Empirical Asset Pricing via Machine Learning : The Global Edition

We examine the cross-section of international equity risk premia with machine learning methods. We identify, classify, and calculate 88 market characteristics and use them to forecast country returns with various machine learning techniques. While all algorithms produce substantial economic gains, a two-layer neural network proves particularly effective. The associated long-short portfolio generates 1.69% per month at a Sharpe ratio of 1.57. Most models select a consistent group of leading predictors: long-run reversal, earnings yield, size, market breadth, and momentum. The return predictability is driven by mispricing rather than risk. In consequence, it is boosted by high limits to arbitrage but gradually diminishes over time as global markets mature