I develop two new types of portfolio efficiency when returns are predictable. The first type maximizes the unconditional Sharpe ratio of excess returns and differs from unconditional efficiency unless the safe asset return is constant over time. The second type maximizes conditional mean-variance preferences and differs from unconditional efficiency unless, additionally, the maximum conditional Sharpe ratio is constant. Using stock data, I quantify and test their performance differences with respect to unconditionally and fixed-weight efficient returns. I also show the relevance of the two new portfolio strategies to test conditional asset pricing models.