Research on crude oil futures price forecasting based on error correction and deep reinforcement learning

Accurate forecasting of crude oil prices has always been the focus of government management decision makers, investment entities and academics. However, due to the interaction of various risk factors such as monetary policy and geopolitics, crude oil price exhibit more complex nonlinear characteristics, making crude oil price forecasting an unprecedented challenge. Our paper conducts an empirical study on INE and WTI crude oil futures markets through a crude oil price forecasting model (PVMD-QSBT-ECS) conducted based on data decomposition, reinforcement learning integration strategy and error correction technology. Firstly, the crude oil futures price series are decomposed by variational mode decomposition (VMD) using particle swarm optimization (PSO) via adaptive weights; Secondly, the Q-learning (QL) algorithm is utilized to determine the optimal weight combination of stacked bidirectional long short-term memory (SBiLSTM), bidirectional gated recurrent unit (BiGRU) and temporal convolutional network (TCN) to build an integrated prediction model, and then the dynamic error correction is applied to the prediction outcomes. Finally, the modified Diebold and Mariano (M-DM) test was utilized to further evaluate the forecasting performance of PVMD-QSBT-ECS. The empirical results indicate that the PVMD-QSBT-ECS model proposed in this paper not only has lower prediction error than other comparative models, but also exhibits superior performance in both emerging and developed markets, and also has obvious advantages in forecasting at different step sizes. © 2023 Systems Engineering Society of China. All rights reserved.