DP15529 Common Component Structural VARs
Structural VAR models produce results that can vary dramatically with the
choice of variables, because information is deficient and/or contaminated by
measured errors. We propose a novel procedure, the Common Component
SVAR (CC-SVAR), which solves both problems. First, the common components
of the variables of interest are estimated using High-Dimensional
Factor techniques. Second, SVAR analysis is performed using such components.
The key feature is that number of common components is larger
than the number of shocks, so that the SVAR is singular. Consistency results
for singular VARs are provided. We apply our procedure to monetary
policy shocks. Our finding is that, with the CC-SVAR, results are robust
and SVAR puzzles disappear.