Crude Oil Prediction Based on Multi-Factor LSTM-Transformer Algorithm

Abstract

The precise prediction of global oil prices holds immense importance in maintaining economic stability and mitigating risks. This paper proposes a novel machine learning-based time series forecasting method using the combination of Long Short-Term Memory (LSTM) and Transformer algorithms. To address the low dimensionality of predictors for the Transformer and preserve the chronological order of data, LSTM networks with a configured time step are employed for transformation. For illustration and verification purposes, a multi-factor LSTM-Transformer model is used to predict the crude oil spot price (return) of West Texas Intermediate (WTI). The empirical results validate the efficacy and superiority of this approach over single-factor and linear models in forecasting oil prices, making it a promising method for time series forecasting.