The VEC-NAR model for short-term forecasting of oil prices

Abstract

The prediction of future crude oil prices is highly challenging due to three characteristics of crude oil prices, namely, their lag, nonlinearity, and interrelationship among different oil markets, which cannot be handled simultaneously by most traditional crude oil price forecasting models. This paper proposes a new hybrid vector error correction and nonlinear autoregressive neural network (VEC-NAR) model to deal with these characteristics simultaneously. Firstly, a VEC model is used to optimize the lag of crude oil prices and determine the interrelationship which distinguishes the endogenous and exogenous variables. Then, the optimal results obtained by the VEC model are combined with a NAR model which effectively depicts nonlinear component, to forecast crude oil prices. The data of Brent oil prices from January 1, 2003 to December 31, 2014 were used as the empirical sample to test the effectiveness of our proposed model which is compared with those well-recognized methods for crude oil price forecasting. The results of Diebold-Mariano test demonstrated that the VEC-NAR model provided superior forecasting accuracy to traditional models such as GARCH class models, VAR, VEC and NAR model in multi-step ahead short-term forecast.