You are on page 1of 18

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/266563181

Artificial Neural Network Model for Business Failure Prediction of Distressed


Firms in Colombo Stock Exchanged

Article · September 2014

CITATION READS

1 206

1 author:

Mudith Sujeewa
University of Kelaniya
10 PUBLICATIONS   2 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

How to maange the fraud risk View project

THE NEW FRAUD TRIANGLE THEORY - INTEGRATING ETHICAL VALUES OF EMPLOYEES View project

All content following this page was uploaded by Mudith Sujeewa on 08 October 2014.

The user has requested enhancement of the downloaded file.


ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

P ublis he d b y: S out h A s ia n A c ade m ic R es e arc h J our nals

ACADEMICIA:
An International
Multidisciplinary
Research Journal
( A D o u b le B l i n d R e fe r e e d & R e v ie we d I nt e r na t io na l J o ur na l)

ARTIFICIAL NEURAL NETWORK MODEL FOR BUSINESS FAILURE


PREDICTION OF DISTRESSED FIRMS IN COLOMBO STOCK
EXCHANGE

Gamlath Mohottige Mudith Sujeewa*

*Department of Accountancy,
University of Kelaniya.

ABSTRACT

The prediction of business failures is one of the important analyses in corporate


world. The research study constructs business failure prediction model for
companies listed in Colombo Stock Exchange (CSE). Data were gathered from
eighty two companies listed in CSE for the sample period of 2011 to 2013. To
achieve the research objectives, Artificial Neural Network (ANN) analysis was
employed as measure of analysis. The results of ANN model for holdout sample, the
model correctly predicted 80% while Type I Error and Type II Error were 40 % and
0% respectively for the year 2011. Thus, the model correctly predicted 90% while
Type I Error and Type II Error were 20% and 0% respectively for the year 2012.
Furthermore, the model correctly predicted 90% while Type I Error and Type II
Error were 20% and 0% respectively for the year 2013. It was found that artificial
neural network analysis predict business failures of firms listed in CSE with high
level of accuracy by identifying non liner relationships of variables of business
failure prediction.

KEYWORDS: Artificial Neural Network, Business Failure Prediction, Type I error,


Type II error.
______________________________________________________________________________

INTRODUCTION

There is a huge demand for prediction of business failures since the results of business failure
leads to heavy losses both financially and non-financially. Thus a model that could accurately
predict business failure in time would be quite useful to managers, shareholders, the government,
suppliers, customers, employees amongst other stakeholders. Further, accurate business failure

South Asian Academic Research Journals


http://www.saarj.com
140
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

prediction models would be extremely valuable to many industry sectors, particularly financial
investment and lending (Sprengers, 2005). The potential value of such models is emphasized by
the extremely costly failure of high-profile companies in the recent past. Therefore the prediction
of business failure is an important and challenging issue that has served as the impetus for many
academic studies over the past couple of decades. Consequently, a significant interest has been
generated in business failure prediction within academia as well as in the finance industry.

A special case of this problem is predicting if firms become ‘bankrupt’ or experience ‘financial
distress’ or not fail. If firms fail then there can be substantial costs. In addition, the bankruptcy
prediction problem has application to a broad base of decision makers. For example, bankers
considering loaning money are interested in determining if the firm to which the loan is being
made will be able to repay the loan. The ability to predict bankruptcy could be used to alert
auditors about the likelihood of a firm failing, possibly influencing potential litigation against the
auditor. Further, regulators could use a bankruptcy prediction model to decide whether a
particular bank should be closed or at least receive increased attention and guidance.
Accordingly, the prediction of bankruptcy probably is one of the most important business
decision making problems facing auditors, consultants, management and government policy
makers etc.Similarly, the basic objective of financial statements is to provide information useful
for making economic decisions. Financial ratios are a popular tool for users who are owner,
shareholders, creditors, investors, government and other users to evaluate the financial condition
and performance of a company by computing a set of key financial ratios from financial
statements.

Corporate failures are a common problem of developing and developed economies (Altman,
1982). Failure is not an impulsive outcome and it grows constantly in stages. There are unique
characteristics of failure in firm’s financial levels prior achieving total failures. An impulsive
protective effort could be accommodated if the company is foreseen to be proceeding in the
direction of potential bankruptcy and this can help dispel the financial distress to all investors
and decrease the costs of bankruptcy. It is clear that notwithstanding the tremendous amount of
research that has gone into this topic around the world, the dilemma of prediction can by no
means be absolutely interpreted. This is because prediction is not an actual science and at best,
purely a calculated estimate (Bhunia, Khan, Mukhuti, 2011).

The number of business failures and bankruptcies increased together with the increase in
corporate distress. Four generic terms that are generally found in literature for corporate distress
are failure, insolvency, default and bankruptcy (Sprengers, 2005).

Failure means that the realized rate of return on invested capital, with allowances for risk
consideration, is significantly and continually lower than prevailing rates on similar investments.
Somewhat different criteria has also been utilized, including insufficient revenues to cover costs
and cases of the average return on investment being below the firm’s cost of capital. A firm
could be an economic failure for many years without failing to cover its current obligations
because of the absence of legally enforceable debt (Sprengers, 2005).

South Asian Academic Research Journals


http://www.saarj.com
141
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

Insolvency is a term used in a more technical way. It indicates lack of liquidity, so it is more cash
based, which happens when a company cannot meet its financial obligations. Technical
insolvency most often is the cause of formal bankruptcy declaration (Sprengers, 2005).

