STGNN for share trend line prediction and change point detection on JSE.

This research explores the application of advanced Spatial-Temporal Graph Neural Networks (STGNNs) for predicting events of interest in the Johannesburg Stock Exchange Market (JSE), specifically future share trend lines; how they can be used to detect events of interest such as trend change points on JSE. Two models are investigated in depth: Attention-based Spatial Temporal Graph Neural Network (ASTGCN), which applies dual attention mechanisms to enhance temporal and spatial feature learning; and Temporal-Correlation Graph Pre-Trained Network (TCGPN), a recently proposed architecture that dynamically constructs inter-stock graphs based on temporal correlations. By evaluating the capabilities of ASTGCN and TCGPN, this study aims to improve both predictive accuracy and the visual explainability of market dynamics, evaluating trend line prediction for both long and short term share trading - contributing a novel application of STGNNs to trend line forecasting and further evaluation of STGNN approaches to financial time-series forecasting.