Topics in structural VAR econometrics (Q1189544)

This short monograph discusses the salient features of structural vector autoregressive models (VAR) in terms of both identification and estimation in the context of full information maximum likelihood (FIML) methods. Structural VAR analysis attempts to identify a set of independent shocks by means of a set of meaningful theoretical restrictions. The random shocks in this framework can be regarded as the ultimate source of stochastic variations of the vector of variables. Let \(y_ t\) be a set of \(n\) economic variables, jointly covariance stationary without any deterministic part and possessing a finite \(p\)-th order autoregressive representation as: \(A(L) y_ t=\varepsilon_ t\), where \(A(L)=I-A_ 1L-\dots-A_ pL^ p\), the roots of the equation \(\text{det}(A(L))=0\) are outside the unit circle and \(\varepsilon_ t\) has an independent multivariate normal distribution with zero mean and variance-covariance matrix \(\Sigma\). For the structural VAR model discussed in this monograph, three different ways of imposing structures into the model are considered: the \(K\)-model, the \(C\)-model and the \(AB\) model. The \(K\)-model introduces an invertible matrix \(K\) such that \(KA(L)y_ t=K\varepsilon_ t=e_ t\) where \(E(e_ te_ t')=I_ n\) and \(I_ n\) is the identity matrix. The \(C\)-model introduces an invertible matrix \(C\) such that \(\varepsilon_ t =Ce_ t\). The \(AB\) model introduces two invertible matrices \(A\) and \(B\) such that \(AA(L)y_ t=A\varepsilon_ t=Be_ t\) with \(E(e_ t)=0\) and \(E(e_ te_ t')=I_ n\). Estimation problems for these three types of SVAR models are discussed and extended to nonstationary contexts. This is followed by an empirical illustration of a five equation macroeconomic model based on Italian data.