Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic
The construction sector is one of the main pillars of an advanced economy. It is the first sector to indicate potential national economic problems. In a similar way it is the first sector to show signs of recovery when an economy is coming out of recession or crisis. The aim of this article is to ap...
Збережено в:
| Опубліковано в: : | Математичне моделювання в економіці |
|---|---|
| Дата: | 2015 |
| Автори: | , |
| Формат: | Стаття |
| Мова: | English |
| Опубліковано: |
Інститут телекомунікацій і глобального інформаційного простору НАН України
2015
|
| Теми: | |
| Онлайн доступ: | https://nasplib.isofts.kiev.ua/handle/123456789/131787 |
| Теги: |
Додати тег
Немає тегів, Будьте першим, хто поставить тег для цього запису!
|
| Назва журналу: | Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| Цитувати: | Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic / M. Vochozka, Z. Rowland // Математичне моделювання в економіці. — 2015. — № 3(4). — С. 62-76. — Бібліогр.: 14 назв. — англ. |
Репозитарії
Digital Library of Periodicals of National Academy of Sciences of Ukraine| id |
nasplib_isofts_kiev_ua-123456789-131787 |
|---|---|
| record_format |
dspace |
| spelling |
Vochozka, M. Rowland, Z. 2018-03-30T19:41:23Z 2018-03-30T19:41:23Z 2015 Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic / M. Vochozka, Z. Rowland // Математичне моделювання в економіці. — 2015. — № 3(4). — С. 62-76. — Бібліогр.: 14 назв. — англ. 2409-8876 https://nasplib.isofts.kiev.ua/handle/123456789/131787 004.942 The construction sector is one of the main pillars of an advanced economy. It is the first sector to indicate potential national economic problems. In a similar way it is the first sector to show signs of recovery when an economy is coming out of recession or crisis. The aim of this article is to apply a neural network to be able to predict potential financial problems in construction companies in the Czech Republic. Будівельна галузь є однією з найважливіших галузей у всіх розвинених економіках світу. Вона першою вказує на потенційні проблеми національної економіки і першою ж сигналізує про ознаки відновлення в економіці, яка виходить з рецесії або навіть кризи. Мета цієї статті полягає у використанні нейронних мереж для прогнозування потенційних фінансових труднощів будівельних компаній в Чеській Республіці. Строительная промышленность является одним из основных столпов всех развитых экономик мира. На ней в первую очередь отражаются возможные проблемы национальной экономики. На ней же одной из первых проявляется возможное улучшение состояния экономики, выходящей из состояния рецессии или даже кризиса. Целью статьи является использование нейронной сети для прогнозирования возможных финансовых затруднений строительных предприятий Чешской Республики. en Інститут телекомунікацій і глобального інформаційного простору НАН України Математичне моделювання в економіці Математичні та інформаційні моделі в економіці Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic Прогнозування розвитку будівельних компаній за допомогою нейронних мереж на основі даних Чеської Республіки Прогнозирование развития строительных компаний с помощью нейронных сетей на основе данных Чешской Республики Article published earlier |
| institution |
Digital Library of Periodicals of National Academy of Sciences of Ukraine |
| collection |
DSpace DC |
| title |
Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic |
| spellingShingle |
Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic Vochozka, M. Rowland, Z. Математичні та інформаційні моделі в економіці |
| title_short |
Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic |
| title_full |
Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic |
| title_fullStr |
Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic |
| title_full_unstemmed |
Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic |
| title_sort |
prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the czech republic |
| author |
Vochozka, M. Rowland, Z. |
| author_facet |
Vochozka, M. Rowland, Z. |
| topic |
Математичні та інформаційні моделі в економіці |
| topic_facet |
Математичні та інформаційні моделі в економіці |
| publishDate |
2015 |
| language |
English |
| container_title |
Математичне моделювання в економіці |
| publisher |
Інститут телекомунікацій і глобального інформаційного простору НАН України |
| format |
Article |
| title_alt |
Прогнозування розвитку будівельних компаній за допомогою нейронних мереж на основі даних Чеської Республіки Прогнозирование развития строительных компаний с помощью нейронных сетей на основе данных Чешской Республики |
| description |
The construction sector is one of the main pillars of an advanced economy. It is the first sector to indicate potential national economic problems. In a similar way it is the first sector to show signs of recovery when an economy is coming out of recession or crisis. The aim of this article is to apply a neural network to be able to predict potential financial problems in construction companies in the Czech Republic.
