Predicting stock price movements in supply chain networks

In today’s interconnected business world, firms are rarely just operating by themselves. Instead, they are connected through different kinds of relationships, and their supply chain systems of buyers and sellers capture one of the most important ones for business operations. A supply chain network approach provides a unique perspective to illustrate the characteristics of a supply chain and make better managerial decisions. Previous research has identified that a firm’s performance is associated with its partners (customers and suppliers) and other supply chain characteristics. However, there is a dearth of understanding of whether and how different types of neighbors, especially those beyond direct partners, can improve the prediction of a firm’s performance.

The main goal of this dissertation is to fill that gap. Using real-world data, this work first builds a supply chain network connecting the firms in the S&P500 and identifies relevant neighbors beyond direct partners. Results show that incorporating relevant neighbors, even though some are not immediate neighbors in the supply chain network, improves the performance of stock movement prediction. Inspired by these findings, the second part of this thesis proposes a deep learning model named HT-GNN to predict a focal firm’s future stock price movement over time. The HT-GNN model combines signals from neighboring firms, historical stock data, and general market trends into the prediction. Overall, this dissertation demonstrates that firms’ performance can be better predicted if their inter-connectedness in a supply chain network is properly incorporated into a machine learning model. The results have implications for investing, risk management as well as our understanding of complex networks among business organizations.