A1858
Title: Partial identification of heteroskedastic structural VARs: Theory and Bayesian inference
Authors: Fei Shang - The University of Sydney (Australia) [presenting]
Abstract: Structural vector autoregressive models are proposed in which the structural parameters are identified via a stochastic volatility process for time-varying conditional variances. Our focus is on the question of how many and what shocks are identified via heteroskedasticity. Therefore, we derive a set of parametric restrictions under which the structural matrix is partially or globally unique, and Savage-Dickey density ratios are used to assess the validity of the identification conditions. We propose a shrinkage prior distribution for conditional log-volatilities and variances that is centred on a hypothesis of homoskedasticity, which assures that the evidence for the identification of the structural shocks is provided by the data. We apply identification through heteroskedasticity to estimate the dynamic output effects of unanticipated changes in tax policy that have been identified in previous studies by exclusion restrictions as well as by using narrative measures as proxies or time-varying volatility.