Principal Portfolios

ABSTRACT We propose a new asset pricing framework in which all securities' signals predict each individual return. While the literature focuses on own‐signal predictability, assuming equal strength across securities, our includes cross‐predictability—leading to three main results. First, we derive optimal strategy closed form. It consists of eigenvectors “prediction matrix,” call “principal portfolios.” Second, decompose problem into alpha and beta, yielding strategies with, respectively, zero positive factor exposure. Third, provide test models. Empirically, principal portfolios deliver significant out‐of‐sample alphas standard factors several data sets.