Будівельна галузь є однією з найважливіших галузей у всіх розвинених економіках світу. Вона першою вказує на потенційні проблеми національної економіки і першою ж сигналізує про ознаки відновлення в економіці, яка виходить з рецесії або навіть кризи. Мета цієї статті полягає у використанні нейронних мереж для прогнозування потенційних фінансових труднощів будівельних компаній в Чеській Республіці.
Строительная промышленность является одним из основных столпов всех развитых экономик мира. На ней в первую очередь отражаются возможные проблемы национальной экономики. На ней же одной из первых проявляется возможное улучшение состояния экономики, выходящей из состояния рецессии или даже кризиса. Целью статьи является использование нейронной сети для прогнозирования возможных финансовых затруднений строительных предприятий Чешской Республики.
|
| issn |
2409-8876 |
| url |
https://nasplib.isofts.kiev.ua/handle/123456789/131787 |
| citation_txt |
Prediction of the future development of construction companies by means of artificial neural networks on the basis of data from the Czech Republic / M. Vochozka, Z. Rowland // Математичне моделювання в економіці. — 2015. — № 3(4). — С. 62-76. — Бібліогр.: 14 назв. — англ. |
| work_keys_str_mv |
AT vochozkam predictionofthefuturedevelopmentofconstructioncompaniesbymeansofartificialneuralnetworksonthebasisofdatafromtheczechrepublic AT rowlandz predictionofthefuturedevelopmentofconstructioncompaniesbymeansofartificialneuralnetworksonthebasisofdatafromtheczechrepublic AT vochozkam prognozuvannârozvitkubudívelʹnihkompaníizadopomogoûneironnihmerežnaosnovídanihčesʹkoírespublíki AT rowlandz prognozuvannârozvitkubudívelʹnihkompaníizadopomogoûneironnihmerežnaosnovídanihčesʹkoírespublíki AT vochozkam prognozirovanierazvitiâstroitelʹnyhkompaniispomoŝʹûneironnyhseteinaosnovedannyhčešskoirespubliki AT rowlandz prognozirovanierazvitiâstroitelʹnyhkompaniispomoŝʹûneironnyhseteinaosnovedannyhčešskoirespubliki |
| first_indexed |
2025-11-25T20:43:44Z |
| last_indexed |
2025-11-25T20:43:44Z |
| _version_ |
1850530948382195712 |
| fulltext |
~ 62 ~
Математичне моделювання в економіці, №3, 2015
UDK 004.942
M. VOCHOZKA, Z. ROWLAND
PREDICTION OF THE FUTURE DEVELOPMENT OF CONSTRUCTION
COMPANIES BY MEANS OF ARTIFICIAL NEURAL NETWORKS ON
THE BASIS OF DATA FROM THE CZECH REPUBLIC
Abstract. The construction sector is one of the main pillars of an advanced
economy. It is the first sector to indicate potential national economic
problems. In a similar way it is the first sector to show signs of recovery
when an economy is coming out of recession or crisis.
The aim of this article is to apply a neural network to be able to predict
potential financial problems in construction companies in the Czech
Republic.
Data on all construction companies in the Czech Republic over the period
2003-2013 were used for the modelling of the neural network. The data file
contained 67,000 records. These records included both financial statements
and non-accounting data (e.g. data on company employees).
The following networks were used for modelling the neural network:
a linear network, a probabilistic neural network (PNN), a generalised
regression neural network (GRNN), a radial basis function network (RBF),
a three-layer perceptron network (TLP) and a four-layer perceptron
network (FLP).
The analysis resulted in a concrete model of an artificial neural network.
The neural network is able to determine with more than ninety per cent
accuracy whether a company is able to overcome potential financial
problems, within how many years a company might go bankrupt, or
whether a company might go bankrupt within one calendar year. The text
also includes the basic statistical characteristics of the examined sample
and the achieved results (sensitivity analysis, confusion matrix, etc.).
The model can be exploited in practice by construction company managers,
investors looking for a suitable company for capital investment,
competitors, etc.
Keywords: construction company, financial problems, prediction, artificial
neural network, model.
Introduction
Traditional methods supporting financial decision making include «consumer
credit scorecards» (Mester, 1997, Reichert at al., 1983, Rosenberg and Gleit, 1994,
[8]) and discrimination models for the assessment of a company´s financial health
(Altman at al., 1995, Reichert at al., 1983, [1]). Both are basically linear models
with multiple variables. A neural network is a flexible non-parametric modelling
tool for designing a prediction model based on logical links (formulae) between
historic data. Schemes in historic data may often exist in both spatial and time
planes. A conventional MLP NN (Multiparametric Neuron Network) focused on
the retrospective revelation of error algorithms is arranged in such a way as to
define constant schemes that do not vary with time.
