This research paper introduces a new approach called DGDNN for predicting stock movement. Traditional methods often ignore the interdependencies between stocks and the hierarchical structure of the stocks themselves. The DGDNN method addresses these issues by automatically constructing dynamic stock graphs using entropy-driven edge generation. It further refines these graphs through a generalized graph diffusion process. The method also employs a decoupled representation learning scheme to capture unique hierarchical features within stocks. The experimental results showed significant improvements over existing methods. The study further demonstrates the effectiveness of the DGDNN method in modeling the time-evolving dynamics between and within stocks.
Publication date: 4 Jan 2024
Project Page: not provided
Paper: https://arxiv.org/pdf/2401.01846
