Forecasting long term exchange rate volatility with stochastic mean-reverting unconditional volatility

Abstract

In this paper, an attempt is made to induce stochastic mean-reversion in the unconditional volatility forecasts from GARCH models by augmenting the models with proxies for levels of market uncertainty and exogenous break variables. Evidence from the results indicates that the augmentation leads to a reduction in information leakage, lower persistence of volatility shocks and improved predictive powers. In addition, it relaxed the restrictive mean-reversion property of GARCH forecasts so that the conditional forecasts revert toward the long-run variance along a non-stochastic path. Furthermore, there are general improvements in forecast accuracies as well as the explanatory powers of the forecasts but where the models are outperformed, the accuracy trade-offs are statistically insignificant in most cases. The approach is recommended as a stand-alone or as a supplementary tool for forecasting medium to long term volatility of exchange rates and other speculative assets