CORPORATE DIVIDEND POLICY DETERMINANTS: INTELLIGENT VERSUS A TRADITIONAL APPROACH
Date | 01 April 2013 |
Author | Pantelis Longinidis,Panagiotis Symeonidis |
DOI | http://doi.org/10.1002/isaf.1338 |
Published date | 01 April 2013 |
CORPORATE DIVIDEND POLICY DETERMINANTS:
INTELLIGENT VERSUS A TRADITIONAL APPROACH
PANTELIS LONGINIDIS
a
*AND PANAGIOTIS SYMEONIDIS
b
a
Department of Engineering Informatics & Telecommunications,University of Western Macedonia, Karamanli& Lygeris Street,
50100 Kozani, Greece
b
Department of Informatics, Aristotle University of Thessaloniki, 54124 Thessaloniki, Greece
SUMMARY
Dividend is the return that an investor receives when purchasing a company’s shares. The decision to pay these
dividends to shareholders concerns several other groups of people, such as financial managers, consulting firms,
individual and institutional investors, government and monitoring authorities, and creditors, just to name a few.
The prediction and modelling of this decision has received a significant amount of attention in the corporate finance
literature. However, the methods used to study the aforementioned question are limited to the logistic regression
method without any implementation of the advanced and expert methods of data mining. These methods have
proven their superiority in other business-related fields, such as marketing, production, accounting and auditing.
In finance, bankruptcy prediction has the vast majority among data-mining implementations, but to the best of
the authors’knowledge such an implementation does not exist in dividend payment prediction. This paper satisfies
this gap in the literature and provides answers that help to understand the so-called ‘dividend puzzle’. Specifically,
this paper provides evidence supporting the hypothesis that data-mining methods perform better in accuracy
measures against the traditional methods used. The prediction of dividend policy determinants provides valuable
benefits to all related parties, as they can manage, invest, consult and monitor the dividend policy in a more
effective way. Copyright © 2013 John Wiley & Sons, Ltd.
Keywords: dividendpolicy; data mining; decisiontree; neural network;logistic regression; Athens stock exchange
1. INTRODUCTION
The primary goal of financial management (FM) is to maximize the current value per share of the
existing stock. One substantial financial decision affecting this value maximization goal is the dividend
policy (DP). According to Baker (2009), dividend decisions, as determined by a firm’s DP, affect the
amount of earnings that a firm distributes to shareholders versus the amount it retains and reinvests.
DP refers to the payout policy that a firm follows in determining the size and pattern of cash distributions
to shareholders over time.
Corporate DP has captured the interest of economists since the middle of the twentieth century
and over the last six decades has been the subject of intensive theoretical modelling and empirical
examination. A number of conflicting theoretical models, which are lacking in strong empirical
support, define current attempts to explain the corporate dividend behaviour (Frankfurter and
Wood, 2002). Brealey et al. (2004) described el oquently the reason fo r this conflict in the DP
modelling landscape. The authors stated that the endearing feature of economics, where it can
* Correspondence to: Pantelis Longinidis, Department of Engineering Informatics & Telecommunications,University of Western
Macedonia, Karamanli & Lygeris Street, 50100 Kozani, Greece. E-mail: logggas@gmail.com
Copyright © 2013 John Wiley & Sons, Ltd.
INTELLIGENT SYSTEMS IN ACCOUNTING, FINANCE AND MANAGEMENT
Intell. Sys. Acc. Fin. Mgmt. 20, 111–139 (2013)
Published online 17 April 2013 in Wiley Online Library (wileyonlinelibrary.com)DOI: 10.1002/isaf.1338
always accommodate not just two but three opposing points of view, is applicable and induces the
controversy about DP. On the one side there is a group whi ch believes that an incre ase in dividend
payout increases firm value. On the opposite side there is a group which believes that an increase
in payout reduces value. And in the centre there is a party which claims that DP makes no
difference. Black ( 1976) characterized this controversy as a puzzle by arguing that the
harder we look at the dividen d picture, the more i t seems like a puzzle with pieces that just do
not fittogether.
Despite exhaustive t heoretical and em pirical analysis to ex plain their pe rvasive presence,
dividends remain one o f the thorniest puz zles in corporat e finance (Baker et al., 2002). The
inability to resolve the dividend puzzle is mainly due to financial economists’efforts to develop
universal models, al though it is proven that DP is s ensitive to factors suc h as market frictio ns,
firm characteristics, corporate governance and legal environments (Frankfurter and Wood, 1997;
Baker et al., 2008).
Frankfurter and Wood (2002) conducted an extensive literature review in order to explore whether
the puzzling reality of corporate dividend behaviour is caused by three factors; namely: (1) method
of analysis employed; (2) sample period; and (3) data frequency. The authors analysed 150 empirical
studies of corporate DP and came to the conclusion that no dividend model, either separately or jointly
with other models, is supported invariably. However, a semantic part of Frankfurter and Wood’s (2002)
analysis is the presentation of the methods utilized for each model. The vast majority of these models
utilized regression and event studies methods, and the synchronous methods of data mining (DM) were
implemented by none of these models.
The DP decision, and more specifically the decision to pay or not dividends, can be regarded as a
typical binary classification problem of assigning new observations to two predefined decisions as
classes (e.g. ‘yes’and ‘no’dividend payment classes). Despite the fact that many DP methods have
been applied in the financial area, an analogous model in the DP literature does not exist to the best
of our knowledge.
This gap in the existing DP literature has stimulated the research interest of this work as it
desires to fulfil the need to employ DM methods in order to model the decision to pay or not
dividends and to explore whe ther these techni ques are capable of pred icting the dividend paymen t
decision better and more precisely than the traditional regression approaches found scattered in
the DP literature. However, by modelling the DP decision, this study aims further to provide a
convenient and effective d ecision-support tool to investors. Investors will understand the financial
and nonfinancial features that paying and nonpaying companies have and will take them into
account when constructing and managing their investment portfolios. Our research effort is
summarized in the following research questions:
RQ1. Are DM methods more accurate than the logistic regression in predicting the dividend payment
decision of corporations?
RQ2. Which are the financial, managerial, and cor porate governance features of corporations paying
and not paying dividends to shareholders?
The rest of the paper is structured as follows. Section 2 reviews the previous and current body of
literature for both DP and DM in the FM area. Section 3 provides insights into the research methodol-
ogy employed, followed by the dataset generation process in Section 4. The available DM techniques
are applied using this dataset and the results reported and commented on in Section 5. The paper ends
with concluding remarks, managerial implications and further research directions.
112 P.LONGINIDIS AND P. SYMEONIDIS
Copyright © 2013 John Wiley & Sons, Ltd. Intell. Sys. Acc. Fin. Mgmt. 20, 111–139 (2013)
DOI: 10.1002/isaf
To continue reading
Request your trial