Using a Latent Variables Representation to Estimate Structural VARs

: A general class of structural vector autoregressions (VAR) is considered containing most variants encountered in empirical practice. Employing a latent variables representation of the system, such models can conveniently be estimated using standard software packages like LISREL. In particular we consider the latent variables representation for structural forms of Johansen's cointegrated VAR. The approach is illustrated using the well known examples of Blanchard and Quah (1989) and King, Plosser, Stock and Watson (1991). Moreover, a latent variables counterpart of the causal graph model suggested by Swanson and Granger (1996) is given. 1 Introduction In recent years, structural versions of vector autoregressive (VAR) systems became increasingly popular in empirical macroeconomic research. Such models were developed to overcome the well known inference problem in reduced form modelling. Early versions of this approach adopt a system of simultaneous equations for the innovations of t..