Efficient Country Selection in G10 Equities: A Machine Learning Approach

National economies are deeply interconnected, to the point that optimizing equity exposure across countries can be difficult. In our latest research, we explored how machine learning could improve the allocation of equity risk premia across G10 countries.

We trained a model on RavenPack News Analytics, which analyzes sentiment in macroeconomic news, to optimize country selection for G10 equity markets.

The model, designed for long-short country rotations, was tested for robustness across holding periods and portfolio sizes. Backtesting confirmed the effectiveness of this approach:

• Higher risk-adjusted returns for mid-frequency trading: rotating between G10 stocks delivers an Information Ratio from 0.84 to 0.61, when holding for 2 days to a week.

• Improved model accuracy to predict next-day equity returns: our ML model outperforms basic sentiment analysis. It increases next-day equity returns by 324 basis points (bps) for a 2-day holding period, 226 bps for a one-week holding period, and 104 bps for a three-week holding period

• Country-specific market movers: the model identifies key news that most impact each market. While news about domestic product is a key factor for all G10 economies, China's market reacts most to balance of payments news, and Italy's to credit conditions, for example.