Ó M. Vochozka, Z. Rowland, 2015
~ 63 ~
Математичне моделювання в економіці, №3, 2015
Neural structures have recently come to the fore as the preferred method for
predicting the collapse of a society (Kumar and Ravi, 2007, [5]). The work by
Odom a Sharda (1990, [9]) was one of the first studies to apply neural networks to
the issue of bankruptcy prediction. Odom and Sharda used Altman’s relative
financial indices as inputs for neural networks to which they subsequently applied
their methods. These methods included MDA to compare a certain number of US
companies, both solvent and insolvent, whereby the data used for bankrupt
companies came from their last financial statements prior to declaring bankruptcy.
They took into account 128 companies and performed several trials. During these
trials the proportion of declining and prospering companies in the examined sample
where changed. The method of artificial neural networks achieved a Type I
classification accuracy within the range 77.8% – 81.5% (depending on the
examined sample) and Type II accuracy within the range 78.6 – 85.7%. The
corresponding results for MDA for Type I accuracy were within the range 59.3% –
70.4% and for the Type II accuracy 78.6% – 85.7%.
Foreign sources offer an overview of various neural network types, including
MLP NN (Multiparametric Neuron Networks), probability neural networks (PNN),
auto-associative neural networks (AANN), self-organizing maps (SOM), learning
vector quantization (LVQ) and cascade correlation (Cancor). A lot of these studies
focused on comparing neural networks with classic statistical techniques like factor
analysis, logit analysis and various forms of discrimination analyses. A lot of these
cases show that neural networks provide a more accurate prediction of bankruptcy
than parametric statistical approaches. However, the results are also often diverse.
Tam and Kiang (1992, [13]) compared different types of models when the
application of neural networks to bankruptcy examination first began. They studied
MDA, LA, K-nearest neighbour (KNN), decision tree classification algorithm
(ID3), single-layer neural networks and multi-layer neural networks. The used
neural networks were the standard of back-propagation (BPNN). The multi-layer
neural network was the most suitable for the prediction of bankruptcy based on
relative financial indices one year prior to bankruptcy. In comparison, logit-
analysis achieved better results for the two-year period prior to bankruptcy. When
Salchenberger, Cinar and Lash (1992, [12]) analysed bankruptcies of thrifts, they
found that BPNN substantially outclassed logit analysis. For example, an 18-month
prediction LA achieved a classification accuracy of 83.3% – 85.4% (depending on
some threshold values), while NN reached 91.7%. Coats and Fant (1993, [4]) found
when they compared BPNN and MDA that BPNN was better in general, although
it had larger variances in the classification of the results depending on the time
period used. Altman et al. (1995) compared the BPNN and MDA methods in the
field of failure prediction on 1,000 Italian companies. According to their
conclusions MDA showed slightly better results than BPNN in its predictions for
the one-year period. Boritz, Kennedy and Albuquerque (1995, [3]) compared
numerous techniques, including various BPNN, LA and MDA procedures. The
comparison results were inconclusive. In a lot of studies BPNN showed better
prediction ability of company failure than MDA and the other aforementioned
techniques. It is for this reason that new hybrid techniques and genetic algorithms
have recently come to the fore. Lee et al. (1996, [6]) tested combinations of models
like MDA, ID3, SOM and BPNN. They compared the predictive abilities on
Korean companies and drew the conclusion that SOM together with neural
~ 64 ~
Математичне моделювання в економіці, №3, 2015
networks achieved the best results. Zhang et al. (1999, [14]) used a fivefold scheme
of mutual validity control on a group of manufacturing companies and compared
BPNN with LA for bankruptcy prediction. BPNN was again substantially better
than LA. McKee and Greenstein (2000, [6]) developed an approach based on
decision trees. They tested this approach on a sample of American companies
based on data a year prior to bankruptcy. Their method achieved better results than
MDA and BPNN for Type II classification errors but worse results for Type I
classification errors. Atiya (2001, [2]) developed new indices obtained from the
securities market. The application of these indices together with traditional relative
financial indices brought substantial improvements in the accuracy of bankruptcy
predictions. The predictions were based on financial data three years prior to
bankruptcy.
The aim of this article is to exploit neural networks for the prediction of
potential financial problems in construction companies in the Czech Republic.
