ABSTRACT
In this chapter we describe structural VAR models and how they can be constructed using the methods of graphical modeling. The form of the structural VAR (SVAR) that we will use differs primarily from that of the canonical VAR model by including regression terms for the dependence between current variables in addition to dependence of current on past variables. These terms explain any correlation between the innovations of the canonical VAR so that the residual series, or structural model innovations, from the SVAR model will be uncorrelated with each other. Our aim is also that structural models be sparse, in the sense that they represent the dependence using a relatively small number of model terms or coefficients. It is reasonable to believe that in a model that reflects the true structure of a system of inter-related series, the current value of each series should depend on a relatively small number of other current and lagged values.
