Subspace Shrinkage in Conjugate Bayesian Vector Autoregressions

TL;DR

This paper introduces a conjugate Bayesian VAR approach with a subspace shrinkage prior that combines VARs and factor models, improving factor detection and forecasting in macroeconomic data.

Contribution

It develops a novel conjugate prior that shrinks towards a factor model subspace, allowing estimation of factor number and shrinkage strength within Bayesian VARs.

Findings

Successfully detects the number of factors in simulations

Improves forecast accuracy in macroeconomic data

Provides a theoretically sound framework for combining VARs and factor models

Abstract

Macroeconomists using large datasets often face the choice of working with either a large Vector Autoregression (VAR) or a factor model. In this paper, we develop methods for combining the two using a subspace shrinkage prior. Subspace priors shrink towards a class of functions rather than directly forcing the parameters of a model towards some pre-specified location. We develop a conjugate VAR prior which shrinks towards the subspace which is defined by a factor model. Our approach allows for estimating the strength of the shrinkage as well as the number of factors. After establishing the theoretical properties of our proposed prior, we carry out simulations and apply it to US macroeconomic data. Using simulations we show that our framework successfully detects the number of factors. In a forecasting exercise involving a large macroeconomic data set we find that combining VARs with factor models using our prior can lead to forecast improvements.