The influence of changes in model on seasonal adjusted data

Abstract

According to recommendations of EUROSTAT on Seasonal Adjustment, frequency of SA revisions of the model and its parameters request to be reduced for a period of up to one year. In the meantime, the model and the parameters should be kept fixed. Model changing or recalculation of its parameters sometimes inescapable for real time series. Possible reasons: available new observations, methodology or basic year changing, other reasons. It was analysed seasonal adjusted statistical date quality from model revision point, or in time series revision case in this work. There was analysed stability of selected SA model on real and modeled time series. The influence of new observation on ARIMA model was tested and three ways of behavior after appearance of new data are singled out: - leaving the same model with it's previous parameters, - re-estimation of model parameters, - choosing a new model. Influence of changes in model parameters on seasonally adjusted data was analysed. There were estimated differences between SA data before and after model revision. The relation between Seasonal Adjustment Quality Index and new time series observation was detected. Stability of models dependence, when changed time series irregular component variance, or seasonal component, or new outlier detected, were also studied in this work.