Cumulative abnormal returns (CARs) computed around acquisition announcements are widely considered to be market‐based assessments of expected value creation. We show, however, that announcement returns do not correlate with commonly used and new measures of ex post outcomes. A simple characteristics‐based model using standard information known at the announcement date can predict these outcomes reasonably well, yet CAR even fails to capture the predictions from this model. Evidence suggests that information about the stand‐alone acquirer dominates CAR, making it virtually impossible to extract deal‐related information. We conclude that CAR is an unreliable measure of expected value creation.