TD-HCN: A trend-driven hypergraph convolutional network for stock return prediction.

Area of Science:

• Quantitative Finance
• Machine Learning
• Time Series Analysis

Background:

• Stock data analysis is challenging due to its dynamic, complex, and nonlinear nature.
• Existing graph-based methods struggle to capture higher-order and dynamic stock relationships.
• This limitation hinders the performance of stock return prediction models.

Purpose of the Study:

• To propose a novel Trend-Driven Hypergraph Convolutional Network (TD-HCN) for stock return prediction.
• To integrate diverse stock data types (prices, industry, wiki relationships) for improved analysis.
• To enhance the identification and utilization of both local dynamic and global static relationships.

Main Methods:

• Developed a Trend-Driven Hypergraph Convolutional Network (TD-HCN).
• Employed a Prior-constrained Relational Learning (PCRL) model to discover latent high-order relationships.
• Utilized a Disentanglement Representation Learning (DRL) mechanism with a dual attention module to capture dynamic trends.

Main Results:

• TD-HCN consistently outperformed state-of-the-art methods on NASDAQ and NYSE datasets.
• Achieved significant improvements in stock return prediction.
• Demonstrated effectiveness in learning dynamic stock relationships and capturing trend changes.

Conclusions:

• The proposed TD-HCN model offers a robust and effective approach for stock return prediction.
• The integration of diverse data and advanced deep learning techniques enhances the capture of complex stock market dynamics.
• TD-HCN provides a significant advancement in analyzing and predicting stock market trends.