Research note: neural network analysis of dividend policy

Summarizes the main hypotheses used in previous research on dividend policy and reports a study of patterns in dividend payouts/growth using neural networks as a data mining technique. Discusses the properties of neural networks, recognizes that they are unsuitable for hypothesis testing and uses sensitivity analysis on 1992‐1997 data from 201 US firms. Presents the results, which do not outperform a previous model based on factor analysis, finds no significant nonlinear relationship in the data; but shows that dividend variability is sensitive to input variables, especially dividend growth.