Tobias Cramer

This study examines the predictive performance of multi-output machine learning models in estimating industry betas. Multi-output predictions improve forecast accuracy by identifying cross-sectional interdependencies between industries that single-output approaches systematically overlook. Two portfolio applications demonstrate the economic value of these improvements: constructing market-neutral anomaly strategies and optimizing minimum variance portfolios. Our results show that multi-output estimates enable more detailed modelling of systematic risk, leading to more effective hedging strategies, better risk management and greater alignment with investor preferences.