Using transformer to predict the price of commodity

Abstract

In this study, we explored the use of the Transformer model for time serial prediction and applied it to commodity price prediction, particularly for specific commodities and gold prices in the market. Given the potential of deep learning in processing large-scale datasets and capturing complex nonlinear patterns, this study aims to explore the application effect of Transformer models in financial time series prediction. Firstly, we reviewed the relevant work in the field of time series prediction, particularly the application of deep learning methods such as recurrent neural networks (RNN), long short-term memory networks (LSTM), and Transformer models. Subsequently, this study improved the standard Transformer model to better adapt to the demand for time series prediction, especially in commodity price prediction. This dissertation first imitates the relevant research of Dr. Hachmi Ben Ameur and predicts the Bloomberg Commodity Index and its component indices: Bloomberg Industrial Metals and Bloomberg Precious Metals. Based on this, the price of gold in the market is predicted. The experimental results show that the model has achieved a high level of prediction accuracy, proving the effectiveness of the Transformer model in such tasks. Specifically, the prediction case of gold prices further validates the robustness and adaptability of the model in handling time series data of different types of commodity prices. This study not only demonstrates the potential of deep learning, especially Transformer models, in financial time series prediction but also provides valuable insights and potential methodological guidance for future research in similar fields.