version uses a centred moving average for trend and period averages for seasonality.

계절 분해는 시계열을 추세, 계절, 잔차 성분으로 나누는 기법으로, 고전적 가법 방식은 중심이동평균으로 추세를, 주기 위치별 평균으로 계절성을 구한다.

How do I separate the trend from the seasonal pattern in my monthly data?

월별 자료에서 추세와 계절 패턴을 어떻게 분리하나요?

Classical seasonal decomposition estimates the trend with a centred moving average, the seasonal pattern as the average deviation at each position in the cycle, and treats what remains as the residual. Strength measures between 0 and 1 then summarise how much of the variation is due to the trend and to seasonality.

When to use Seasonal Decomposition

• Describing trend and seasonality in monthly, quarterly or weekly data
• Checking whether seasonality is strong enough to model before forecasting
• Producing a seasonally adjusted view of a series

When to use something else

• Seasonal patterns that change over time or series with outliers: use STL decomposition
• Forecasting future values: use exponential smoothing (Holt-Winters) or a SARIMAX model

Assumptions and requirements

• Regularly spaced observations covering at least two full seasonal cycles
• Additive structure, with a seasonal effect of constant size
• A fixed seasonal period, inferred from the date frequency or set by the user

What the analysis reports

• Seasonal period used
• Seasonal strength (0 to 1)
• Trend strength (0 to 1)
• Table of trend, seasonal and residual values for each time point

How to report it

An additive seasonal decomposition with a period of [value] indicated [strong] seasonality (seasonal strength = [value]) and a [moderate] trend (trend strength = [value]).

Related methods

• Exponential Smoothing (Holt-Winters)
• SARIMAX Model
• KPSS Test

Run Seasonal Decomposition on Quorant