This paper uses a dynamic partial equilibrium model to explain a puzzle of dividend smoothing. In contrast to the Modigliani-Miller theory, I show that firm value depends on payout policy. The analysis implies that firms with more stable dividend stream are more valuable. This explains why dividends are rigid over time. A volatile component of dividends is introduced to reduce the likelihood of dividend omission in bad times while keeping the same historical average dividends. I show that the empirically observed positive relation between dividends and future firm performance is a statistical artifact driven by dividend smoothing. Thus, the empirical tests of dividend signaling theory might be misspecified.