Contrary to the central prediction of signaling models, changes in profits do not empirically follow changes in dividends. We show both theoretically and empirically that dividends signal safer, rather than higher, future profits. Using the Campbell (1991) decomposition, we are able to estimate expected cash flows from data on stock returns. Consistent with our model's predictions, cash-flow volatility changes in the opposite direction from that of dividend changes and larger changes in volatility come with larger announcement returns. We found similar results for share repurchases. Crucially, the data supports the prediction unique to our model that the cost of the signal is foregone investment opportunities. We conclude that payout policy conveys information about future cash flow volatility. Our methodology can be applied more generally to overcoming empirical problems in testing theories of corporate financing.