How Important Are Dividend Signals in Assessing Earnings Persistence

We build and test a Bayesian model that shows how investors revise their earnings persistence expectations after dividend announcements. When dividend changes confirm preceding earnings changes, our model predicts inverse u-shaped investor revisions for noisy dividend signals. As the dividend signal becomes more informative our model predicts that investor revisions will become more skewed converging to a monotonically decreasing relation for perfect dividend signals. When dividend changes contradict preceding earnings changes, our model predicts u-shaped investor revisions. In empirical tests, we find results generally consistent with our model predictions. To validate our proxy for the informativeness of confirming dividend changes (the ratio between dividend changes and earnings changes), we first demonstrate that it can be derived from a model suggested by prior literature and second we show that compared to several alternative proxies together with one other proxy it is most consistent with investor reactions and best predicts future earnings persistence. Our findings are important because they explain conflicting results in the literature and can support hypotheses development and model specification in empirical research.