Title: Real-time macroeconomic forecasting with a heteroskedastic inversion copula
Authors: Michael Smith - University of Melbourne (Australia) [presenting]
Abstract: Accounting for asymmetries in the forecast densities of macroeconomic variables can improve their accuracy. In multivariate time series, this can be achieved by using a copula to capture both serial and cross-sectional dependence, allowing the margins to be modeled directly as nonparametric. Yet most existing high-dimensional copulas cannot capture heteroskedasticity. To do so, we propose a new copula created by the inversion of a multivariate unobserved component stochastic volatility model, and show how to estimate it using Bayesian methods. We study its real-time forecasts and their accuracy for four quarterly U.S. macroeconomic variables. The copula model captures heteroskedasticity, dependence in the level, time-varying asymmetry and heavy tails, bounds on the variables and other features. Over the window 1975Q1 -- 2016Q3, the point and density forecasts out-perform those from a range of benchmark models, particularly for the nowcast of GDP growth.