Title: Structured regularization of high-dimensional panel vector autoregressions
Authors: Stephan Smeekes - Maastricht University (Netherlands) [presenting]
Lenard Lieb - Maastricht University (Netherlands)
Mathias Staudigl - Maastricht University (Netherlands)
Abstract: Inference is investigated on large panel vector autoregressions using regularization techniques. Rather than relying on standard sparsity assumptions, our regularization makes use of structured sparsity that exists naturally by similarity of the dynamics within and between groups in the panel. We explicitly allow for spillover effects between groups as our estimation method endogenously provides a measure of the strength of the (dynamic) interconnectedness between the groups. We develop efficient algorithms for estimation and explore the consistency, oracle and inferential properties of our estimator for plausible economic models, while finite sample properties are investigated in a simulation study. Finally, the generality and interpretability of our method is illustrated by an application to the analysis of fiscal policy in a multi-country macroeconomic panel dataset.