2025年第39期(总第1080期)

演讲主题：Predicting Future Earnings Changes Through Supply Chains: A Machine Learning Approach

主讲人：程茁 香港理工大学副教授

主持人：鲍玉昆 信息管理与数据科学系教授

活动时间：2025年6月12日（周四）10：00-12：00

活动地点：管院大楼119室

主讲人简介：

程茁，香港理工大学商学院副教授，于俄亥俄州立大学Fisher商学院获得博士学位。学术研究主要集中在信息技术的商业价值，社交网络，信息技术与会计金融领域的交叉课题，研究成果已发表在The Accounting Review, Management Science, Information Systems Research, Decision Support Systems, and Information Technology and Management, Internet Research, China Accounting and Finance Review等学术期刊。

活动简介：

Accurate forecasting of company earnings is critical for financial decision-making, yet many traditional methods overlook the interconnectedness of businesses within supply chains. This study proposes a machine learning approach that leverages supply chain network information to improve the prediction of the direction of one-year-ahead earnings changes. We develop a hybrid model—boosting graph neural networks (BGNN)—by integrating gradient boosting decision trees (GBDT) and graph neural networks (GNN). This integrated framework captures both firm-level financial information and the structural relationships among firms within supply chains. Our results indicate that incorporating supply chain information enables the model to achieve an F1 score of 67.13% in predicting the direction of one-year-ahead earnings changes. This performance outperforms two baseline methods: a random forest model with an F1 score of 54.09% and a stochastic gradient boosting model with an F1 score of 56.19%. We also observe significant gains in AUC, recall, precision, and accuracy, underscoring the value of incorporating supply chain network information in earnings forecasting. Further, an examination of predictor importance shed light on the model’s inner workings. Overall, this study advances the literature by incorporating supply chain information into earnings predictions, offering practical insights for investors and analysts.