1. Material and Methodology
The information on companies given below comes from the Albertina database.
The data covers all the construction companies which operated on the Czech
market between the years 2003-2013 and which fall under the CZ-NACE
classification under Section F. The following activities are included – construction
of buildings, civil engineering and specialized construction activities.
The file contains a total of 67,492 rows of data in columns labelled:
– company name (15,189 companies);
– region;
– list of annual financial statements between 2003-2013;
– list of additional data.
MS Excel was used for the preparation of the data file. The data (financial as
well as non-financial) for each company for each year was always presented on one
line. The file containing the 67,492 records of construction companies for
individual years, including 100 characteristics on each company, was imported into
Statistica software by DELL. The data was subsequently processed by an
intelligent problem solver.
An artificial neural structure was sought that would be able to classify each
company on the basis of the input data into one of the following groups:
– solvent company;
– company that will go bankrupt in the current year;
– company that will go bankrupt in 2 years;
– company that will go bankrupt in the future.
First we determined the characteristics of the individual companies. We had to
define the output category quantity. These were defined on the basis of the values
presented in the column «final status» in the excel sheet. The category output
quantities are «Financial Statement Extent», «Financial Statement Structure» and
the «Auditor’s Statement». All the items shown from the financial statements are
continuous quantities.
~ 65 ~
Математичне моделювання в економіці, №3, 2015
Once this exercise was completed, 1,000 artificial neural structures1 were
generated, of which the 10 most suitable were retained. For the model, linear neural
networks, probabilistic neural networks, radial basis function neural networks,
three-layer perceptron networks and four-layer perceptron networks, were utilised.
For the radial basis function neural network we used 1 up to 15,998 hidden
neurons. The 2nd layer of the three-layer perceptron network contained 1 up to 100
hidden neurons. The 2nd and the 3rd layers of the four-layer perceptron network
both contained 1 up to 100 hidden neurons.
2. Results – Production Function
1,000 artificial neural networks were generated on the basis of the set parameters.
10 artificial neural networks showing the best characteristics were retained for
further assessment and subsequent processing. The results of the analysis are given
in Table 1.
Table 1 – Models of artificial neural networks showing the best characteristics2
Index
Profile Train
Perf.
Select
Perf.
Test
Perf.
Train
Error
Select
Error
1 MLP 1:4-33-61-4:1 0.943588 0.945240 0.944865 1.430218 1.541917
2 MLP 2:7-88-63-4:1 0.038504 0.037695 0.038320 1.002620 1.040046
3 MLP 15:15-54-66-4:1 0.943182 0.945052 0.944427 0.586112 0.591007
4 Linear 84:86-4:1 0.944432 0.945490 0.944865 0.160929 0.163963
5 Linear 90:98-4:1 0.944307 0.945490 0.944615 0.160849 0.162028
6 PNN 88:93-31997-4:1 0.944245 0.946052 0.945427 0.162359 0.160338
7 PNN 87:92-31997-4:1 0.944245 0.946052 0.945427 0.162360 0.160335
8 RBF 61:69-328-4:1 0.943870 0.945490 0.944802 0.159620 0.158623
9 RBF 61:69-359-4:1 0.943713 0.945115 0.944802 0.159637 0.158612
10 RBF 61:69-360-4:1 0.943870 0.945427 0.945052 0.159324 0.158588
Index Test Error Training/Members Inputs Hidden(1) Hidden(2)
1 1.453480 BP100,CG20,CG0b 1 33 61
2 1.014261 BP100,CG20,CG0b 2 88 63
3 0.583500 BP100,CG20,CG0b 15 54 66
4 0.160875 PI 84 0 0
5 0.160743 PI 90 0 0
6 0.160657 88 31,997 0
7 0.160657 87 31,997 0
8 0.159088 SS,KN,PI 61 328 0
1Unless the improvement in the individual trained networks is significant the training of neural networks can be
shortened.
2A linear neural network is indicated as Linear, probabilistic neural network as PNN, generalized regression neural
network as GRNN, radial basis network as RBF and multi-layer perceptron network as MLP.
~ 66 ~
Математичне моделювання в економіці, №3, 2015
Continued (Table 1)
Index Test Error Training/Members Inputs Hidden(1) Hidden(2)
9 0.159930 SS,KN,PI 61 359 0
10 0.159093 SS,KN,PI 61 360 0
Multiple perceptron networks with two hidden layers were retained among the
ten best networks, see lines 1 – 3 of the table. Two linear neural networks, two
probabilistic neural networks and three radial basis function neural networks then
follow.
