Contrary to the standard practice of using past average realized returns when testing asset pricing models, this paper analyzes the factor structure and the cross-sectional variability of expected returns. We show that the first two principal components explain 99.5% of the variability of (lower bound) expected returns. Quality, funding illiquidity, default premium and market-wide variance risk premium explain most of the time-varying behavior of the first principal component. Market-wide illiquidity significantly explains the second principal component. The cross-sectional fit of several asset pricing models using expected returns is consistently better than the one with average realized returns. The factor loadings of the two principal components explain more than 90% of the cross-sectional variability of expected returns. The most successful model is a multi-factor model with the market, and the four aggregate factors that explain the first principal component. This model is able to explain more than 95% of the cross-sectional variability of expected excess returns, with a corrected R-square of 60%, which is (asymptotically) different from zero.