International Journal of Scientific Research and Engineering Development

The stock market has captivated academics for years because of its dynamic, nonlinear, and seemingly random behavior. Earlier research using statistical and econometric models has yielded to machine learning (ML), deep learning (DL), graph neural networks (GNNs), and large language models (LLMs) in the recent literature. In this review, I provide a reflection on more than fifteen of the last several years' papers from 2019-2025 that consider the emerging empirical approaches and the increasing inclusion of alternative data sources, such as sentiment data, in their studies, alongside addressing the underlying issues of reproducibility and explainability of how or why models produce their respective predictions. I observed that while the novel approaches generally report improved predictive performance over baseline models, these results are often based on unrealistic evaluation conditions. The primary takeaway from my reflection is that hybridized profiling approaches (econometric energy, ML exploratory flexibility, and alternative context data) seem to be the best suited for future research on stock market predictions, while the practical deployment of hybrid systems remains limited due to nonstationarity (union and/or intersection of spatial and temporal data/event characteristics) and a lack of transparency in the evaluations of the models.