Firm-level variables that predict cross-sectional stock returns, such as price-to-earnings and short interest, are often averaged and used to predict the time series of market returns. We extend this literature and limit the data-snooping bias by using a large population of the literature’s cross-sectional return predictors. We find the literature has ignored several cross-sectional variables–such as asset turnover and Z-score–that contain strong in-sample predictability when examined in isolation. However, after accounting for the number of predictors and their interdependence, we find only weak evidence that cross-sectional predictors make good time-series predictors, especially out-of-sample.