From 1988 to 2003, the average change in managerial ownership is significantly negative every year for American firms. We find that managers are more likely to significantly decrease their ownership when their firms are performing well, but not more likely to increase their ownership when their firms have poor performance. Because investors learn about the total change in managerial ownership with a lag, changes in Tobin's q in a period can be affected by changes in managerial ownership in the previous period. In an efficient market, it is unlikely that changes in managerial ownership in one period are caused by future changes in q. When controlling for past stock returns, we find that large increases in managerial ownership increase q. This result is driven by increases in shares held by officers, while increases in shares held by directors appear unrelated to changes in firm value. There is no evidence that large decreases in ownership have an adverse impact on firm value. We argue that our evidence cannot be wholly explained by existing theories and propose a managerial discretion theory of ownership consistent with our evidence.