Forecasting crude oil prices: A Gated Recurrent Unit-based nonlinear Granger Causality model

Abstract

Crude oil prices substantially impact the global economy, national security, and international politics. Accurate forecasting of crude oil prices and understanding the causality mechanism are crucial for policymakers, investors, and academia. Given the multitude of influence factors and oil prices’ complex nonlinearity, we propose a novel Gated Recurrent Unit-based non-linear Granger Causality (GRU-GC) model to provide precise oil price forecasting and causality determination. First, we collect numerous potential influence factors, including supply and consumption, global economic development, financial factors, and the energy market. We then harness the powerful predictive capabilities of the GRU, coupled with the insightful interpretation offered by the linear Granger model. This enables point prediction, probability density estimation, and causal inference simultaneously. The experimental results on the Brent and West Texas Intermediate (WTI) oil data sets show a significant superiority in multi-step-ahead forecasting. Furthermore, the unexpected variability of oil prices spurred by the COVID-19 pandemic validates the need and efficacy of integrating a probability density function for robust prediction. Additionally, our model reveals the non-linear Granger causality that differentiates Brent and WTI oil prices.