From Regression Models to Machine Learning Methods: A Bibliometric Review on Stock Market Prediction

Abstract

Stock market predictability remains a central question in financial economics, positioned at the intersection of asset pricing theory, econometric forecasting, and artificial intelligence. Using bibliometric methods, this study maps the intellectual history of this field by analyzing 12,495 publications from the Web of Science (WoS) Core Collection between 1994 and 2024. The analysis reveals a methodological shift in the literature: a transition from testing linear hypotheses anchored in classical market efficiency to engineering high‐dimensional representations of market environments using computational methods. Validated through triangulation of document co‐citation networks and keyword burst dynamics, the results trace the field's evolution from foundational stochastic volatility and factor models to contemporary deep learning and natural language applications, a transition accelerated by exogenous macroeconomic shocks. Furthermore, the research frontier is driven by three dimensions: measuring latent state variables via multimodal data, modeling strategic market interactions through agent‐based simulations, and imposing theoretical discipline on complex predictive architectures. This review offers a framework for integrating artificial intelligence with financial economics to develop more adaptive and interpretable pricing models.