Forecasting industrial production using structural time series models

Industrial production data series are volatile and often also cyclical, Hence, univariate time series models which allow for these features are expected to generate relatively accurate forecasts of industrial production. A particular class of unobservable components models - structural time series models - is used to generate forecasts of Austrian and German industrial production. A widely applied ARIMA model is used as a baseline for comparison, The empirical results show that the basic structural model generates more accurate forecasts than the ARIMA model when accuracy is measured in terms of size of error or directional change; and that the basic structural model forecasts better than the structural model with a cyclical component included on the basis of numerical measures, and tracking error for month-to-month changes. (C) 1998 Elsevier Science Ltd. All rights reserved.