Download
Abstract
There is an increasing interest in applying variational Bayes techniques to estimating large Bayesian vector autoregressions (VARs) with stochastic volatility. However, less attention has been paid to the development of appropriate tools for comparing these high-dimensional models, especially among those designed to address COVID-19 outliers. This paper develops a marginal likelihood estimator that combines importance sampling and variational approximation for comparing large VARs with different time-varying volatility specifications and outlier adjustments. Through a Monte Carlo study, we show that the proposed approach is fast and able to identify the correct models. The effectiveness of the proposed method is further illustrated through an empirical application of comparing a variety of 180-variable VARs.