PACF Patterns for Common Models
You'll learn: How to recognize the signature "fingerprints" that AR, MA, and ARMA models leave in their PACF plots.
Why Patterns Matter
Just as doctors recognize disease patterns in medical tests, data scientists identify time series models by their characteristic ACF and PACF signatures. The PACF plot acts like a diagnostic tool—each model type leaves a distinct pattern that helps you choose the right model structure.
AR(p) Models: Sharp Cutoff
An AR(p) process shows a sharp cutoff in the PACF after lag p. The first p lags will be significant, then all later lags drop to zero (within significance bounds). For example, an AR(2) model shows two significant spikes at lags 1 and 2, then nothing.
Think of it like echoes in a canyon: only the first few bounces (direct connections) matter; indirect echoes fade completely.
MA(q) Models: Gradual Decay
An MA(q) process shows gradual exponential decay or a damped sinusoidal pattern in the PACF—no clean cutoff. This mirrors how MA models show sharp cutoffs in the ACF but diffuse patterns in the PACF.
ARMA(p,q) Models: Mixed Behavior
ARMA models combine both components, so their PACF shows gradual decay (influenced by the MA part) rather than a clean cutoff. These are trickier to identify from PACF alone—you need both ACF and PACF together.
Quick Reference Table
| Model | PACF Pattern |
|---|---|
| AR(p) | Cuts off after lag p |
| MA(q) | Gradual decay/tails off |
| ARMA(p,q) | Gradual decay/tails off |
Key Takeaway: AR models have clean PACF cutoffs at lag p; MA and ARMA models show gradual decay. Use PACF and ACF together to confidently identify your model structure.