Enhancing Risk Identification with GNN: Edge Classification in Risk Causality from Securities Reports

•Study aims to extract implicit risk information in financial disclosures.•Emphasizes needs for advanced risk assessment due to evolving corporate policies.•Understanding causalities in risk graphs aids in accurate risk structure assessment.•Proposes an edge classification method for risk causality in graphs.•Enhances risk analysis effectiveness, aiding diverse industries and investors. In the evolving business landscape, the scope of risk factors is extremely wide, making it impossible for all business-related risks to be captured within publicly available financial disclosures. Previous studies have predominantly focused on understanding causal relationships and risk chains based on the risks that are explicitly documented. Thus, risks that are not explicitly listed are often overlooked. The aim of this study was to analyze risk chains and extract implicit information from disclosed documents. We focused on edge classification and suggested suitable labels for the edges of a risk chain graph. Furthermore, we proposed an edge-type classification in heterogeneous graphs using Graph Neural Networks (GNN). This was accomplished by defining six risks and constructing risk-chain graphs. The outcomes demonstrated the edge-type classification proved to be an effective approach compared with existing method. This method holds the potential to aid investors in enhancing their profits and making more informed decisions.