CMStatistics 2022: Start Registration
View Submission - CFE
A1631
Title: Learning based uncertainty quantification in set identified SVARs Authors:  Tilmann Haertl - University of Konstanz (Germany) [presenting]
Abstract: Inference in set-identified structural vector autoregressions requires capturing both the uncertainty about the structural models in the identified set as well as the estimation uncertainty regarding the reduced form parameters of the model. Learning the identified set based on a bootstrapped set of reduced form parameters enables uncertainty quantification regarding the structural parameters directly via the learning guarantees of the learning algorithm. This approach is silent with regard to the underlying identification strategy or combinations of identification strategies and is not computationally costly. Hence, it is also applicable for cases in which conventional inference is potentially not worked out yet or tedious and offers an alternative way to quantify the uncertainty inherent in the identification. Simulations for the structural impulse responses show that the learned sets yield informative value for the structural parameters.