The future of time-series forecasting

Time-series forecasting methods may be classified into univariate (projection) methods and multivariate methods. The choice amongst methods depends on a variety of considerations including the objectives, the type of data, and whether an automatic or nonautomatic approach is to be used. The value of forecasting competitions is reviewed and suggestions are made regarding the forthcoming M2-competition. Various unsolved research problems relating to different univariate and multivariate methods are outlined. The difficulties of fitting multivariate models to economic data are discussed and it is suggested that univariate procedures will continue to have wide use in practice. The impact of proliferating computer software is considered and some desirable features of ‘good’ packages are listed.