Energy prices forecasting using nonlinear univariate models

This study analyses whether nonlinear methods are powerful enough to outperform consistently the no-change forecast for prices of key energy commodities, i.e. Brent crude oil, WTI crude oil, natural gas and coal. Six classes of nonlinear models are tested: threshold models (both self-exciting and external threshold variable model approach), smooth transition models (self-exciting and external threshold variable model approach), Markov regime switching models and neural networks. The forecasting competition is designed to simulate a real-time forecasting scheme. The analysis provides some evidence for predictive capabilities of nonlinear methods, but only in short-term horizons.