Testing dividend signaling models

This paper exploits a key monotonicity property common to dividend signaling models-the greater the rate that dividend income is taxed relative to capital gains income, the greater the value of information revealed by a particular dividend yield-to distinguish the hypothesis that dividends are used as a signaling device from the hypothesis that dividends contain information but are not used as Spencian signals. The monotonicity conditions are tested with robust nonparametric techniques. While the monotonic relationship predicted by signaling theory can be found, a more careful inspection reveals that it does not hold for different levels of the dividend signal, as required by signaling theory. This strongly suggests that existing signaling models cannot explain the dividend policy choices of firms.