Course contentsShow
← Data Science
Lesson 730 of 2,14519. Time Series FoundationsFree lesson

Interpreting PACF Plots

Reading PACF correlograms, identifying significant lags using confidence intervals, and recognizing cutoff patterns.

Interpreting PACF Plots

You'll learn how to read PACF plots to identify which lags have direct relationships with your time series, using visual clues and confidence intervals.

What Am I Looking At?

A PACF plot shows vertical bars (or points) representing the partial autocorrelation at each lag. Think of it like looking at a bar chart where each bar tells you: "Does lag k have a direct relationship with the current value, after removing indirect effects?"

The Confidence Bands

Every PACF plot includes horizontal dashed lines (usually at ±1.96/√n). These are your confidence intervals. Any bar that pokes outside these bands is statistically significant—meaning that lag probably has a genuine direct relationship with your series, not just random noise.

Bars that stay within the bands? Likely just noise. Ignore them.

The Cutoff Pattern

Here's the magic: PACF plots often show a sharp cutoff. You'll see significant spikes at lag 1, maybe lag 2, and then... everything drops inside the confidence bands.

Example interpretation:

  • Spike at lag 1, then cutoff → Strong direct relationship only at lag 1
  • Spikes at lags 1 and 2, then cutoff → Direct relationships at lags 1 and 2
  • No significant spikes → No clear direct autoregressive pattern

This cutoff pattern is your friend for identifying how many lagged values directly influence the current value—crucial information for building models later.

Why This Matters

The PACF helps you distinguish between direct and indirect relationships. A value at time t might correlate with t-5 only because t-1 through t-4 are carrying that information forward. PACF cuts through that chain to show you the real drivers.

Key Takeaway: Read PACF plots by identifying which lags break through the confidence bands—those are your significant direct relationships—and note where the pattern cuts off sharply.