Integrating the Industrial Chain Information Using Heterogeneous Graph for Portfolio Selection

Abstract

Price prediction plays a crucial role in consumer management and portfolio selection and significant advancements have been made in this domain by leveraging the capability of deep learning to handle complex relationship information. Nevertheless, existing methods predominantly analyze homogeneous stock relationships, such as industry affiliations, overlooking heterogeneous correlation information in industrial chains’ upstream and downstream segments. To address this gap, we propose a novel approach employing heterogeneous graph attention networks for price prediction and portfolio selection. This method integrates a sequential information process module with a heterogeneous graph attention network that extracts relationships within industrial chains. We employ a comprehensive empirical study of the Chinese stock market, indicating that our model enhances investment and offers novel data-driven business insights for consumers.