Loading...

Stock Prediction and Portfolio Management in Iran Stock Market using Dynamic Graph Neural Network

Gharabeyk, Shirin | 2024

0 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: English
  3. Document No: 57520 (52)
  4. University: Sharif University of Technology, International Campus, Kish Island
  5. Department: Science and Engineering
  6. Advisor(s): Hemmatyar, Ali Mohammad Afshin
  7. Abstract:
  8. The stock market is vital to the economy as it serves as a key indicator of economic health. AI algorithms can process vast amounts of market data at high speeds, identifying patterns and trends that are often invisible to human analysts. Forecasting future stock trends poses significant challenges due to the complex inter-stock and intra-stock dynamics that influence stock prices. Recently, graph neural networks have shown promising results by modeling multiple stocks as graph-structured data. However, many of these approaches depend on manually defined factors to construct static stock graphs, which fail to capture the rapidly changing interdependencies between stocks. In this work, Enhanced Renyi Transfer Entropy (ERTE) is employed that requires no expert knowledge to overcome these limitations. The method dynamically constructs stocks graph using entropy-driven edge generation, where nodes represent stocks and edges quantify the information transfer between them. ERTE can determine leading and lagging stocks, highlighting those that drive market movements and those that follow. To enhance the model further, task-optimal dependencies are learned between stocks through a graph diffusion process and a decoupled representation learning scheme is used to capture hierarchical intra-stock features. Also, focal loss is employed to improve predictive accuracy. This loss function down-weights well-classified examples and emphasizes misclassified ones, enhancing the model's sensitivity to rare events and subtle patterns. Experimental results show significant improvements over the state-of-the-art baseline on real-world datasets
  9. Keywords:
  10. Stock Market Prediction ; Graph Neural Network ; Trausfer Entropy ; Focal Loss ; Portfolio Management ; Dynamic Neural Network

 Digital Object List

 Bookmark

No TOC