What drives stock returns across countries? Insights from machine learning models

We employ machine learning techniques to examine cross-sectional variation in country equity returns by aggregating information across multiple market characteristics. Our models reveal significant return predictability, which translates into discernible patterns in portfolio performance. In addition, variable importance analysis uncovers a sparse factor structure that varies across forecast horizons. A handful of critical predictors—such as long-term reversal, momentum, earnings yield, and market size—capture most of the return differences, while country risk measures play a minor role. Consistent with the partial segmentation perspective, return predictability persists in small, illiquid, and unintegrated markets and weakens over time as the constraints on capital mobility diminish. As a result, attempts to forge them into profitable strategies can be challenging at best. • Machine learning uncovers significant predictability in country equity returns • Critical predictors include long-term reversal, momentum, earnings yield, and market size. • Country risk measures have a much smaller impact on return predictability. • Predictability persists in small, illiquid, and unintegrated markets. • Challenges exist in translating predictability into profitable strategies.