From Earnings Dialogue to Market Dynamics: A Machine Learning and NLP Approach to Abnormal Return Prediction

This dissertation presents a comprehensive framework that integrates textual analysis of earnings call transcripts with traditional financial metrics to improve forecasting accuracy in complex, data-rich financial environments. The research specifically addresses the combination of structured financial data with unstructured textual data from corporate earnings calls, expanding beyond conventional financial forecasting models, which typically focus on quantitative measures such as stock prices and financial ratios. Traditional methods often underutilize qualitative insights obtained from managerial discussions and related textual sources. Leveraging recent advancements in artificial intelligence and natural language processing, this study develops a unified framework incorporating sentiment analysis and topic modeling into conventional financial indicators. The dissertation consists of four studies. Study 1 includes a systematic literature review, identifying research gaps concerning the integration of unstructured textual data in financial forecasting models. Study 2 evaluates a hybrid forecasting model that combines traditional financial indicators with qualitative features extracted from over 150,000 earnings call transcripts, showing significant improvements in predicting abnormal stock returns across multiple indices and forecasting horizons. Study 3 conducts an in-depth feature importance analysis, highlighting that thematic clusters related to financial performance, strategic management, and communication practices consistently enhance predictive performance. Study 4 updates existing literature regarding recent developments in artificial intelligence in financial applications, emphasizing advancements in transformer-based architectures and the integration of alternative datasets. Overall, the findings indicate that combining structured and unstructured datasets significantly improves forecasting precision, providing a robust foundation for enhanced risk management and investment decision-making.