Zero-Beta (Factor-Neutral) Portfolios
Zero-beta portfolios (factor-neutral portfolios) are constructed to eliminate exposure to systematic risk within a factor model. This paper examines their novel role in empirical asset pricing. I develop a unified framework for testing and comparing factor models based on the maximum Sharpe ratio of zero-investment zero-beta portfolios, which is broadly applicable and robust to practical frictions. Although all models are formally misspecified, machine learning–based approaches dominate conventional ones. I further propose an optimal zero-beta investment strategy that exploits model mispricing, delivering robust out-of-sample performance and outperforming most established strategies after accounting for transaction costs. I demonstrate that, in practice, leveraging modern asset pricing models may be more effective by systematically trading model mispricings rather than harvesting factor risk premia. Finally, I show that using unit-investment zero-beta portfolios to estimate the zero-beta rate introduces an upward bias proportional to the degree of model misspecification.