This study examines whether global or regional factor models are more effective in predicting stock returns, using an innovative machine learning approach. By moving away from traditional methods that heavily rely on past data and linear models, the researchers employ an ex-ante forecasting strategy with a variety of machine learning techniques, both linear and non-linear. The analysis, which utilizes data from 24 developed markets between 1990 and 2021, reveals that while regional models perform slightly better with simple linear methods, they lack statistical significance. In contrast, complex machine learning algorithms trained on global data consistently outperform, offering both economic and statistical advantages. This challenges previous findings and provides a fresh perspective on financial market integration and stock return predictability. The study’s strengths include a comprehensive comparison of predictive performance across various models and a robust empirical design. However, it could improve by addressing model selection, economic relevance, emerging market analysis, interpretability, and performance across different market regimes. Overall, the research advances the field by highlighting the benefits of globally trained models using advanced algorithms, suggesting areas for future exploration to enhance financial forecasting.
Are Global or Regional Models Better for Predicting Stock Returns?
