Do dividend changes signal future earnings?

Signaling models contributed to the corporate finance literature by formalizing "the informational content of dividends" hypothesis. However, these models are under criticism as the empirical literature found weak evidences supporting a central prediction: the positive relationship between changes in dividends and changes in earnings. We claim that the failure to verify this prediction does not invalidate the signaling approach. The models developed up to now assume or derive utility functions with the single-crossing property. We show that, in the absence of this property, signaling is possible, and changes in dividends and changes in earnings can be positively or negatively related.