Fast and accurate variational Bayesian methods for high-dimensional TVP-VAR models

HDR Candidate: Ni (Bobby) Ying

Milestone: Confirmation

Title: Fast and accurate variational Bayesian methods for high-dimensional TVP-VAR models

When: 9-11am, Wednesday 5 May 2021.

Where: Zoom - https://uqz.zoom.us/j/86568016519

Abstract:

This thesis explores variational Bayesian (VB) methods' applicability in a range of high-dimensional macroeconomic problems of interest (where the conventional MCMC methods are computationally challenging). The first chapter of this thesis will focus on applying a VB procedure with theoretical guarantees to impose sparsity (with a bi-level spike-and-slab prior) on the time-variation of macroeconomic series. The second chapter is interested in using a VB method (with a hierarchical shrinkage prior) to model high-dimensional cross-country macroeconomic datasets with unrestricted time-varying parameter Panel VARs (TVP-PVARs). The last chapter of this thesis will be focusing on performing model selection in large TVP-VARs with a VB model-selection criterion.