Non-technical summary

Research Question

The COVID-19 pandemic outbreak in 2020 has stimulated the development of new weekly economic indices in many countries to track rapid economic changes more timely. The calculation of such indices often involves seasonal adjustments of daily and weekly time series, which were initially carried out with experimental methods due to the lack of time for constructing more sophisticated alternatives. Meanwhile, several new and more elaborate approaches for seasonally adjusting infra-monthly data have been suggested. Sound theoretical descriptions of these approaches are already available; however, their performance has not been evaluated empirically in great detail so far.

Contribution

To fill this gap, we consider real-time data vintages of several infra-monthly economic time series for Germany and analyse the revisions in various signal estimates obtained from concurrent seasonal adjustments with both experimental and more elaborate methods. The speed of convergence from preliminary to final revisions is also studied for selected signals in daily data. In addition, we assess computation times as well as the cross-vintage stability of estimated holiday-related deterministic effects and their standard errors.

Results

Our main findings are threefold. First, the elaborate methods are often significantly faster than the experimental ones. Second, despite some differences with respect to automatic outlier detection, the estimated holiday effects obtained from the elaborate and experimental methods are similarly stable across the considered real-time data vintages. Third, one particular model-based elaborate method tends to generate lower and less volatile revisions in most signal estimates than any other method considered; on the flip side, however, those revisions may occasionally converge significantly slower.