Identifying Seasonal Patterns
What you'll learn: How to use ACF and PACF at seasonal lags to detect seasonal autoregressive and moving average components in your time series.
The Core Idea
You've already learned how ACF and PACF help identify regular ARIMA patterns at short lags (lag 1, 2, 3...). But seasonal time series have another layer of patterns that repeat at fixed intervals—like monthly sales peaking every December or quarterly earnings following a yearly cycle.
To detect these seasonal patterns, you look at ACF and PACF values at the seasonal lags: lag 12 for monthly data (annual seasonality), lag 4 for quarterly data, or lag 7 for daily data (weekly patterns).
Reading Seasonal Spikes
Think of ACF and PACF plots as having two "zones" of interest:
- Near-term lags (1, 2, 3...): Show regular AR/MA structure
- Seasonal lags (12, 24, 36... for monthly data): Reveal seasonal structure
Seasonal MA terms show up as significant spikes in the ACF at seasonal lags (12, 24, 36) while cutting off after a certain seasonal lag. If you see a spike at lag 12 only, that suggests one seasonal MA term.
Seasonal AR terms appear as significant spikes in the PACF at seasonal lags that cut off, while the ACF shows a gradual decay at those seasonal intervals. A spike at lag 12 in PACF suggests one seasonal AR term.
The Detective Work
When examining plots, look for patterns that repeat at multiples of your seasonal period. A spike at lag 12 and lag 24 in monthly data tells you there's structure repeating annually and bi-annually—a strong seasonal signal worth modeling.
Key Takeaway: Identifying seasonal patterns requires examining ACF and PACF at seasonal lag multiples (12, 24, 36 for monthly data) to determine the presence and order of seasonal AR and MA components, just as you do for non-seasonal terms at early lags.