AD ALTA
JOURNAL OF INTERDISCIPLINARY RESEARCH
FORMATION OF COMPLEX COMPANY EVALUATION METHOD THROUGH NEURAL
NETWORKS BASED ON THE EXAMPLE OF CONSTRUCTION COMPANIES´ COLLECTION
a
MAREK VOCHOZKA
The Institute of Technology and Business in České Budějovice,
Okružní 517/10, 370 01
České Budějovice, Czech Republic
email:
a
vochozka@mail.vstecb.cz
Abstract: The main goal of this contribution is to create a model through the use of
neural networks, which will be able to predict the company´s ability to survive a
prospective financial crisis. Artificial neural networks are able to conduct non-linear
statistical modelling and offer a completely suitable alternative to individual financial
indicators within complex methods of evaluation. The contribution examines the basic
data on companies coming from the Albertina database. The collection includes both
financial and non-financial indicators of all construction companies in the Czech
Republic within the period of 2008 to 2014. The object is to find an artificial neural
network, which can classify each company based on the input data. Three neural
networks are given and described, proving positive results. The best results are
achieved by MLP 15:15-54-66-4:1. Through this network the Czech construction
companies´ ability to survive a possible distress is consequently evaluated.
Keywords: complex company evaluation, artificial neural networks, construction
company, company bankruptcy, financial and non-financial indicators, predictive
model
1 Introduction
In all companies, the meaning of enterprise evaluation keeps
growing within today´s constantly changing economic
environment (Fotr and Kislingerova, 2009). Enterprise evaluatio
n is the basic element for understanding the sources of company
competition and at the same time, it is a source for company´s
strategy implementation support. It is obvious that the
knowledge of a company´s financial position is necessary.
Reverse information is able to discover areas in which the
enterprise was successful and how or where it has fulfilled the
expectations and its aims. They may also point to situations not
expected or managed by the enterprise and to situations, which
may occur in the closest future (Vochozka et al., 2017).
According to Wang, Stockton and Baguley (2010), success of
the enterprise is even directly dependent on an exact prediction
of future development.
T
he process of a complex enterprise evaluation represents an
objective, just and exact evaluation of enterprise function using
mathematical statistics and operative research principles (Zhang
and Zhong, 2015, p. 178). A correct enterprise evaluation may
be ensured only by relevant methods. The last fifty years have
brought a varied consideration range of approaches, methods,
and tools of its measurements (Wagner, 2011, p. 776).
According to Vlachy (2009, p. 147) traditional methods of
financial analysis are insufficient. For instance, ratio analysis,
using balance sheet and profit and loss statement data is still a
widely used method, which may thus easily interpret the
enterprise´s financial situation (Savvidis and Ginoglou, 2013).
But not even this enterprise evaluation based on the analysis of
financial data is sufficient (Smeureanu et al., 2011). Modern
enterprises produce huge amounts of data, and traditional
analytical tools and methods are no longer able to process such
amounts of information collectively (Yan, Wang and Liu, 2012,
p. 275). Enterprise evaluation should use both financial and non-
financial indicators. (Hsiang et al. 2013). The truth is that
information nowadays may represent relatively precious
company wealth. A huge amount of data may also fundamentally
influence complex enterprise evaluation (Machek and Hnilica,
2012). The ability to analyse and use massive amounts of
information still keeps lagging behind the ability to collect and
keep them (Wang, Rees and Liao, 2002).
Complex enterprise evaluation methods are a specific group of
tools used for suitable enterprise evaluation – mainly
multidimensional models working with several criteria assigned
specific weight (importance). The enterprise´s situation is then
collectively expressed by one number, which evaluates the level
of the enterprise´s financial health (Vochozka, 2010, p. 675).
Artificial neural networks are able to carry out non-linear
statistical modelling in these models, and thus provide a suitable
alternative for simple financial indicators including a frequently-
used logistical regression or discrimination analysis (García,
Giménez and Guijarro, 2013). These collective indexes serve
according to Vochozka (2010) mainly investors and owners of
the enterprise to determine the performance of the given
enterprise from the perspective of value creation, or serve
creditors in predicting whether the enterprise is not reaching
bankruptcy in the nearest future.
The issue of artificial neural networks related to enterprise
evaluation belongs among rather young subjects. Their
development and especially wide application expansion is being
observed since 1980´s (Du Jardin, 2010). Nowadays, still new
types of networks keep appearing, as well as massive
development of information technologies and computing
technologies for their implementation (Synek, Hoffmann and
Mackenzie, 2013). Neural networks belong, together with fuzzy
sets, expert systems, gnostic theory in uncertain data or genetic
algorithms, etc., among non-static higher methods of financial
analysis (Vochozka et al., 2016). Most simple indicators, but
also mathematical-statistical or non-statistical methods prove
shortcomings that implement a certain level of inaccuracy into
the result. They often do not take into account specific
differences – for instance, the level of inflation or tax policy.
They also have difficulties capturing causes of problems and are
not able to work with intangible assets, know-how for instance
(Kuzey, Uyar and Delen, 2014). Modern methods try to get rid
of these shortcomings. So-called higher methods of financial
analysis demand high-quality software equipment and
knowledge of mathematical statistics. Data availability and
ability to provide the model with information wanted are also
necessary. Neural networks require a certain set of data to refine
the network outcome, that is why they are not able to evaluate
enterprise performance correctly without model data (Amusan et
al., 2013). Savvidis and Ginoglou (2013) state that the
performance of artificial neural networks and of complete
company evaluation depends mainly on data. If there is enough
data it is possible to claim that artificial neural network is the
correct choice for enterprise evaluation (Ghodsi, Zakerinia and
Jokar, 2011).
The main advantage predicting artificial neural networks for
application in economy is, according to Vesely (2011) the ability
to work with non-linear data, too. In complex enterprise
evaluation, there are countless non-linear relations or structures
(Ciobanu and Vasilescu, 2013). A non-linear enterprise
evaluation model assembled on the basis of neural networks may
stimulate economic phenomena better, and its results are
objective, relatively exact and have a practical referential value
(Zhang and Zhong, 2015, p. 178). This advantage of artificial
neural networks is confirmed also by Wu et al. (2011) claiming
that networks are able to learn, and having learned, they are able
to capture the hidden, and even strongly non-linear
dependencies. They use distributed parallel processing of
information and reach high speed processing of large data
volumes. According to Mostafa (2009), artificial neural network
models have a great potential in classifying the relative
enterprise performance thanks to their robustness and algorithm
modelling flexibility.
A model based on artificial neural networks, evaluating
enterprise performance may be set in many ways. Input data is
often represented by significant items which are usually a part of
a balance sheet or profit and loss statement. Shi, Bian and Zhang
(2010) for instance, classify the value of total enterprise assets,
the amount of workers, main enterprise costs, net fixed assets,
net profit, main enterprise income, total asset turnover indicator
and income per share among input information. Zhang and
Zhong (2015) use up to 20 enterprise financial indicators for the
purposes of education and testing of back propagation type
neural network samples. They include, for instance, net income
- 232 -