ABSTRACT
In its simplest form, a vector autoregression (VAR) is an unrestricted reduced-form model that expresses each variable as a linear function of a constant and the lags of that and each other variable in the system. Since each equation in a VAR has the same regressors, they can be estimated separately by OLS. However, even in moderately sized systems with, say, six variables and four lags of each, and a constant term, there are 25 parameters to be estimated in each equation so that over-parameterization has often been cited as the main cause of poor forecasting performance.
