Firm-Specific Determinants on Dividend Changes: Insights from Data Mining

This paper aims at investigating the performance of state-of-the-art Data Mining techniques in identifying important firm-specific determinants of dividend changes. Since announcements of dividend changes are said to be informative and likely to affect stock prices, an accurate prediction of dividend changes is of vital interest. Therefore, we compare Data Mining techniques like Classification Trees, Random Forests or Support Vector Machines with classical methods like Multinomial Logit or Linear Discriminant Analysis. This comparison is done on data of the dividend payout of German Prime Standard Issuers during the years 2007–2010, as in this phase of financial turmoil many dividend changes can be observed. To our best knowledge this is the first application of Data Mining techniques in this research field concerning the German Stock Market.