Exploring stock returns in financial markets with interpretable financial variables and graph neural networks

We integrate classic financial variables with financial-network structures in a Graph Convolutional Network (GCN) framework to address transparency in machine-learning applications to finance. By stratifying and grouping data based on known variables, our GCN model improves both the interpretability and accuracy of financial analysis. This approach clarifies the interconnected nature of financial markets and shows the predictive value of combining traditional financial insights with advanced modeling techniques.