Data-driven selection criteria for X-13ARIMA-SEATS seasonal adjustment algorithms

• Activity Number: 24
• Type: Topic Contributed
• Date/Time: Sunday, August 4, 2013 : 2:00 PM to 3:50 PM
• Sponsor: Business and Economic Statistics Section
• Author(s): Karsten Webel*+
• Companies: Deutsche Bundesbank
• Keywords: ARIMA model-based approach ; signal extraction ; unobserved components ; X-11 approach

Abstract

When choosing a software package for conducting seasonal adjustment as part of their daily routines, decisions of many statistical agencies are based on pragmatic reasons. Then, usually all time series, or at least broad subsets thereof, are seasonally adjusted according to the approach implemented in the chosen software package. Recent releases of X-13ARIMA-SEATS and (J)Demetra+ may change habits as these programs include both a combination of the nonparametric X-11 approach with ARIMA forecasts and the parametric SEATS approach. Hence, users may choose between both methods for each particular time series under review. Accordingly, the question naturally arises which criteria this selection should be based on. We develop a decision tree that relies on both theoretical and empirical issues. In particular, the latter include visual inspection of squared gains of final X-11 and SEATS seasonal adjustment filters as well as calculation of diverse revision measures, which is illustrated using German turnover data. Thereby, we also show that running the SEATS algorithm in X-13ARIMA-SEATS with default options may lead to results that are somewhat misleading.