Printer Friendly Version | Back

Length of a Time Series for Seasonal Adjustment: Some Empirical Experiments

Year: 2011       Vol.: 60       No.: 1      

Authors: Lisa Grace S. Bersales

Abstract:

Use of 5 to 15 years of quarterly or monthly data is suggested when doing seasonal adjustment using X11 and its variants. This is meant to address changes in the structure of the time series. Philippine time series are good candidates for this practice since they usually exhibit frequent changes in patterns. Empirical validation of the suggested length of series is done for seasonal ARMA processes. Different quarterly series were simulated for the following situations and seasonal adjustment was done for various lengths of time series: (1) processes without any structural change; (2) processes with abrupt permanent change in structure; (3) processes with gradual permanent change in structure. For all types of processes, both weak and strong seasonality were considered. Regression models were used in testing the effect of length of series used in seasonal adjustment to the error in estimating the seasonal factor. Results show that the length of series used does not have significant effect on the seasonal adjustment for processes without structural change and with abrupt permanent structural change. On the other hand, for processes with gradual permanent change, use of longer lengths of series for seasonal adjustment is better.

Keywords: seasonal adjustment, seasonal factor, X11-ARIMA, seasonal ARMA processes

Download this article:

Back to top