C49. Identification and residual checks

Topic: Forecasting · more questions on this topic
‹ C47

How can ACF/PACF patterns suggest AR or MA orders? What should the fitted residual ACF be checked for?

Worked answer (click to show)

Definitions/notation. For the usual causal stationary AR(p) model the population PACF vanishes beyond lag p; for MA(q), the population ACF vanishes beyond lag q. Sample patterns are noisy clues rather than proofs of order, and the residual ACF should be checked for remaining serial correlation that the fitted model has not captured.