We propose a theory of self-selection in which fund managers choose two important dimensions of fund activeness: level of activeness and tracking error. To facilitate investor learning, skilled managers choose a level of activeness that maximizes the information ratio relative to that of unskilled managers. The latter, to hinder investor learning, scale up their tracking error (TE) with positive probability. In equilibrium, low-TE funds outperform high-TE funds, exhibit larger cross-sectional dispersion in performance, and higher flow-performance sensitivity. Improvements in fund reputation induce unskilled managers to rely more on investments with high TE. Our motivating evidence is consistent with these findings.