Since their inception, VAR models have been at the centre of many controversies associated with econometric modelling. The recurring criticism throughout history is due to the model being a-theoretic. There is not an underlying theory which is attached to the model therefore it incorporates no prior information. Asteriou and Hall (2011) argue that since there are no initial restrictions on the models parameters, one could deduce that all variables have an effect on everything. However, statistical inference, namely, causality testing can be used to eliminate those variables which are deemed insignificant from the model. Another problem which can occur when using VAR is that the parameters will consume much of the degrees of freedom, insist Gujarati & Porter (2009). For example; if we had a 4 variable model and had chosen to use 8 lags, 32 lagged parameters will exist, which in addition to the constant would total 33. Should the sample size not be sufficiently large, the parameters will use too many degrees of freedom. This will lead to problems withthe estimation.
Finally, VAR cannot be used for policy analysis due to the lack of prior information and a-theoretic nature of the VAR model; therefore the results are incredibly problematic to interpret. However there are methods to overcome this problem. Supporters of the model suggest estimating the impulse response functions thus ascertaining the effects on the variables, should one of them be subject to a shock. There are some advantages to using VAR models. They are very simple to setup. Econometricians need not worry about which variables are endogenous and which are exogenous i.e. originating within or outside the model. Additionally the estimation is extremely simple; each equation can be estimated using the standard OLS method.