FORECASTING TIME-SERIES WITH COMMON SEASONAL PATTERNS

This paper proposes a methodology for modelling and forecasting series which possess common patterns at seasonal and/or other frequencies. The approach is in the Bayesian autoregression tradition originally developed by Litterman (1980), Doan, Litterman, and Sims (1984), and Sims (1989), and builds common patterns directly into the prior of the coefficients of the model by means of a set of uncertain linear restrictions. To gauge the usefulness of the approach, the procedure is applied to the problem of forecasting a small vector of national industrial production indices.