Title: Model-based seasonal adjustment in case of seasonal heteroskedasticity
Authors: Jean Palate - National Bank of Belgium (Belgium) [presenting]
Abstract: In many time series, the seasonal pattern presents specific characteristics following the period of the year. Typical examples are the higher variability of activity in winter or rapid changes in the production behaviour during summer. In the context of seasonal adjustment, a simple solution to deal with such seasonal heteroskedasticity consists in using the non-parametric X-11 method with specific filters for the different periods. However, decompositions based on periodic models offer a more powerful approach. Several model-based solutions are reviewed, with different strategies for identifying automatically parsimonious models. The methods were applied on large data sets and the outcomes of that large-scale study are discussed. Finally, some characteristics of the model-based approach are stressed through specific cases. All the estimations were done by means of JDemetra+, which is a free open source software.