Forecasting with Global Vector Autoregressive Models: a Bayesian Approach

What is it about?

A global vector autoregression (GVAR) is a model of the world economy that models economic and financial interactions among economies. It can be used to analyze the response of a country to a particular shock (say the global financial crisis) or forecasting. In this paper we propose estimating these models using so-called Bayesian shrinkage priors and show that these significantly improve out-of-sample forecasts.

Featured Image

Why is it important?

GVARs have become increasingly popular, not only for forecasting but also for structural (policy) analysis. Still, classical estimation is prone to issues of overfitting, and confidence bounds / credible sets of these models are typically large. Bayesian shrinkage priors can severely reduce estimation uncertainty and makes the model more attractive to a broad range of researchers