This paper compares the traditional Census X-11 method for seasonal adjustment with two recent alternative methods using ARIMA models, viz. X-11 ARIMA and Burman's signal extraction method. No strong preference results for one of these methods when applied to a number of macro-economic time series for the Netherlands. IntroductionThe comparative study of seasonal adjustment by Fase, Koning and Volgenant showed the Census X-11 method to be among the best of nine adjustment methods tested. This method has been used since by the Netherlands Bank for seasonal adjustment of a large number of macro-economic and financial data.The fast evolution of time series analysis in the last decade has now prompted a reconsideration of this choice. Box and Jenkins' ARIMA models provide an adequate description of seasonal movements in time series and, therefore, these models can be used in seasonal adjustment. First, they may be used to modify and, hopefully, improve the Census X-11 method. Second, it is conceivable to decompose a time series according to its ARIMA model into a seasonal and a non-seasonal component.This paper presents an empirical comparison between the Census X-11 method and two alternatives, viz. the X-11 ARIMA method developed by Dagum [1975] and Burman's signal extraction method. X-11 ARIMA is a Census X-11 modification, while Burman's method is based on decomposition of ARIMA models. The X-11 ARIMA method has already been tested extensively [see e.g. Dagum, 1978; Dagum/ Morry;Kuiper] and is used in practice, while the decomposition methods using ARIMA models are still on the eve of leaving the laboratory. The Burman method is considered here because this method was operational and its computer program available to us at the start of this research project.The Census X-11 method [see Shiskin/Young/Musgrave] separates the trend-cycle component of a time series from the random and seasonal components by repeated application of weighted moving averages. Census X-11 enjoys a worldwide popularity as a seasonal adjustment method mainly because of the wide applicability and flexibility with which shifts in the seasonal pattern can be described. A disadvantage, however, is that addition of fresh data often leads to a substantial revision of the most recent adjustments. This is due to the use of asymmetrical weights in the moving averages at the end of the observation period. A number of procedures have been designed Wallis, K.F.: Seasonal adjustment and revision of current data: linear filters for the X-I1 method.

Amsterdam, respectively. This study was completed when the first author was at the Nederlandsche Bank. We gratefully acknowledge the useful comments of a referee on a previous version of this article.