A1040
Title: A study on graph neural network-based stock forecasting methods in stock market
Authors: Mingyu Go - Chungnam National University (Korea, South) [presenting]
Minsu Park - Chungnam National University (Korea, South)
Abstract: Predicting stock prices involves a complex process fraught with various uncertainties, encompassing a comprehensive consideration of diverse economic, corporate, and market factors. Therefore, by considering the intricate relationships among these factors, the accuracy of stock price predictions can be enhanced. A new approach is proposed to future stock price prediction by integrating traditional time series analysis with inter-stock network structures. To calculate inter-stock similarities, multidimensional analysis reflecting the characteristics of each stock was conducted, and inter-stock adjacency was defined using a graph generation algorithm. The constructed network and market indicators were effectively learned through the graph neural network (GNN), enabling the prediction of short-term future stock prices for each stock. As a result, the network-based approach has demonstrated a more precise reflection of market trends and proven superior predictive performance compared to traditional time series analysis. The importance of understanding and leveraging interplay is underscored among stocks in stock market analysis, and methodologies that can contribute to predicting future market values are proposed.
