Ensemble probabilistic prediction approach for modeling uncertainty in crude oil price

Abstract

The quantification of the uncertainty in crude oil price is of significance to improve the related financial decision-making. However, studies in this field have remained limited because the nonlinearity inherent in the crude oil price makes it challenging to model its uncertainty. In this paper, a novel learning system of ensemble probabilistic prediction combining five popular machine learning methods and an improved optimizer is presented to effectively model the uncertainty in crude oil price and establish the corresponding prediction interval with satisfactory reliability and resolution. An improved grey wolf optimizer based on the adaptive Cuckoo search algorithm (AGWOCS) is proposed in the learning system to integrate the prediction intervals produced by the above machine learning methods. In addition, the superiority of the proposed AGWOCS is validated based on an algorithm test, compared to three benchmark optimizers. To validate the effectiveness of the proposed learning system, the uncertainties in daily and weekly Europe Brent spot prices are modeled as a case study. The evaluation results based on the reliability, resolution, and sharpness demonstrate that the proposed learning system can yield the prediction interval with a higher quality, which has distinct advantages over eight benchmarks as a whole. The convergence and scalability of the learning system are also investigated, which reveals its feasibility.