Seasonality, Nonstationarity and the Structural Forecasting of the Index of Industrial Production

Summary

In this paper we focus on two STS models suitable for forecasting the index of industrial production. The first model requires that the index be transformed with a first and seasonal difference filters. The second model considers the index in its second difference filter, while seasonality is modeled with a constant and seasonal dummy variables. Tests designed to discriminate empirically between these two models are also conducted. Our results prefer the performance of the second model, particularly when the conventional ML estimation procedure is replaced by the ALS procedure. This process together with appropriate seasonal adjustment advances the possibility of using the suggested index forecasts to help to predict business cycle turning points.

Key words

The authors express their gratitude to Professors A. C. Harvey (University of Cambridge, UK), S. J. Koopman (Tilburg University, The Netherlands), J. Goldstein (Bowdoin College, USA), D. Bolduc (University of Laval, Canada), Dr. C. C. A. Winder (De Nederlandsche Bank NV, Amsterdam, The Netherlands), and all the participants to the 39^th Canadian Economic Association Meeting in 1999 in Hull (Quebec, Canada) for helpful discussions and suggestions. They also wish to acknowledge the helpful comments and detailed and valuable reports on an earlier version of this paper by two anonymous referees and an editor. The usual disclaimer applies.