A0308
Title: Semiparametric forecasting using non-Gaussian ARMA models based on s-vines
Authors: Jialing Han - University of York (United Kingdom) [presenting]
Alexander Alexander John McNeil - University of York (United Kingdom)
Alexandra Dias - University of York (United Kingdom)
Martin Bladt - University of Copenhagen (Denmark)
Abstract: A semiparametric method for forecasting time series based on the s-vine copula approach for stationary time series developed recently is proposed. By combining a parametric s-vine process to describe serial dependence with a nonparametric model of the marginal distribution, the method offers improved modelling and forecasting for time series that have a non-Gaussian distribution and a nonlinear dependence on past values. The methodology gives a clear meaning to the concept of a non-Gaussian autoregression moving average (ARMA) model in which a parametric object known as the Kendall partial autocorrelation function plays the central role. To demonstrate the potential forecasting gains that can be obtained by using nonGaussian models, an approach to comparing distributional forecasts is applied. The methodology is illustrated with an application to forecasting the force of inflation in the US.
