Identifying Shocks via Time-Varying Volatility

Under specific parametric assumptions, an n−variable structural vector auto-regression (SVAR) can be identified (up to n! shock orderings) via heteroskedasticity of the structural shocks (Rigobon, 2003, Sentana & Fiorentini, 2001). I show that misspecification of the heteroskedasticity process can bias results derived from these identification schemes. I propose a different method that identifies the SVAR up to n! shock orderings using only moment equations implied by a stochastic process for the variance. Unlike previous work, this result requires only weak technical conditions. In particular, it requires neither parametric assumptions nor mutual independence of the shocks. I propose intuitive criteria to select among the orderings and show that this selection does not impact inference asymptotically. As an empirical exercise, I apply this method to Bernanke, Boivin, & Eliasz (2005) and reject their Cholesky assumptions on the structural shocks. However, I confirm their conclusion of monetary non-neutrality and offer several extensions.