Figure No. 1 shows a schematic illustration of a multiple perceptron network
with two hidden layers, namely MLP 1:4-33-61-4:1.
Figure 1 – Graph of artificial neural network (MLP 1:4-33-61-4:1)
The first layer (from the left), in the form of triangles, represent the inputs for
the models, namely continuous and category qualities. White indicates a positive
quantitative value, whereas red a negative one. Two hidden layers follow. The
resulting classification is finally defined by the output layer, whereby a company is
allocated to one of the four groups identified in Section 1 of this article. The
percentage success rate of the training, validation and verification sample should in
general be almost the same in order to be able to say that the network is of a good
quality and has the characteristics required to apply it in practice. For the network
represented above the values for all three samples were above the 94% level.
Figure 2 shows a graphic illustration of a multi-layer perceptron network
MLP 2:7-88-63-4:1.
~ 67 ~
Математичне моделювання в економіці, №3, 2015
Figure 2 – Graph of artificial neural network (MLP 2:7-88-63-4:1)
The values of all the parts of this network achieve a level of approximately 4%.
This means that the network is of poor quality and not applicable in practice.
Figure 3 shows a graphic illustration of a multi-layer perceptron network
MLP 15:15-54-66-4:1.
Figure 3 – Graph of artificial neural network (MLP 15:15-54-66-4:1)
In the network represented above the values of the training, validation and
verification data oscillate above the 94% level.
Figure 4 shows a graphic illustration of a linear neural network Linear 84:86 -
4:1.
~ 68 ~
Математичне моделювання в економіці, №3, 2015
Figure 4 – Graph of artificial neural network (Linear 84:86-4:1)
In this case the network has two types of neurons - input and output layers.
There is no hidden layer. In the case of the linear network represented above the
values of the training, validation and verification data oscillate above the 94%
level.
Figure 5 shows a graphic illustration of a linear neural network Linear 90:98 -
4:1.
Figure 5 – Graph of artificial neural network (Linear 90:98-4:1)
In the case of the linear network represented above the values of the training,
validation and verification data oscillate above the 94% level.
Figure 6 shows a graphic illustration of a probabilistic neural network
PNN 88:93-31997-4:1.
~ 69 ~
Математичне моделювання в економіці, №3, 2015
Figure 6 – Graph of artificial neural network (PNN 88:93-31997-4:1)
A probabilistic neural network works with one hidden layer of neurons.
According to the calculations the central layer of the neural network represented
above hides 31,997 neurons. In this case the values of the training, validation and
verification data oscillate above the 94% level.
Figure 7 shows a graphic illustration of a probabilistic neural network
PNN 87:92-31997-4:1.
Figure 7 – Graph of artificial neural network (PNN 87:92-31997-4:1)
In the case of the probabilistic network represented above the values of the
training, validation and verification data oscillate above the 94% level.
Figure 8 shows a graphic illustration of a radial basis function neural network
RBF 61:69-328-4:1.
~ 70 ~
Математичне моделювання в економіці, №3, 2015
Figure 8 – Graph of artificial neural network (RBF 61:69-328-4:1)
A radial basis function neural network works with one hidden layer of neurons.
According to calculations the central layer of the neural network represented above
hides 328 neurons. In this case the values of the training, validation and
verification data oscillate above the 94% level.
Figure 9 shows a graphic illustration of a radial basis function neural network
RBF 61:69-359-4:1.
Figure 9 – Graph of artificial neural network (RBF 61:69-359-4:1)
In the case of the neural network represented above the values of the training,
validation and verification data oscillate above the 94% level.
Figure 10 shows a graphic illustration of a neural network of radial basis
function RBF 61:69-360-4:1.
~ 71 ~
Математичне моделювання в економіці, №3, 2015
Figure 10 – Graph of artificial neural network (RBF 61:69-360-4:1)
In the case of the radial basis function neural network represented above the
values of the training, validation and verification data oscillate above the 94%
level.
Nevertheless, it was not possible to determine anything substantive from the
graphic illustration. The confusion matrix (see Table 2) that was subsequently
drawn up helped to clarify the situation.
The table describes the success rate and consequently the predictions of the
individually generated artificial neural networks. The confusion matrix calculates
the absolute value of the correctly classified quantities. This enabled the
identification of the network with the highest success rate in predicting the future
development of the companies in the examined sample.
