Conversation Modeling: Analyzing Earnings Conference Calls for Financial Risk Prediction

Predicting the risk of public firms is a fundamental task in investment and risk management. In financial economics, the risk of a public firm is commonly measured as the stock price volatility during a period of time. Literature has shown that a firm’s risk is predictable using publicly available data such as the firm’s earnings conference call, where managers and analysts discuss key business issues. However, these conference call transcripts are often very long and contain a diverse range of non-risk relevant topics, which poses challenges for the text-based risk forecasting. Therefore, an effective understanding of the complex and latent text semantics within conference call transcripts is the key to risk prediction. In this study, we aim to investigate the structural dependency within a conference call transcript by explicitly modeling the dialogue between managers and analysts. Specifically, we decompose a long conference call transcript into a conversation graph, where nodes are different discussion units and the edge between two nodes encodes their semantic similarity. This novel design improves the transcript representation performance, which in turn reduces the risk forecast error. Computational experiments conducted on a large-scale firm risk prediction dataset show that our approach significantly reduces the risk forecast error compared to several state-of-the-art text-based methods. We also extensively evaluate the effectiveness of our designed conversation graph. We believe that our design can benefit researchers and practitioners and illuminate possible approaches for other financial prediction tasks.

Speaker Bio

Yi Yang is an Associate professor in the Department of Information Systems, Business Statistics and Operations Management at Hong Kong University of Science and Technology. He received his PhD in computer science from Northwestern University. He designs machine learning methods in his research to solve challenging business and Fintech problems. His work has been published in business discipline journals such as Information Systems Research, Management Information Systems Quarterly, Journal of Marketing, Contemporary Accounting Research and INFORMS Journal on Computing. His work has also been published in top-tier machine learning and natural language processing conferences such as Annual Meeting of the Association for Computational Linguistics (ACL), Conference on Empirical Methods in Natural Language Processing (EMNLP) and International Conference on Artificial Intelligence and Statistics (AISTATS).

Please note: this event is for University of Edinburgh staff and students only and is not open to the public. Guests will be required to sign in and show staff ID.