Flexible Priors and Restrictions for Structural Vector Autoregressions
Francesca Loria, Christiane Baumeister and Junior Maih

This paper introduces an innovative approach for estimating Bayesian vector autoregressions (VAR) in structural form, enhancing flexibility in incorporating various priors and identification strategies. The method accommodates zero, sign, and narrative restrictions, as well as identification via proxy variables, offering a unified framework for replicating prominent strategies in the VAR literature. Unlike existing methods, it directly elicits informative priors and restrictions on structural parameters, ensuring transparency and avoiding unintended beliefs. The approach is versatile, scalable to larger models and eliminates the need for separate algorithms for different identification schemes. By employing endogenous priors, it overcomes issues related to non-uniform priors over identified sets, extending beyond impulse responses. Additionally, the methodology allows for imposing (in-)equality restrictions on VAR parameters, providing a robust means to incorporate strong beliefs. Overall, this user-friendly framework addresses key challenges in the current literature, offering a valuable tool for empirical researchers, with the method accessible through the RISE toolbox.