Ljung-Box Test for Residual Independence
You'll learn how to verify that your ARIMA model's residuals behave like random noise using a statistical test.
What It Is
After fitting an ARIMA model, you need to check whether the residuals (the differences between actual and fitted values) are independent—meaning there's no leftover pattern the model missed. The Ljung-Box test is a formal hypothesis test that checks whether residuals show significant autocorrelation at multiple lags at once.
Think of it like checking that you've squeezed all the juice from an orange. If your model captured all the patterns in your time series, the residuals should look like white noise—completely random with no predictable structure.
How It Works
The test examines the first h autocorrelations of your residuals together:
- Null hypothesis (H₀): Residuals are independently distributed (no autocorrelation exists)
- Alternative hypothesis (H₁): Residuals show autocorrelation
The test produces a p-value. If the p-value is greater than your significance level (typically 0.05), you fail to reject the null hypothesis—good news! Your residuals behave like white noise, suggesting your model captured the time dependencies adequately.
If the p-value is low (< 0.05), your residuals still contain patterns, meaning your ARIMA model may need refinement.
When to Use It
Apply the Ljung-Box test after fitting your ARIMA model as part of residual diagnostics. It's more powerful than examining individual ACF lags because it tests multiple lags simultaneously, giving you a comprehensive verdict on residual independence.
Key Takeaway: The Ljung-Box test confirms whether your ARIMA residuals are white noise; a high p-value (> 0.05) indicates good model fit with no remaining autocorrelation patterns.