Density forecasting using Bayesian global vector autoregressions with stochastic volatility

Does joint modelling of the world economy pay off? Evaluating global forecasts from a Bayesian GVAR

“…They find that the SSVS prior, on which we concentrade in this paper, exhibit the best forecasting ability among a wide range of priors. Huber (2014) shows how a Bayesian GVAR model can be augmented to account for stochastic volatility. Finally, Dovern and Huber (2015) show, in a complementary paper, that the GVAR model used in the paper at hand yields better turning point prediction than country-specific time-series models.…”

“…They conclude that imposing a factor structure on the latent log-volatilities helps to improve the accuracy of the density forecasts at little additional costs in terms of computational demands. Huber (2014) proposes a GVAR model with a factor SV structure. He reports that allowing each country's volatility to be driven by a country-specific latent factor improves forecasts of GDP and short-term interest rates, while leading to no improvements for forecasts of inflation, real exchange rates and long-term interest rates.…”

“…Finally, we enhance the Bayesian estimation of GVAR models. We modify the Bayesian GVAR model with stochastic volatility as outlined by Huber (2014) and introduce a variant of the stochastic search variable selection (SSVS) prior. Using a restricted stochastic volatility model in the spirit of Carriero et al (2012) and a conjugate version of the SSVS prior leads to significant computational gains that are essential when estimating large-scale models such as the GVAR.…”

The shortage of safe assets in the US investment portfolio: Some international evidence

“…Our modeling approach is based on Feldkircher and Huber (2016) and Crespo Cuaresma, Feldkircher, and Huber (2016), who put forward a Bayesian variant of the GVAR model to alleviate the curse of dimension- Punzi and Kauko (2015). 4 See Caballero and Krishnamurthy (2008a) and Caballero and Krishnamurthy (2008b).…”

“…Using a global-local shrinkage prior in the spirit of Griffin and Brown (2005) and applied to the VAR case in Huber and Feldkircher (2016) allows us to shrink the highdimensional parameter space towards a simpler model specification. To cope with the fact that the volatility of macroeconomic shocks displayed pronounced movements over time, we adopt a stochastic volatility (SV) specification in the spirit of Cogley and Sargent (2005) and utilized in Huber (2016). To identify structural demand and supply shocks we follow Caballero, Farhi, and Gourinchas (2016) and use sign restrictions imposed on the US responses and the average international responses of certain key macroeconomic quantities.…”