Empirical Comparisons of Seasonal ARIMA and ARIMA Component (Structural) Time Series Model
William R. Bell
This paper compares seasonal ARIMA models as presented in Box and Jenkins (1970) with
ARIMA component (structural) models as presented in Harvey (1989). Both models are
augmented as appropriate with the same regression variables to account for calendar effects, level
shifts, and additive outliers. The models are compared on a set of 40 Census Bureau monthly
time series in regard to fit using AIC and related statistics. Bell and Pugh (1990) made similar
comparisons of ARIMA models with the basic structural model (BSM). This paper extends their
work by also considering ARIMA component models with trigonometric seasonal components.
For the 40 time series considered, AIC ARIMA models over the ARIMA component models.