Abstract:Stock trend classification remains a fundamental yet challenging task, owing to the intricate time-evolving dynamics between and within stocks. To tackle these two challenges, we propose a graph-based representation learning approach aimed at predicting the future movements of multiple stocks. Initially, we model the complex time-varying relationships between stocks by generating dynamic multi-relational stock graphs. This is achieved through a novel edge generation algorithm that leverages information entropy and signal energy to quantify the intensity and directionality of inter-stock relations on each trading day. Then, we further refine these initial graphs through a stochastic multi-relational diffusion process, adaptively learning task-optimal edges. Subsequently, we implement a decoupled representation learning scheme with parallel retention to obtain the final graph representation. This strategy better captures the unique temporal features within individual stocks while also capturing the overall structure of the stock graph. Comprehensive experiments conducted on real-world datasets from two US markets (NASDAQ and NYSE) and one Chinese market (Shanghai Stock Exchange: SSE) validate the effectiveness of our method. Our approach consistently outperforms state-of-the-art baselines in forecasting next trading day stock trends across three test periods spanning seven years. Datasets and code have been released (

What problem does this paper attempt to address?

This paper attempts to address two major challenges in stock market prediction: 1) the dynamic and asymmetric relationships among stocks, and 2) the hierarchical structure of internal characteristics within stocks. Specifically: 1. **Dynamic and Asymmetric Relationships**: Traditional graph neural network methods usually use static graphs to represent the relationships among stocks, which are constructed based on industry - or company - specific relationships. However, this method ignores the dynamic changing characteristics of relationships in the stock market, that is, the relationships among stocks change rapidly over time. Such static graphs cannot capture the complex dynamics of the stock market. 2. **Hierarchical Structure of Internal Characteristics within Stocks**: Existing graph neural network methods often closely combine information propagation and graph representation learning in the process, resulting in the loss of hierarchical structure characteristics. These characteristics are crucial for learning the patterns of individual stocks and the overall market. To meet these challenges, the paper proposes a new graph learning framework - Multi - relational Graph Diffusion Neural Network with Parallel Retention (MGDPR). The main contributions of this framework are as follows: 1. **Dynamic Multi - relational Graph Generation**: Quantify the connectivity and strength among stocks on each trading day through information entropy and signal energy to generate a dynamic multi - relational stock graph. This provides a more fine - grained description of the complex dynamic relationships among stocks. 2. **Multi - relational Graph Diffusion**: Adaptively learn the task - optimal edges through a multi - relational diffusion process, further filter out noisy edges and restore task - relevant edges. This helps to alleviate the inconsistency between the task objective and the input graph. 3. **Graph Representation Learning and Parallel Retention**: Adopt a decoupled graph representation learning scheme and introduce a parallel retention mechanism to preserve the hierarchical characteristics within stocks and the long - term dependencies of time series. Through experimental verification on three real - world data sets, MGDPR performs excellently in predicting future stock trends and outperforms the existing state - of - the - art baseline methods. These results indicate that MGDPR can more effectively capture the complex dynamic relationships and internal characteristics of the stock market.

Multi-relational Graph Diffusion Neural Network with Parallel Retention for Stock Trends Classification

DGDNN: Decoupled Graph Diffusion Neural Network for Stock Movement Prediction

MDGNN: Multi-Relational Dynamic Graph Neural Network for Comprehensive and Dynamic Stock Investment Prediction

Dynamic Graph Representation with Contrastive Learning for Financial Market Prediction: Integrating Temporal Evolution and Static Relations

Stock Trend Prediction with Multi-Granularity Data: A Contrastive Learning Approach with Adaptive Fusion

Multi-Graph Convolutional Network for Relationship-Driven Stock Movement Prediction

Hierarchical Adaptive Temporal-Relational Modeling for Stock Trend Prediction

A Dynamic Attributes-driven Graph Attention Network Modeling on Behavioral Finance for Stock Prediction

A novel graph convolutional feature based convolutional neural network for stock trend prediction

Temporal-Relational hypergraph tri-Attention networks for stock trend prediction

Temporal and Heterogeneous Graph Neural Network for Financial Time Series Prediction

Multi-Channel Temporal Graph Convolutional Network for Stock Return Prediction

Temporal Relational Ranking for Stock Prediction

A graph neural network-based stock forecasting method utilizing multi-source heterogeneous data fusion

HATR-I: Hierarchical Adaptive Temporal Relational Interaction for Stock Trend Prediction

Graph-based stock prediction with multisource information and relational data fusion

GraphCNNpred: A stock market indices prediction using a Graph based deep learning system

Graph-Based Learning for Stock Movement Prediction with Textual and Relational Data

Stock Index Prediction Based on Multi-Time Scale Learning with Multi-Graph Attention Networks

A model based LSTM and graph convolutional network for stock trend prediction

ML-GAT:A Multilevel Graph Attention Model for Stock Prediction