Title: Heteroskedastic proxy-SVARs
Authors: Luca Fanelli - University of Bologna (Italy) [presenting]
Abstract: Identification strategies are discussed for Structural Vector Autoregressions (SVARs) which combine the use of external instruments, the so-called proxy-SVAR or SVAR-IV approach, with the heteroskedasticity found in the data, the so-called identification-via-heteroskedasticity. The focus is on the case in which $r$ valid instruments are used to identify $g$ (larger of equal than one) structural shocks of interest, with $r$ larger or equal than $g$, and there are $m$ structural breaks in the VAR error covariance matrix which give rise to $m+1$ volatility regimes. It is shown that the combination of the two approaches enhances the identification possibilities for practitioners and produces overidentified, testable models, denoted HP-SVARs. Two types of heteroskedasticity are considered. In one case, the structural breaks do not affect the on-impact coefficients hence the impulse response functions (IRFs) are constant across volatility regimes. In the other case, the structural breaks affect the on-impact coefficients and IRFs are regime-dependent. General identification results for HP-SVARs are derived for these two cases. Estimation can be carried out through maximum likelihood.