A data-driven selection of an appropriate seasonal adjustment approach

Recent releases of X-13ARIMA-SEATS and JDemetra+ enable their users to choose between the non-parametric X-11 and the parametric ARIMA model-based approach to seasonal adjustment for any given time series without the necessity of switching between different software packages. To ease the selection process, we develop a decision tree whose branches combine conceptual differences between the two methods with empirical issues. The latter primarily include a thorough inspection of the squared gains of final X-11 and Wiener-Kolmogorov seasonal adjustment filters as well as a comparison of various revision measures. We finally illustrate the decision tree on selected German macroeconomic time series.