Econometrics Seminar - Jamie Cross (BI Norwegian Business School)

Title: Large Bayesian vector autoregressions with stochastic volatility in mean

Abstract: Recent research has shown that incorporating volatility feedback effects and large datasets into vector autoregression (VAR) models are useful for macroeconomic forecasting and structural analysis. Despite this fact, computational complexities have made it challenging to estimate VARs with both of these features.  In this paper, we propose an efficient Bayesian posterior simulator for estimating large VARs that allow for stochastic volatility in mean (SVM) dynamics. We show that the new algorithm is more efficient (both computationally and statistically) compared to conventional particle-filter based algorithms for VAR-SVM models. In two empirical applications on the US economy, the large VAR-SVM model provides (1) novel macroeconomic insights about multi-sectored spillovers of uncertainty and (2) competitive out-of-sample forecasts to conventional large VAR models.