RDP 2022-04: The Unit-effect Normalisation in Set-identified Structural Vector Autoregressions Appendix D: Additional Empirical Results

Figure D1 presents additional results obtained under Restriction (2) when the horizon over which the sign restrictions are imposed (H) is varied. For H = 2, the identified set for ${\eta }_{1,1,0}$ includes zero in only 1.2 per cent of draws from the posterior. However, the set of posterior medians and 68 per cent robust credible intervals for the output response are extremely wide. Increasing H reduces the proportion of draws where the identified set for ${\eta }_{1,1,0}$ includes zero: for H = 5 (the assumption in the main text), the identified set includes zero in 0.6 per cent of draws; and for $H\in \left\{11,23\right\},$ there are no draws where the identified set includes zero. Increasing the number of sign restrictions also appreciably tightens the set of posterior medians and robust credible intervals.