This paper investigates whether firms manage analyst forecasts to generate positive earnings surprises and the consequences of such forecast management. We first document that firms quot;talk downquot; forecasts. Forecasts of quarterly earnings issued later in the forecasting horizon grow increasingly pessimistic on average. More importantly, the exact timing of changes in earnings forecasts turn out to be a key determinant of whether a firm indeed succeeds at generating positive earnings surprises. In particular, (i) changes in consensus early in the forecast horizon have no effect on the probability that earnings will exceed the consensus, (ii) late forecasts that raise the consensus sharply reduce the probability of a positive earnings surprise, and (iii) late forecasts that lower the consensus sharply raise the probability of a positive earnings surprise. These last two findings are the opposite of what would be predicted if deviations of late forecasts from the consensus were due to new information arrival. We then find evidence that investors are systematically quot;misledquot; by late arriving forecasts. In particular, downward revisions in the consensus lead to large positive cumulative abnormal returns following the earnings announcement. Finally, while the finding that investors reward firms that successfully manage forecasts down might seem to provide a rationale for downward forecast management, this is not so. Specifically, controlling for the extant earnings-consensus forecast differential, the negative impact of downward forecast revisions on stock price dominates the stock price appreciation following the earnings announcement. This begs the question: Firms manage analyst forecasts (down), but why?