Capturing macroeconomic tail risks with Bayesian vector autoregressions

Event Date: 13 November 2019

Speaker:  Massimiliano Marcellino, Bocconi University

 Date:  13 November 2019 / Time:  4.15 pm / Location:  CW506b

A rapidly growing body of research has examined tail risks in macroeconomic outcomes.  Most of this work has focused on the risks of significant declines in GDP, and relied on quantile regression methods to estimate tail  risks. In this paper we examine the ability of Bayesian VARs (BVARs) with stochastic volatility (SV) to capture asymmetries and tail risks in macroeconomic forecast distributions and outcomes. We find, first, that the evidence of skewness in output growth is not all that strong, statistically speaking. Second, with our BVAR specifications featuring time-varying volatility, we are able to capture the same kind of distributional asymmetries in the predictive  distributions of output growth as resulting from quantile regression with, in addition, some gains in standard point and density forecasts. Finally, while the BVAR results for asymmetries are particularly evident when using a common volatility factor that is a function of past financial conditions, they also emerge with a conventional stochastic volatility specification. Monte Carlo experiments indicate that this finding stems from the flexibility of the BVAR-SV to allow periods of correlation between shocks to the levels of variables and their volatilities.

Published: 7 November 2019