Empirical Dynamic Stochastic Models for Firm Dividends

This paper develops a framework for the empirical analysis and quantitative modeling of firm dividends viewed as discrete stochastic processes. The framework, built around the concept of sojourn intervals - the length of time that dividends remain unchanged - is first used to identify a number of striking dynamic patterns in the dividend payouts of US corporations, some of which are linked to the length of time that the firm has been paying regular dividends. For example, while it is well known that stock splits tend to be followed by dividend increases, we find a similar tendency following dividend cuts. Furthermore, changes in dividend payouts are very poor predictors of such cuts, and their frequency is inversely related to the length of time the firm has been paying dividends. The paper then provides a class of empirical models of firm dividends, based on absorbing Markov chains with phase-type structure, that are consistent with dividend discreteness, and are also sufficiently flexible to address the dynamic patterns observed in the data. An appropriately calibrated empirical model is shown to perform quite well in matching sample moments, and in simulating actual sample paths. The calibration also suggests that the empirical transition matrix of firm dividends has properties that may be generally useful in developing accurate but tractable dynamic specifications for them