Asymptotic normal and bootstrap inference in structural VAR analysis

The aim of the paper is to examine the performance of bootstrap and asymptotic parametric inference methods in structural VAR analysis. The results obtained through a Monte Carlo experiment suggest that the two approaches are largely equivalent in most, but not all, cases. While the asymptotic method turns out to be surprisingly robust with respect to the distribution of the errors, the bootstrap does deliver results superior in terms of both length of the confidence interval and coverage when highly non-linear statistics (such as the components of the variance of the forecast error) are considered.