Selección combinada de modelos ARIMA para el desestacionalizado de las ramas de actividad del EMAE[Combined ARIMA Model Selection for the Seasonal Adjustment of EMAE Activity Branches]

The seasonal adjustment of the series comprising the Monthly Estimator of Economic Activity (EMAE) requires an appropriate selection of the ARIMA models used in the pre-adjustment stage. Since the characteristics of the series may change over time, the specifications automatically selected by X-13ARIMA-SEATS should be reviewed periodically. This paper proposes a model selection procedure based on a restricted search over ARIMA specifications and compares it with the automatic procedure implemented in the seas() function in R. Model selection is initially based on the corrected Akaike information criterion (AICc) and subsequently on a hierarchy of diagnostics that includes residual seasonality (QS), the quality of the seasonal adjustment (Q(M)), the M_i statistics, residual normality, and model parsimony. The results show that the proposed procedure generally selects models with lower AICc values, whereas the automatic procedure tends to select more parsimonious specifications. However, differences in the quality of the seasonal adjustment are small in most cases. The results suggest that combining both procedures makes it possible to exploit their respective advantages and provides a suitable strategy for the systematic review of the models used in the seasonal adjustment of the EMAE activity branches.