Title: On conditional dynamic skewness and directional forecast of currency exchange rates
Authors: Julien Hambuckers - University of Liege - HEC Liège (Belgium) [presenting]
Abstract: A GARCH-in-mean model is formulated where the innovations follow a non-Gaussian sinh-arcsinh distribution, with time varying skewness parameter. The structural equation of this parameter is assumed to be of the ARMA-X type, allowing for an effect of past stochastic shocks, autoregressive terms and interest rate differential on conditional skewness dynamic. This model is used to predict the direction of change of three major currency pairs (USD/EUR, USD/GBP and USD/CHF) over the period 1999-2016. We find a dynamic skewness specification to be highly significant and able to improve the correct classification rate between up and down periods. Robust bootstrap tests support these findings. Economically speaking, we show that, notwithstanding a better in-sample performance compared to benchmark models for all pairs, we are able to obtain a better and positive out-of-sample performance for the USD/CHF pair.