ABSTRACT
This chapter discusses some multivariate forecasting procedures, focusing particularly on the use of multiple regression models. It looks at some multivariate time-series models, giving particular attention to vector autoregressive (VAR) models. The enormous improvement in computing capability has made it much easier to fit a given multivariate model from a computational point of view. The chapter discusses the main time-series alternative to econometric simultaneous equation models is VAR models. With multivariate time-series data, the modelling process is complicated by the need to model the serial dependence within each series, as well as the interdependence between series. A time plot for each variable will indicate the presence of trend, seasonality, outliers and discontinuities. For stationary series, it will also be helpful to calculate the autocorrelation function for each series in order to suggest an appropriate univariate model for each series.
