Machine learning and deep learning methodologies are commonly utilized for forecasting stock prices. Huynh et al. (2023) introduced the ESTIMATE model in a recent study, utilizing hypergraph convolutional networks to link stocks through industry and price correlation matrices. The approach led to notable enhancements in the precision of stock price predictions. However, the conventional methods of industry classification and price correlation might not comprehensively depict the intricate relationships among stocks. Furthermore, existing graph convolution networks and hypergraph convolution techniques, which are primarily developed in Euclidean space, may not adequately handle the complexities and hierarchical nature of the stock market. To address this, the current research integrates hyperbolic matrices and contrast learning into hypergraph convolution models. This integration enables the capturing of the inherent hierarchical relationships within the market using hyperbolic matrices and discerning subtle market behaviors under varying conditions through contrast learning. An empirical analysis was conducted on the Standard & Poor’s 500 index of U.S. stocks, obtained from the Yahoo Finance database spanning from January 1, 2016, to May 1, 2022, encompassing 1593 trading days. The outcomes were compared with the ESTIMATE model, demonstrating a considerable enhancement in prediction accuracy across multiple metrics.