11.6 ARIMA

ARIMA models use autoregressive terms, differencing, and moving-average error terms to represent serial dependence. Seasonal ARIMA extends the structure to regular seasonal cycles.

ARIMA can be effective when historical dependence contains useful predictive information. The analyst should examine stationarity, differencing, residual autocorrelation, parameter stability, and forecast plausibility. A well-fitted ARIMA model does not automatically explain the operational cause of change.