Title: Exchange rate predictive densities: An application of stochastic model specification search
Authors: Anthony Garratt - University of Warwick (United Kingdom)
Emi Mise - University of Leicester (United Kingdom)
Emi Mise - University of Leicester (United Kingdom) [presenting]
Abstract: The predictive densities for exchange rates computed from TVP-VECMs incorporating stochastic volatility (SV) are evaluated. The long-run components of the VECMs consider well-known sets of fundamentals-based models of exchange rates, including purchasing power parity, uncovered interest parity, monetary fundamentals as well as Taylor rules. We compute out-of-sample predictive densities by Bayesian model averaging using the posterior model likelihood as model weights. In order to address the well-documented problem of over-parameterisation leading to poor predictive performance, a prior shrinkage procedure is adopted. Various methods of prior shrinkage have been proposed in the literature. We proceed by comparing the predictive densities of TVP-VECM with SV, with and without the application of the shrinkage method. This approach permits a significantly more parsimonious model in which some of the parameters are restricted not to vary over time. In our application, monthly data for U.K. Sterling against U.S. dollar, over the 40-year period after the Bretton-Woods system is used. Our forecast evaluation uses both statistical and economic criteria.