ABSTRACT
This chapter discusses differences between Vector autoregressive (VAR) models and simultaneous equations models and importance of the lag order determination for VAR models. It describes use of VAR models for Granger-causality test, impulse-response analysis, variance decomposition analysis and forecasting. VAR models can be viewed as unrestricted reduced-form equations of the correct but unknown structural models. Considering that structural models are often misspecified in defining exogenous variables, use of VAR models may be an alternative approach. Granger states that the causality has two components: the cause occurs before the effect; and the cause contains information about the effect that is unique, and is in no other variable. A shock to a variable in a VAR model directly affects the variable and is also transmitted to the other variables through the dynamic structure of a VAR model. An impulse-response function traces effects of a shock to one variable on the current and future values of all variables included in a VAR model.
