We propose that fund performance can be predicted by its R-2, obtained from a regression of its returns on a multifactor benchmark model. Lower R-2 indicates greater selectivity, and it significantly predicts better performance. Stock funds sorted into lowest-quintile lagged R-2 and highest-quintile lagged alpha produce significant annual alpha of 3.8%. Across funds, R-2 is positively associated with fund size and negatively associated with its expenses and manager's tenure.