The neural networks 1, 2, 6 and 7 showed relatively good results in predicting
solvent companies, however the results were very poor when it came to identifying
the potential bankruptcy of a company. The other networks were also able to
predict solvency (with slightly lower levels of accuracy). However, of these other
networks some were better in predicting the potential bankruptcy of a company in
the current business year, within two years, or further into the future.
The results are therefore not definitive. However, if the networks are compared
on the basis of their prediction success rate for the individual classified groups,
neural network 5 is the best and the most applicable in practice. It is a linear neural
network Linear 90:98-4:1. It is of interest that 90 input data entered the calculation.
Compared to the other generated networks this is the highest number of inputs.
This means that a possible sensitivity analysis should determine the result
sensitivity accurate to ninety inputs (of various weights). It is also of interest that
for example the first two perceptron networks worked with only two inputs. In the
end they were able to accurately define solvent companies. However, the low
number of inputs meant they were unable to predict the future potential bankruptcy
of a company. The third function, which took 15 inputs into account provided
better results. Once again, the results were not optimal and it is therefore not
applicable in practice.
~ 72 ~
Математичне моделювання в економіці, №3, 2015
Table 2 – Confusion matrix (neural networks 1 – 10)
T.
So
lv
en
t
co
m
pa
ny
T.
B
an
kr
.
in
cu
rre
nt
y
ea
r
T.
B
an
kr
.
in
th
e
fu
tu
re
T.
B
an
kr
.
in
tw
o
ye
ar
s
T.
Ba
nk
r.
ne
xt
y
ea
r
S.
So
lv
en
t
co
m
pa
ny
S.
B
an
kr
.
in
cu
rre
nt
y
ea
r
S.
B
an
kr
.
in
th
e
fu
tu
re
S.
B
an
kr
.
in
tw
o
ye
ar
s
S.
Ba
nk
r.
ne
xt
y
ea
r
1 2 3 4 5 6 7 8 9 10 11
Solvent
company.1 30192 1232 569 0 4 15121 603 270 3 0
Bankr. in
current year.1 0 0 0 0 0 0 0 0 0 0
Bankr. in the
future.1 0 0 0 0 0 0 0 0 0 0
Bankr. in two
years.1 0 0 0 0 0 0 0 0 0 0
Solvent
company.2 30192 1232 569 0 4 15121 603 270 3 0
Bankr. in
current year.2 0 0 0 0 0 0 0 0 0 0
Bankr. in the
future.2 0 0 0 0 0 0 0 0 0 0
Bankr. in two
years.2 0 0 0 0 0 0 0 0 0 0
Solvent
company.3 30177 1229 569 0 4 15118 598 270 3 0
Bankr. in
current year.3 1 2 0 0 0 1 0 0 0 0
Bankr. in the
future.3 14 1 0 0 0 2 5 0 0 0
Bankr. in two
years.3 0 0 0 0 0 0 0 0 0 0
Solvent
company.4 30182 1198 566 0 4 15110 588 269 3 0
Bankr. in
current year.4 10 34 0 0 0 11 15 1 0 0
Bankr. in the
future.4 0 0 3 0 0 0 0 0 0 0
Bankr. in two
years.4 0 0 0 0 0 0 0 0 0 0
Solvent
company.5 30176 1196 564 0 4 15109 587 268 3 0
Bankr. in
current year.5 16 36 2 0 0 12 16 2 0 0
Bankr. in the
future.5 0 0 3 0 0 0 0 0 0 0
Bankr. in two
years.5 0 0 0 0 0 0 0 0 0 0
Solvent
company.6 30192 1213 567 0 4 15120 589 270 3 0
Bankr. in
current year.6 0 19 0 0 0 0 14 0 0 0
~ 73 ~
Математичне моделювання в економіці, №3, 2015
Continued (Table 2)
1 2 3 4 5 6 7 8 9 10 11
Bankr. in the
future.6 0 0 2 0 0 1 0 0 0 0
Bankr. in two
years.6 0 0 0 0 0 0 0 0 0 0
Solvent
company.7 30192 1213 567 0 4 15120 589 270 3 0
Bankr. in
current year.7 0 19 0 0 0 0 14 0 0 0
Bankr. in the
future.7 0 0 2 0 0 1 0 0 0 0
Bankr. in two
years.7 0 0 0 0 0 0 0 0 0 0
Solvent
company.8 30190 1222 568 0 4 15120 598 270 3 0
Bankr. in
current year.8 2 10 0 0 0 1 5 0 0 0
Bankr. in the
future.8 0 0 1 0 0 0 0 0 0 0
Bankr. in two
years.8 0 0 0 0 0 0 0 0 0 0
Solvent
company.9 30190 1227 568 0 4 15118 602 270 3 0
Bankr. in
current year.9 2 5 0 0 0 3 1 0 0 0
Bankr. in the
future.9 0 0 1 0 0 0 0 0 0 0
Bankr. in two
years.9 0 0 0 0 0 0 0 0 0 0
Solvent
company.10 30181 1212 569 0 4 15115 594 270 3 0
Bankr. in
current
year.10
11 20 0 0 0 6 9 0 0 0
Bankr. in the
future.10 0 0 0 0 0 0 0 0 0 0
Bankr. in two
years.10 0 0 0 0 0 0 0 0 0 0
X
.S
ol
ve
nt
co
m
pa
ny
X
.B
an
kr
.
