We show that portfolio sorts, as widely used in empirical asset pricing, tend to misattribute cross-sectional return predictability to the firm characteristic underlying the sort. Such misattribution arises if the sorting variable correlates with a firm-specific effect capturing unobservable...