Structure Over Signal: A Globalized Approach to Multi-relational GNNs for Stock Prediction
Amber Li, Aruzhan Abil, Juno Marques Oda
TL;DR
The paper tackles stock return prediction under macroeconomic shocks by building OmniGNN, a multi-relational GNN that combines metapath attention, Transformer-based temporal encoding with ALiBi, and a global industry node to enable fast shock propagation. The approach leverages two metapaths, $\mathcal{SS}$ and $\mathcal{SIS}$, and a temporal sequence model to capture long-range dependencies while mitigating oversmoothing through a star-like topology. Empirical results on 10 IT stocks from 2019–2022 show OmniGNN outperforming GAT and Transformer baselines in IC, IR, CR, and Precision@K, with COVID-19 ablations highlighting the macro-structure’s importance. The method offers a scalable framework for multi-relational financial graphs, robust to market stress, and applicable to larger stock universes.
Abstract
In financial markets, Graph Neural Networks have been successfully applied to modeling relational data, effectively capturing nonlinear inter-stock dependencies. Yet, existing models often fail to efficiently propagate messages during macroeconomic shocks. In this paper, we propose OmniGNN, an attention-based multi-relational dynamic GNN that integrates macroeconomic context via heterogeneous node and edge types for robust message passing. Central to OmniGNN is a sector node acting as a global intermediary, enabling rapid shock propagation across the graph without relying on long-range multi-hop diffusion. The model leverages Graph Attention Networks (GAT) to weigh neighbor contributions and employs Transformers to capture temporal dynamics across multiplex relations. Experiments show that OmniGNN outperforms existing stock prediction models on public datasets, particularly demonstrating strong robustness during the COVID-19 period.