The Effects of Seasonal Adjustment in Econometric Models

The effects of seasonal adjustment in econometric models has received little attention in the literature, despite the fact that deseasonalised data, adjusted by various means, are widely used. This paper provides a valuable insight into the relative effects of three adjustment procedures – the X-11 method, the Durbin-Wymer DSEAS method, and the seasonal dummy variable approach. A basic problem in studying seasonality is that the true seasonal pattern is not observable. This is overcome by using an econometric model and data whose structures are completely prespecified, so that experiments can measure exactly the effects of adjustment procedures. This is the basis of the Monte Carlo method used here. The model and data chosen are such as to give the results wie applicability. The first part of the study compares the procedures of X-11 and DSEAS in terms of their ability to deseasonalise a single economic series. It shows that the errors involved in producing a seasonally adjusted series are relatively small for both methods. If the seasonal pattern changes gradually over time, then the X-11 method gives smaller errors than DSEAS. The second aspect of the study considers the effects of adjustment on parameter estimates of an econometric model. X-11 and DSEAS both adjust data prior to their inclusion in the model. On the other hand, the dummy variable approach uses constructed seasonal variables in the equations of the model to capture the effects of seasonality. The experiments show that, for the given model, the dummy variable approach is best, while there is little difference between the use of X-11 or DSEAS.