Course contentsShow
← Data Science
Lesson 786 of 2,14520. Time Series ModelsFree lesson

ACF and PACF of Residuals

Examining autocorrelation and partial autocorrelation plots of model residuals to check for remaining patterns.

ACF and PACF of Residuals

You'll learn: How to use ACF and PACF plots of model residuals to verify your time series model has captured all the patterns in your data.

Why Check Residuals with ACF and PACF?

After fitting a time series model (like ARIMA), your residuals should behave like white noise—completely random with no patterns left over. If your model was good, it "ate up" all the structure in the data, leaving only unpredictable randomness behind.

Think of it like cleaning a dirty dish: if you've done a thorough job, there shouldn't be any food residue left behind. Similarly, if you've built a good model, there shouldn't be any autocorrelation patterns remaining in the residuals.

What to Look For

When you plot the ACF of residuals, you want to see:

  • All lags falling within the significance bounds (those dashed blue lines)
  • No systematic patterns or spikes
  • Random scatter around zero

When you plot the PACF of residuals, you're checking the same thing:

  • No significant spikes at any lag
  • Random noise pattern

Why This Matters

If you see significant spikes in the ACF or PACF of your residuals:

  • Your model missed some autocorrelation structure
  • You might need a higher AR order (if PACF shows spikes)
  • You might need a higher MA order (if ACF shows spikes)
  • Consider refitting with different parameters

This diagnostic step is crucial because it tells you whether your model is complete—have you really extracted all the signal, or is there still information hiding in the errors?

Key Takeaway: Clean ACF and PACF plots of residuals (all values within significance bounds) confirm your model captured all temporal patterns, leaving only random noise behind.