Zum Problem der Saisonbereinigung ökonomischer Zeitreihen

Summary

In this paper various aspects of seasonal adjustment of economic time series are discussed. Firstly, criteria relating to the adequacy of adjustment procedures based on spectral analysis of stationary stochastic processes are reviewed critically. In this context, some of the well-known adjustment procedures are critizised from a basic point of view: in general, they mostly lean heavily on the theory of stationary stochastic processes if a well-defined theoretical background can be detected at all. However, this very theory is of questionable value only in analyzing or adjusting especially economic time series since it implies a seasonal component of constant variance. Since no real seasonal component is of constant variance, adjustment procedures based on this theory have to be manipulated in various tricky ways so as to meet the fact of non-constancy of the variance. But this leads to a state of affairs which is unsatisfactory both from a practical and a theoretical point of view.

The main purpose of the paper consists in the presentation of an adjustment procedure which avoids the defects mentioned above. It is formulated in terms of the theory of non-stationary stochastic processes where “non-stationarity” is defined suitably so as to allow a changing seasonal pattern of non-constant variance. Besides overcoming in this way the restrictive assumption of stationarity, the proposed procedure allows the estimation of the seasonal component directly, i. e. without estimating a smooth component in a first step. This is an important advantage of this procedure over the conventional procedures which is of a highly practical relevance. A practical example is given.