Title:
Forecasting Changing Seasonal Components in German and US Unemployment using Periodic Correlations
Authors:
Philip Hans Franses
Marius Ooms

Abstract:
An important underlying idea in seasonal adjustment procedures is the presumed orthogonality of seasonal and nonseasonal components. In this paper we examine this orthogonality assumption for official quarterly unemployment figures in Germany and the USA. Although nonperiodic correlations do not seem to reject the orthogonality assumption, a periodic analysis based on correlation functions that vary with the seasons indicates the violation of orthogonality. For example, we find the correlation between the seasonal and nonseasonal components to be significantly positive in the first quarter and negative in other quarters. This implies that seasonally adjusted figures of the number of unemployed are likely to be ``too optimistic'' in the first quarter in times of rising unemployment. We find significant predictability of seasonal components when we base such forecasts on periodic autoregressions with unit roots. Similar results are obtained when we construct forecasts for seasonal components of so-called structural time series models [Harvey(1989)]. In addition to these empirical findings, we discuss the usefulness of periodic models with unit roots to assess the relative adequacy of additive and multiplicative models of (long-run) seasonality.
Keywords:
seasonal adjustment, periodicity, unit roots, predictability

• TI 95-77 Franses, P.H., and Ooms, M.(1995), ``Forecasting Changing Seasonal Components in German and US Unemployment using Periodic Correlations''
  • post-script article, 2401K,
  • zipped post-script article (.zip) and ascii abstract, 248K,