From News to Returns: A Granger-Causal Hypergraph Transformer on the Sphere

TL;DR

The paper introduces CSHT, a Causal Sphere Hypergraph Transformer for financial time series that unifies Granger-causal hypergraphs, spherical embeddings, and causally masked attention to model how news and sentiment drive asset returns. By encoding directional multivariate dependencies as hyperedges on a hypersphere and constraining attention with geodesic masks, CSHT achieves robust generalisation across market regimes and provides transparent attribution pathways from macro events to stock-level responses. Empirical results on 2018–2023 S&P 500 data show CSHT outperforms strong baselines in next-day return prediction, regime classification, and top-asset ranking, with higher causal alignment and interpretability demonstrated through case studies of a Fed rate hike and the COVID-19 crash. The approach offers a principled, geometry-aware framework for trustworthy financial forecasting under uncertainty, with potential extensions to interventional reasoning and cross-market applications.

Abstract

We propose the Causal Sphere Hypergraph Transformer (CSHT), a novel architecture for interpretable financial time-series forecasting that unifies \emph{Granger-causal hypergraph structure}, \emph{Riemannian geometry}, and \emph{causally masked Transformer attention}. CSHT models the directional influence of financial news and sentiment on asset returns by extracting multivariate Granger-causal dependencies, which are encoded as directional hyperedges on the surface of a hypersphere. Attention is constrained via angular masks that preserve both temporal directionality and geometric consistency. Evaluated on S\&P 500 data from 2018 to 2023, including the 2020 COVID-19 shock, CSHT consistently outperforms baselines across return prediction, regime classification, and top-asset ranking tasks. By enforcing predictive causal structure and embedding variables in a Riemannian manifold, CSHT delivers both \emph{robust generalisation across market regimes} and \emph{transparent attribution pathways} from macroeconomic events to stock-level responses. These results suggest that CSHT is a principled and practical solution for trustworthy financial forecasting under uncertainty.

Figures (7)

• Figure 1: Overview of the causal forecasting problem setup.
• Figure 2: Causal influence during the COVID-19 regime: News $x_{t-2}^{\text{COVID}}$ and sentiment $s_{t-1}^{\text{fear}}$ drive returns $r_t^{\text{health}}$. CSHT captures such patterns via hyperspherical hyperedges and causal attention masks.
• Figure 3: CSHT architecture: Granger-causal links connect news, sentiment, and returns. Nodes are embedded on a hypersphere and processed via masked geodesic attention for interpretable forecasting.
• Figure 4: Illustrative causal flow: news affects sentiment, which drives index movements (SPY) and individual returns (e.g., JPM, DAL).
• Figure 5: Runtime scaling of CSHT with number of assets. The model maintains near-linear growth, ensuring scalability to large portfolios.