Abstract:
|
Virtually all seasonal adjustment programs include a variety of statistical tests for assessing whether a given time series is a candidate for seasonal adjustment. However, any ensemble of seasonality tests is certain to be either consistent or inconsistent. In the former case, the tests arrive at the same decision, raising the question if they are redundant. In the latter case, the tests arrive at different decisions, raising the question if there is a method that is capable of identifying the most informative tests and finding a final decision. Using simulated seasonal and non-seasonal ARIMA models that are representative of the Deutsche Bundesbank's time series database, we use random forests of conditional inference trees in order to quantify the informational content of the seasonality test implemented in JDemetra+ 2.1. We find that the modified QS- and the Friedman-test yield by far the most informative results. Finally, we discuss a conceptual design of a combined seasonality test.
|