This paper delves into the application of machine learning to forecast international stock returns based on firm characteristics. We reveal that market-specific training significantly enhances the performance of neural networks (NNs), as NNs trained on individual markets outperform those using a global model based on U.S. data.

Our analysis shows that NNs excel in predicting stock return rankings and in forming profitable portfolios. In contrast, regression trees demonstrate weaker performance, particularly in markets with limited data availability. This finding suggests that the choice of machine learning model is critical in the context of international finance, where data characteristics can vary widely across different markets.

Additionally, we explore the impact of incorporating U.S. firm characteristics into market-specific models. Our results indicate that these additional variables can enhance the return predictability of NNs, highlighting the interconnectedness of global markets. This research contributes to the growing literature on machine learning in finance and provides valuable insights for investors seeking to leverage advanced modeling techniques for international stock returns.