Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GRU-PFG: Extract Inter-Stock Correlation from Stock Factors with Graph Neural Network

Yonggai Zhuang, Haoran Chen, Kequan Wang, Teng Fei

TL;DR

The paper proposes GRU-PFG, a stock-prediction method that learns inter-stock correlations solely from Alpha360 factors by projecting GRU-derived representations onto a graph and modeling both primary and latent relationships. It introduces a three-stage scheme (Preliminary Information Extraction, Primary Relationship Extraction, and Secondary Relationship Extraction) and computes a final stock feature $F_{last}$ which drives the daily return prediction via $p^t = W_l F_{last} + b_l$, with a loss $L$ based on mean squared error. On CSI300, GRU-PFG achieves $IC = 0.134$, outperforming multi-source models like HIST ($IC = 0.131$) while using only stock-factor inputs, indicating strong inter-stock relational learning and potential for better generalization. Overall, the work demonstrates that graph-based inter-stock correlation learning from standardized factors can rival more data-intensive approaches and offers a pathway to more robust, scalable stock prediction.

Abstract

The complexity of stocks and industries presents challenges for stock prediction. Currently, stock prediction models can be divided into two categories. One category, represented by GRU and ALSTM, relies solely on stock factors for prediction, with limited effectiveness. The other category, represented by HIST and TRA, incorporates not only stock factors but also industry information, industry financial reports, public sentiment, and other inputs for prediction. The second category of models can capture correlations between stocks by introducing additional information, but the extra data is difficult to standardize and generalize. Considering the current state and limitations of these two types of models, this paper proposes the GRU-PFG (Project Factors into Graph) model. This model only takes stock factors as input and extracts inter-stock correlations using graph neural networks. It achieves prediction results that not only outperform the others models relies solely on stock factors, but also achieve comparable performance to the second category models. The experimental results show that on the CSI300 dataset, the IC of GRU-PFG is 0.134, outperforming HIST's 0.131 and significantly surpassing GRU and Transformer, achieving results better than the second category models. Moreover as a model that relies solely on stock factors, it has greater potential for generalization.

GRU-PFG: Extract Inter-Stock Correlation from Stock Factors with Graph Neural Network

TL;DR

The paper proposes GRU-PFG, a stock-prediction method that learns inter-stock correlations solely from Alpha360 factors by projecting GRU-derived representations onto a graph and modeling both primary and latent relationships. It introduces a three-stage scheme (Preliminary Information Extraction, Primary Relationship Extraction, and Secondary Relationship Extraction) and computes a final stock feature which drives the daily return prediction via , with a loss based on mean squared error. On CSI300, GRU-PFG achieves , outperforming multi-source models like HIST () while using only stock-factor inputs, indicating strong inter-stock relational learning and potential for better generalization. Overall, the work demonstrates that graph-based inter-stock correlation learning from standardized factors can rival more data-intensive approaches and offers a pathway to more robust, scalable stock prediction.

Abstract

The complexity of stocks and industries presents challenges for stock prediction. Currently, stock prediction models can be divided into two categories. One category, represented by GRU and ALSTM, relies solely on stock factors for prediction, with limited effectiveness. The other category, represented by HIST and TRA, incorporates not only stock factors but also industry information, industry financial reports, public sentiment, and other inputs for prediction. The second category of models can capture correlations between stocks by introducing additional information, but the extra data is difficult to standardize and generalize. Considering the current state and limitations of these two types of models, this paper proposes the GRU-PFG (Project Factors into Graph) model. This model only takes stock factors as input and extracts inter-stock correlations using graph neural networks. It achieves prediction results that not only outperform the others models relies solely on stock factors, but also achieve comparable performance to the second category models. The experimental results show that on the CSI300 dataset, the IC of GRU-PFG is 0.134, outperforming HIST's 0.131 and significantly surpassing GRU and Transformer, achieving results better than the second category models. Moreover as a model that relies solely on stock factors, it has greater potential for generalization.

Paper Structure

This paper contains 14 sections, 8 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Stock Factors
  4. Prediction Models Purely Based on Alpha360
  5. Prediction Models Based on Multiple Sources of Information
  6. Fundmentals
  7. Stock trend-related concepts
  8. Limitations of GRU in Stock Prediction
  9. Methodology
  10. Experiments
  11. Experiment Setup
  12. Evaluation Results
  13. Impact of Model Parameters and Structure
  14. Conclusion

Figures (4)

  • Figure 1: GRU processing workflow on CSI300 data: Input 360 factors, output one-dimensional daily return
  • Figure 2: Prediction Workflow of GRU-PFG model
  • Figure 3: Comparison of GRU-PFG with other models (just based on Alpha360) on monthly average IC and monthly average Precision@N at different time points, with metrics ordered from top left to bottom right as IC, Precision@5, Precision@10, and Precision@30
  • Figure 4: Comparison of monthly average IC and monthly average Precision@N with different methods for measuring stock similarity