Abstract:
Official statistics routinely employs the X-13-ARIMA method to seasonally adjust economic time series. A key step is choosing the length of the seasonal moving av- erage. Traditionally, this choice relies on ad hoc criteria and expert judgement. We propose a cross-validation-based filter selection scheme that offers greater flexibility, including the possibility of incorporating novel filters. This approach is particularly promising for the seasonal adjustment of weekly, daily, and high-frequency time series. We demonstrate how to integrate cross-validation into the X-13-ARIMA method and discuss the advantages of various implementation options. Evaluation on monthly and quarterly time series demonstrates that this selection method performs at least as well as, and often better than, conventional selection criteria.
