Abstract:
The COVID-19 outbreak in 2020 has fostered in many countries the development of new weekly economic indices for the timely tracking of pandemic-related turmoils and other forms of rapid economic changes. Such indices often utilise information from daily and weekly economic time series that normally exhibit complex forms of seasonal behaviour. The latter dynamics were initially removed with ad hoc or experimental methods due to the urgent need of instant results and hence the lack of time for inventing and approving more sophisticated alternatives. This, never- theless, has in turn inspired recent developments of seasonal adjustment methods tailored to the specifics of infra-monthly time series. Although sound theoretical descriptions of these tailored methods are already available, their performance has not been evaluated empirically in great detail so far. To fill this gap, we consider real-time data vintages of several infra-monthly economic time series for Germany and analyse the cross-vintage stability of holiday-related deterministic pretreatment effects as well as the revisions in various concurrent signal estimates obtained with experimental STL-based and selected elaborate methods, such as the extended ARIMA model-based and X-11 approaches. Our main findings are that the tai- lored methods tend to outperform the experimental ones in terms of computational speed, that the considered pretreatment routines yield generally stable parameter estimates across data vintages, and that the extended ARIMA model-based approach generates the smallest and least volatile revisions in many cases.