ABSTRACT
Forecasting the future values of an observed time series is an important problem in many areas, including economics, production planning, sales forecasting and stock control. Forecasting methods may be broadly classified into three groups: subjective, univariate and multivariate. The chapter focuses on calculating point forecasts, where the forecast for a particular future time period consists of a single number. It provides a brief introduction to some multivariate forecasting procedures. For long-term forecasting of non-seasonal data, it is often useful to fit a trend curve to successive values and then extrapolate. The main stages in setting up a Box-Jenkins forecasting model are: model identification, estimation, diagnostic checking, and consideration of alternative models. Point forecasts are adequate for many purposes, but a prediction interval is often helpful to give a better indication of future uncertainty. A prediction interval consists of upper and lower limits between which a future value is expected to lie with a prescribed probability.
