Can machine learning (ML) enhance portfolio optimization by improving return forecasts? We tackle this question using elastic net models trained on 147 firm-level features across 37 countries, feeding their predictions into a Markowitz framework. The results are clear: ML-guided portfolios consistently outperform benchmarks, delivering higher returns, lower volatility, and significantly better Sharpe ratios. This outperformance holds across firm sizes, markets, and implementation choices. Globally, ML portfolios achieve monthly returns above 2% with Sharpe ratios up to 1.48—more than triple those of passive strategies. In short, ML forecasts breathe new life into traditional portfolio theory.
Cakici et al. (Thu,) studied this question.