Forecasting Australian unemployment rates using spectral analysis

Abstract

Univariate spectral analysis is used to model seasonally unadjusted quarterly unemployment rate data for Australia, 1978(2) to 2002(3). Data are tested for three categories: persons, males and females. Dynamic out-of-sample forecasts are made for 8 quarters using spectral analysis models evaluated against ARIMA model counterparts. It is found that the spectral analysis models achieve higher levels of forecasting accuracy than ARIMA counterparts, including turning point forecast accuracy. These resuIts emerge in spite of weaker in-sample explanatory power of the spectral models against the ARIMA models. It is concluded the results suggest that the spectral model is ultimately better attuned to the various cyclical forces of the past unfolding into the future.