Using ACF and PACF Together
What you'll learn: How to combine ACF and PACF plots to determine the right ARIMA model components for your time series.
The Power of Pattern Matching
Now that you understand ACF and PACF separately, the real skill is using them together to identify your model. Think of ACF and PACF as two different diagnostic tools—like an X-ray and an MRI—that together reveal what's happening inside your time series.
The Classic Identification Patterns
When you plot ACF and PACF side-by-side, specific patterns point to specific model types:
Pure AR (Autoregressive) Process:
- ACF: Gradually decays (exponentially or with damped sine waves)
- PACF: Sharp cutoff after lag p
- Conclusion: Use AR(p) model, where p is the last significant lag in PACF
Pure MA (Moving Average) Process:
- ACF: Sharp cutoff after lag q
- PACF: Gradually decays
- Conclusion: Use MA(q) model, where q is the last significant lag in ACF
Mixed ARMA Process:
- Both ACF and PACF: Gradual decay (no clear cutoff)
- Conclusion: You likely need both AR and MA components—experiment with different combinations
The Practical Workflow
- Plot both ACF and PACF for your (stationary) time series
- Look for cutoffs versus gradual decay
- Note which lags exceed the significance bounds
- Match the pattern to the table above
- Start with the suggested model order as your initial guess
Key Takeaway: ACF and PACF together form a pattern-matching toolkit—ACF cutoffs suggest MA order, PACF cutoffs suggest AR order, and when both decay gradually you need a mixed ARMA model.