Crude Oil Price Prediction

Abstract

In this paper, we propose linear and nonparametric models to predict one month, three months, six months, one year, eighteen months and two years ahead crude oil price in out-of-sample background. Mainly, our forecast depends on three predictor variables, the change in crude oil inventories, its previous prices and product spread. By employing mean-squared prediction error (MSPE) and stochastic dominance (SD) tests, we find that the prediction result of our nonparametric models is significantly better than the random walk model, while the corresponding linear models' performance is better than the random walk model only for longer horizon forecasts (one to two years). In General, for the sample period from 1995.1 to 2015.4, the conclusion is that our model applying nonparametric estimation always outperforms all other models in different horizon forecasting. And for the nonparametric model including all three predictors, we document MSPE reduction as high as 62.6% and directional accuracy ratio as high as 77.5% at the two years horizon compared to the random walk model.