Forecasting crude oil price (revisited)

The recent changes in crude oil price behaviour between 2007 and 2009 revived the question about the underlying dynamics governing crude oil prices. Even more importantly, the outstanding question over whether we can forecast crude oil price and returns or not needs to be readdressed. The goal of this paper is to present an analysis of crude oil spot daily price/returns. The aim is to find if structural changes in the crude oil market have had an effect on the ability to forecast daily returns. Also, we argue that there is still a gap between computational methods and traditional statistical methods for time series forecasting; hence, in this paper we try to make an effort to give due consideration to the statistical properties of the time series in the building process of softcomputing models. As such, our investigation starts with testing for non-linearity in the structure of these series using the most trusted test for iid, the BDS test. The Fuzzy Classifier System for non-linearity (FCS) proposed by Kaboudan (1999) and a time-domain test for non-linearity introduced by Barnett and Wolff (2005) are also used. Finally, we estimate the Lyapunov exponents to investigate the existence of chaos in crude oil price and return. Our tests show consistently over time that the dynamical forces driving crude oil price and returns are non-linear ones, of possibly low dimension. Moreover, the FCS test shows evidence of high level of noise which means that smoothing or noise reduction is necessary for achieving any level of forecast accuracy. To forecast the short-term of crude oil spot returns we compared the performance of ARIMA(p,d,q), EGARCH(p,q) and ANN models. We conclude that it is possible to forecast crude oil price using non-linear models providing noise control measures are used. Our results also show some evidence of effective multi-step forecasting (up 26 steps) for smoothed daily returns.