Abstract <p>Quantitative investment increasingly relies on complex models like temporal graph networks for asset return prediction, yet these models often function as “black boxes,” hindering user trust and effective decision-making. Concurrently, the multi-source, dynamic nature of financial data poses significant analytical challenges. To address these issues, we present QuantVisExplorer, a visual analytics system integrating a novel Multi-Relational Temporal Graph Fusion (MRTGF) model. Our core contributions are threefold: (1) a systematic data processing pipeline to construct multi-relational graphs from heterogeneous financial data; (2) the innovative MRTGF model, which improves prediction accuracy by fusing static industry-based structures with dynamic, news-driven event relationships; (3) the QuantVisExplorer system itself. The system is designed around a multi-level analytical workflow (Market, Industry, and Asset) that guides users from high-level overviews to granular details. To foster model trust, it provides deep explainability by visualizing feature contributions for prediction attribution and includes an interactive backtesting module for strategy validation. A comprehensive evaluation—including quantitative model comparisons, ablation studies, two in-depth case studies with domain experts, and a formal user study—demonstrates that QuantVisExplorer significantly enhances insight generation, model understanding, and decision-making confidence in quantitative investment.</p> Graphical abstract <p></p>