Revisions in concurrent seasonal adjustments of daily and weekly economic time series

The COVID-19 pandemic outbreak in 2020 has fostered in many countries the development of weekly economic indices for the timely tracking of rapid economic changes. Such indices typically utilise information from daily and weekly data that often exhibit complex seasonal dynamics which were initially removed on experimental grounds due to the urgent need for instant results and hence the lack of time for devising more elaborate approaches. Nevertheless, several methodological advances in the seasonal adjustment of infra-monthly data have been made since that time. Although sound theoretical descriptions are already available, those advanced methods have not been evaluated empirically in great detail so far. To fill this gap, we analyse the revisions in various concurrent signal estimates obtained with an experimental STL-based method and selected elaborate approaches, such as extensions of the popular ARIMA model-based and X-11 approaches. Using daily and weekly real-time economic time series for Germany, the cross-vintage stability of estimated regression parameters required for data pretreatment is also assessed. Our main findings are that the elaborate methods are often computationally faster, that the considered pretreatment routines yield generally stable parameter estimates, and that the extended ARIMA model-based approach often generates the smallest and least volatile revisions.