AD ALTA
JOURNAL OF INTERDISCIPLINARY RESEARCH
The year of establishing the company,
Business contact receivables,
Short-term financial property in thousands CZK,
Other current assets,
Other short-term obligations,
Revenues for sale of goods in thousands CZK,
Return interest in thousands CZK.
At the same time, it is suitable to submit the result to the
modification of vector weights between individual vectors. The
aim is an increase in the efficiency of the obtained model. With
regard to the amount of variables, this is rather an attempt. In
this case, a significant increase in classification (prediction)
accuracy has not occurred, not in one of the three most suitable
neural structures (MLP 15:15-54-66-4:1, Linear 84:86-4:1, and
Linear 90:98-4:1).
4 Conclusion
When processing this paper, three neural structures were
determined and described, showing similar positive results (MLP
15:15-54-66-4:1, Linear 84:86-4:1 a Linear 90:98-4:1),
respectively the best results from the 10 preserved neural
structures. Based on the reached reliability values, it is
impossible to unambiguously determine the one neural structure
with the best parameters. If we focus on the calculated error,
preferring both linear networks, while, during a detailed testing
the Linear 90:98-4:1 network will be preferred. On the other
hand, the other tool, confusing matrix, pretends a completely
different result. All four situations, i.e. that the enterprise is not
going bankrupt, it is going bankrupt in two years and i tis going
bankrupt in the future, are best predicted by the multilayer
perceptron MLP 15:15-54-66-4:1 network.
With regard to the usability of the model and minimal deviations
from the other two models which were being taken into account
we may judge that the best results are shown by the MLP 15:15-
54-66-4:1. Thanks to its parameters, we may claim that the result
is applicable in practice. Via MLP 15:15-54-66-4:1 we will
judge the ability of a building enterprise in the CZ to survive
possible financial distress.
A comparison of the obtained model to already renowned and
used bankruptcy models (such as Altman indexes, the Neumaier
IN indexes, and Taffler index) occurs. A range of expert papers
has dealt with their predictive value, such as Vochozka (2010).
Generally, it may be concluded that they show the following
shortcomings (Vochozka, 2010):
Assumption of bipolar dependent variables,
Data choice method in model enterprises,
Assumption of data stationarity and instability,
Choice of independent variables,
The use of annual financial statements,
Time dimension.
In case of individual variables, it is clear that their absolute size
is in question. Nevertheless, we must understand the result not as
individual variables, but as a file of variables within which the
individual variables interact. To make it clearer, we are only
indicating the most significant variables. But also among them
there are quantities characterizing the enterprise size – e.g.
´revenues for sales of goods in thousands CZK´. Less significant
variables, such as numbers of employees or total assets are not
mentioned.
Suggested Solution: the neural structure shows some
shortcomings, as well as models constructed via multiple
discrimination analysis do. Some are eliminated, specifically the
assumption of bipolar dependent variables (the model works
with four values), the choice of independent variables (the model
has allowed using all available variables – it was not necessary
to eliminate some), and the time dimension (the enterprise´s
neural networks, respectively recording lines do classify. Thus, it
is possible to work with the history of individual enterprises).
Regarding the specific comparison, we may refer to Vochozka
(2010), Delina and Packova (2013), Kubenka and Slavicek
(2014) or Mertlova (2015). The suggested neural structure shows
significantly better values of prediction, 15-20% higher accuracy
on the average.
Interesting results have been brought by sensitivity analysis.
Based on their results we may arrive to these partial conclusions:
The year of the enterprise´s establishment tells us that an
enterprise with a longer history has gained greater experience,
and thus will be probably able to survive possible financial
distress.
An enterprise, which generates greater business-contact
receivables will, with a greater c, be able to survive possible
financial distress. This claim is relatively courageous, as we are
unable to analyse claim structure out of financial statements.
They may be expired claims, or even impregnable claims.
Business-contact claims may be a false positive indicator.
A higher value of short-term financial property expressed in
thousands CZK indicates the enterprise´s ability to survive
probable financial distress.
An enterprise that creates other higher current assets will
probably survive possible financial distress.
A higher value of short-term obligations means a higher ability
of the enterprise to survive possible financial distress. Optically,
it may seem to be a false positive indicator. But, if we look at the
result through money supply creation, the indicator makes sense.
The enterprise, thanks to a longer due date of its obligations,
accumulates short-term financial property. The indicator thus
complements point No. 3 more than appropriately.
Higher revenues for sale of goods in thousands CZK create an
assumption that the enterprise will probably survive possible
financial distress. It is interesting that the overview also includes
sale revenues in the building industry section. It might be
assumed revenues for own products and services will be
calculated with a greater probability. Nevertheless, the indicator
is certainly not false positive.
A higher value of return interest in thousands CZK means a
higher ability of the enterprise to survive possible financial
distress. Even in this case it may be a matter of a rather
negligible item in profit and loss statement within a building
enterprise. But, the value again is certainly not false positive.
The determined aim to create, via neural networks, a model,
which will be able to predict a building-enterprise´s ability to
survive possible financial distress, has been fulfilled.
Literature
1. Al-Shayea, Q. K. and G. A. El-Refae. Evaluation of Banks
Insolvency Using Artificial Neural Networks. Proceedings of the
11th WSEAS international conference on Artificial Intelligence,
Knowledge Engineering and Data Bases. 2012, 113-118.
2. Amusan, L. M., Ch. K. Ayo, T. Mosaku, T. Fagbenle, O.
Tunji-Olayeni, J. Owolabi, I. Omuh, and A. Ogunde. Neural
Network-Ant Colony Optimization Model of Residential
Building Project Cost: Exploratory Approach. Creating global
competitive economies: 2020 vision planning implementation,
Int. Business information management assoc-ibima. 2013, USA,
1-3, 2104-2117.
3. Bishop, C. Neural Networks for Pattern Recognition. Oxford:
University Press.1995, 482 p. ISBN 0198538642.
4. Ciobanu, D. and M. Vasilescu. Advantages and Disadvantages
of Using Neural Networks for Predictions. Economic Sciences
Series. 2013, 13(1), 444-449.
5. Delina, R. and M. Packova. Prediction bankruptcy models
validation in Slovak business environment. E & M Ekonomie a
Management. 2013, 16(3), 101-112.
6. Du Jardin, P. Predicting bankruptcy using neural networks and
other classification methods: The influence of variable selection
techniques on model accuracy. Neurocomputing. 2010, 73(10-
12), 2047-2060.
- 236 -