The particular role of outlier detection and replacement in seasonal adjustment of a time series is to allow unbiased estimation of typical seasonal and calendar effects. Only if abrupt, strong and atypical changes are explicitly modelled, it is possible to estimate typical calendar influences and seasonal effects, which recur with similar intensity in the same season each year. The seasonal and calendar components are then eliminated from the unadjusted data in order to get the seasonally adjusted results.
It is most likely that problems will arise if outliers are not modelled. It may be the case, for example, that first estimates of seasonally adjusted figures will identify turning points too late and be misleading over a period of months – compared with the final results which are established years later. Furthermore, one year after wrongly not identifying outliers, problems may arise at the current end of the time series because parts of the one-off special effects have been ascribed incorrectly to the seasonal component as recurring each year. This would result in inaccurate movements of the seasonally adjusted series which have to be corrected later. In order to prevent such problems and the corresponding revisions, it is advisable to treat abrupt, strong and atypical events as outliers in the process of seasonal adjustment.
The reasoning behind this is illustrated using empirical examples such as Argentinean currency in circulation during and after the Argentine crisis in 2001-02 as well as German economic indicators since the beginning of the financial and economic crisis in 2008. It is shown that applying the Guidelines on Seasonal Adjustment of the European Statistical System helps to avoid flawed modelling of the seasonal and calendar components and, thus, of the seasonally adjusted results.
Keywords: Seasonal adjustment; Outlier
Biography: Head of Division General Economic, Capital Markets and Financial Statements Statistics of the Deutsche Bundesbank