in
cu
rre
nt
y
ea
r
X
.B
an
kr
.
in
th
e
fu
tu
re
X
.B
an
kr
.
in
tw
o
ye
ar
s
X
.B
an
kr
.
ne
xt
y
ea
r
I.S
ol
ve
nt
co
m
pa
ny
I.B
an
kr
.
in
cu
rre
nt
y
ea
r
I.B
an
kr
.
in
th
e
fu
tu
re
I.B
an
kr
.
in
tw
o
ye
ar
s
I.B
an
kr
.
ne
xt
y
ea
r
1 2 3 4 5 6 7 8 9 10 11
Solvent
company.1 15115 613 263 1 5 0 0 0 0.00 0.00
Bankr. in
current year.1 0 0 0 0 0 0 0 0 0.00 0.00
Bankr. in the
future.1 0 0 0 0 0 0 0 0 0.00 0.00
~ 74 ~
Математичне моделювання в економіці, №3, 2015
Continued (Table 2)
1 2 3 4 5 6 7 8 9 10 11
Bankr. in two
years.1 0 0 0 0 0 0 0 0 0.00 0.00
Solvent
company.2 15115 613 263 1 5 0 0 0 0.00 0.00
Bankr. in
current year.2 0 0 0 0 0 0 0 0 0.00 0.00
Bankr. in the
future.2 0 0 0 0 0 0 0 0 0.00 0.00
Bankr. in two
years.2 0 0 0 0 0 0 0 0 0.00 0.00
Solvent
company.3 15107 611 263 1 5 0 0 0 0.00 0.00
Bankr. in
current year.3 2 1 0 0 0 0 0 0 0.00 0.00
Bankr. in the
future.3 6 1 0 0 0 0 0 0 0.00 0.00
Bankr. in two
years.3 0 0 0 0 0 0 0 0 0.00 0.00
Solvent
company.4 15101 599 263 1 5 0 0 0 0.00 0.00
Bankr. in
current year.4 14 14 0 0 0 0 0 0 0.00 0.00
Bankr. in the
future.4 0 0 0 0 0 0 0 0 0.00 0.00
Bankr. in two
years.4 0 0 0 0 0 0 0 0 0.00 0.00
Solvent
company.5 15099 600 262 1 5 0 0 0 0.00 0.00
Bankr. in
current year.5 16 12 1 0 0 0 0 0 0.00 0.00
Bankr. in the
future.5 0 1 0 0 0 0 0 0 0.00 0.00
Bankr. in two
years.5 0 0 0 0 0 0 0 0 0.00 0.00
Solvent
company.6 15115 604 263 1 5 0 0 0 0.00 0.00
Bankr. in
current year.6 0 9 0 0 0 0 0 0 0.00 0.00
Bankr. in the
future.6 0 0 0 0 0 0 0 0 0.00 0.00
Bankr. in two
years.6 0 0 0 0 0 0 0 0 0.00 0.00
Solvent
company.7 15115 604 263 1 5 0 0 0 0.00 0.00
Bankr. in
current year.7 0 9 0 0 0 0 0 0 0.00 0.00
Bankr. in the
future.7 0 0 0 0 0 0 0 0 0.00 0.00
Bankr. in two
years.7 0 0 0 0 0 0 0 0 0.00 0.00
~ 75 ~
Математичне моделювання в економіці, №3, 2015
Continued (Table 2)
1 2 3 4 5 6 7 8 9 10 11
Solvent
company.8 15113 612 263 1 5 0 0 0 0.00 0.00
Bankr. in
current year.8 2 1 0 0 0 0 0 0 0.00 0.00
Bankr. in the
future.8 0 0 0 0 0 0 0 0 0.00 0.00
Bankr. in two
years.8 0 0 0 0 0 0 0 0 0.00 0.00
Solvent
company.9 15111 610 263 1 5 0 0 0 0.00 0.00
Bankr. in
current year.9 4 3 0 0 0 0 0 0 0.00 0.00
Bankr. in the
future.9 0 0 0 0 0 0 0 0 0.00 0.00
Bankr. in two
years.9 0 0 0 0 0 0 0 0 0.00 0.00
Solvent
company.10 15109 604 263 1 5 0 0 0 0.00 0.00
Bankr. in
current
year.10
6 9 0 0 0 0 0 0 0.00 0.00
Bankr. in the
future.10 0 0 0 0 0 0 0 0 0.00 0.00
Bankr. in two
years.10 0 0 0 0 0 0 0 0 0.00 0.00
Conclusion
The aim of this article was to apply neural networks to predict potential financial
problems in construction companies in the Czech Republic.
