An Alternative Approach to Modeling and Forecasting Seasonal Time Series

Abstract

This article proposes an alternative methodology for modeling and forecasting seasonal series. The approach is in the Bayesian autoregression tradition pioneered by Doan, Litterman, and Sims and builds seasonality directly into the prior of the coefficients of the model by means of a set of uncertain linear restrictions. As an illustration, the method is applied to 10 U.S. quarterly macroeconomic series. For each series, I compare the forecasting performance of a univariate time-varying autoregressive model with seasonality built in the prior of the coefficients with five other widely used models.