Hermes: A Multi-Scale Spatial-Temporal Hypergraph Network for Stock Time Series Forecasting

TL;DR

Hermes tackles the challenge of stock time series forecasting by explicitly modeling inter-industry lead-lag dynamics and multi-scale information within a unified spatial-temporal hypergraph framework. It introduces a hyperedge-based moving aggregation to capture lead-lag relationships and a cross-scale, edge-to-edge fusion mechanism that preserves scale-specific information while enabling cross-scale interaction. The approach combines multi-scale feature extraction, adaptive hypergraph construction, and a prediction head, achieving state-of-the-art performance on NASDAQ, NYSE, and S&P 500 datasets with strong ablation support for each component. The resulting framework offers improved forecasting accuracy and practical efficiency, enabling more reliable decision-making for investors and regulators.

Abstract

Time series forecasting occurs in a range of financial applications providing essential decision-making support to investors, regulatory institutions, and analysts. Unlike multivariate time series from other domains, stock time series exhibit industry correlation. Exploiting this kind of correlation can improve forecasting accuracy. However, existing methods based on hypergraphs can only capture industry correlation relatively superficially. These methods face two key limitations: they do not fully consider inter-industry lead-lag interactions, and they do not model multi-scale information within and among industries. This study proposes the Hermes framework for stock time series forecasting that aims to improve the exploitation of industry correlation by addressing these limitations. The framework integrates moving aggregation and multi-scale fusion modules in a hypergraph network. Specifically, to more flexibly capture the lead-lag relationships among industries, Hermes proposes a hyperedge-based moving aggregation module. This module incorporates a sliding window and utilizes dynamic temporal aggregation operations to consider lead-lag dependencies among industries. Additionally, to effectively model multi-scale information, Hermes employs cross-scale, edge-to-edge message passing to integrate information from different scales while maintaining the consistency of each scale. Experimental results on multiple real-world stock datasets show that Hermes outperforms existing state-of-the-art methods.

Figures (5)

• Figure 1: (a): Intra-industry correlation. Relations between stocks within an industry (hyperedge) can be represented by a hypergraph. (b): Lead-lag correlation. (c): Multi-scale correlation.
• Figure 2: The architecture of Hermes.
• Figure 3: Parameter sensitivity studies of main hyper-parameters in Hermes.
• Figure 4: The specific lead-lag interaction process within the window, with a lead-lag step of 3 as an example.
• Figure 5: Three types of message passing: hypernodes for stocks and hyperedges for industries.