1000 artificial neural networks were generated on the basis of data obtained on
construction companies for the period 2003 – 2013. Ten of these neural networks
were retained for further processing. Analysis of the results of a confusion matrix
determined that neural network 5 was the most successful. It provided the optimal
ratio of prediction success rate for all the possible results, namely «solvent
company», «bankruptcy in the current year», «bankruptcy in the future» and
«bankruptcy in two years». It is a linear neural network Linear 90:98-4:1. The
artificial neural network is able to predict the future development of a construction
company in the Czech Republic with a success rate higher than 94%. Eight other
artificial neural networks achieved similar results, however they were not able to
predict the potential bankruptcy of a company. It is for this reason that it was more
suitable to select an artificial neural network that made slightly worse predictions
with regards to solvent companies but which was able to more accurately predict
(in tens of percent) potential bankruptcy.
The model is applicable in practice due to its characteristics. It can be utilised
by construction companies, financial analysts, banks, competitors or potential
investors alike.
~ 76 ~
Математичне моделювання в економіці, №3, 2015
REFERENCES
1. Altman, E. I., Eom Y. H., Kim D. W., 1995. Failure prediction: Evidence from Korea.
Journal of International Financial Management and Accounting. 6(3), 230–249.
2. Atiya, A. F., 2001. Bankruptcy prediction for credit risk using neural networks:
A survey and new results. IEEE Transactions on Neural Networks. 12(4), 929–935.
3. Boritz, J., Kennedy, D., Albuquerque, A., 1995. Predicting corporate failure using
a neural network approach, Intelligent Syst. in Accounting,Finance, and Manag. 4, 95–111.
4. Coats, P., Fant, L., 1993. Recognizing financial distress patterns using a neural network
tool. Financial Management. 22, 142–155.
5. Kumar, P., Ravi, V., 2007. Bankruptcy prediction in banks and firms via statistical and
intelligent techniques – a review. European Journal of Operational Research.180, 1–28.
6. Lee, K., Han, I., Kwon, Y., 1996. Hybrid neural network models for bankruptcy
predictions, Decision Support Syst.18, 63–72.
7. McKee, T. E., Greenstein, M., 2000. Predicting bankruptcy using recursive partitioning
and a realistically proportioned data set. J. Forecasting.19, 219–230.
8. Mester, L. J., 1997. What’s the point of credit scoring? Business Review. 3–16.
9. Odom, M., Sharda, R., 1990. A neural network model for bankruptcy prediction. In
Proc. Int. Joint Conf. Neural Networks, San Diego, CA.
10. Reichert, A. K., Cho, C. C., Wagner, G. M., 1983. An examination of the conceptual
issues involved in developing credit-scoring models. Journal of Business and Economic
Statistics. 1(2), 101–114.
11. Rosenberg, E., Gleit, A., 1994. Quantitative Methods in Credit Management:
A Survey; Operations research.42(4), 589–613.
12. Salchenberger, L., Cinar, E., Lash, N., 1992. Neural networks: A new tool for predicting
thrift failures, Decision Sciences.23, 899–916.
13. Tam, K., Kiang, M., 1992. Managerial applications of the neural networks: The case of
bank failure predictions. Management Science.38, 416–430.
14. Zhang, G. at al., 1999. Artificial neural networks in bankruptcy prediction: General
framework and cross-validation analysis. European J. Oper. Res.116, 16–32.
Стаття надійшла до редакції 09.07.2015
|