Bankruptcy comes along when the insolvency of a company becomes critical, when the total
liabilities of a company exceed a fair value valuation. Default is another condition that is
inescapably associated with distress. A firm is not always immediately in default when it misses
a loan payment or its interest payments. Finally the term bankruptcy will be discussed. A firm
can go bankrupt when the total liabilities exceed a fair value of the total assets of that firm
(Sprengers, 2005).

Many business decision-making problems involve the classification of data into different groups.
The business community uses bankruptcy prediction models directly in order to identify
companies that are likely to experience future financial distress, including bankruptcy. The
development of financial analysis models to predict business failures can be thought of as ‘early
warning systems’, which proves to be very helpful for managers and relevant authorities who can
prevent the occurrence of failures.

COLOMBO STOCK EXCHANGE (CSE)

The CSE operates the only share market in Sri Lanka and is responsible for providing a
transparent and regulated environment where companies and investors can come together. The
CSE is a company that is limited by guarantee established under the Laws of Sri Lanka. The CSE
is licensed by the Securities and Exchange Commission of Sri Lanka (SEC) and is a mutual
exchange consisting of 15 Members and 14 Trading Members.

It is important to pay attention on the current Sri Lankan Regulatory Framework on Non-
Performing Listed Companies (NPLCs) since this study deals with business failure prediction.
The consultation paper issued by Securities and Exchange Commission, (2011), discussed
following observations pertaining to Non Performing Listed Companies during the course of
review of Annual Reports of the Listed Companies.

Companies continue to be listed even though they are incurring heavy losses for several years
and the reserves being depleted.

The dividends have been paid by a company, which had ceased business operations, partly out of
the issued capital violating the principles of the Companies Act No 17 of 1982.

Companies remain listed even though significant assets have been disposed or operations of the
business have ceased.

The auditors of certain companies have modified their reports on the company’s or group’s
ability to continue as a going concern and have issued a disclaimer of opinion on the financial
statements.

The consultancy paper is concluded that according to the current regulatory requirements, neither
Securities and Exchange Commission Act nor Listing Rules of the Colombo Stock Exchange

South Asian Academic Research Journals


http://www.saarj.com
142
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

lays down a mechanism to closely monitor Non Performing Listed Companies. However, with
the introduction of new Companies Act in 2007, a new provision was introduced to deal with the
companies with serious loss of capital. As per the Section 220 of the Companies Act No 07 of
2007, a director of a company shall call an extraordinary general meeting of shareholders if the
company’s net assets are less than half of its stated capital. Further, it is noted that Non
Performing Listed Companies (NPLC’s) of Colombo Stock Exchange should be closely
regulated in order to maintain a fair and transparent market.

REVIEW OF LITERATURE

A wide-ranging literature review was conducted by the researcher to identify the direction of this
study and to understand the broader perspectives of business failure prediction. Throughout this
chapter, researcher review literature over the development of business failure prediction models.
Further, researcher uses and highlights the findings of the studies conducted by the other
researchers all over the world covering different sectors to build, support and rationalize the
arguments, concepts and constructs derived in the current study.

Factors which can contribute to the understanding of corporate bankruptcy can be found in the
fields of Finance and theory of Business Management. Probably because of the diversity of the
problem, we still have not seen a fully successful model to predict bankruptcy. A possible
explanation could be the fact that all firms are unique and information about firms is limited and
most of the time not uniform.

Altman (1968) developed pioneer business failure prediction model, and this study is based on
a sample of 66 publicly traded, manufacturing firms using Multiple Discriminant Analysis
(MDA). The model correctly predicted financial failure for 95% of the firms, one year prior to
their demise. Accuracy decreases to 83% two years prior to insolvency.

An Analysis of the Applications of Neural Networks (NN) in Finance for the last ten years
(Fadlalla & Lin, 2001), neural networks has been increasingly applied to various areas of
finance. Neural networks are more often applied on the assets side than on the liabilities side of
the balance sheet. Some major characteristics of the areas of these applications are their data
intensity, unstructured nature, high degree of uncertainty, and hidden relationships. Most of the
applications use the back propagation model with one hidden layer. In most of these applications,
neural networks outperformed traditional statistical models, such as discriminant and regression
analysis. Furthermore, these applications have shown significant success in financial practice, for
example, in bankruptcy prediction, forecasting T-bills, in asset management, in portfolio
selection, and in fraud detection.

Recognizing Financial Distress Patterns Using a Neural Network Tool, (Coats and Fant, 1993),
they compare the results of the established MDA approach which makes a prior assumptions
about the discriminating variables against a new, more robust neural network approach. Their
objective was to showcase the advantages of the NN for recognizing complex patterns in data.
Test results suggest that the NN approach is more effective than MDA for pattern classification.

Leary (1998) provided a ‘meta analyses’ of the use of neural networks to predict corporate
failure in the study, Using Neural Networks to Predict Corporate Failure. Fifteen papers are
South Asian Academic Research Journals
http://www.saarj.com
143
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

reviewed and compared in order to investigate ‘what works and what doesn’t work’. The studies
are compared for their formulations including aspects such as the impact of using different
percentages of bankrupt firms, the software they used, the input variables, the nature of the
hidden layer used, the number of nodes in the hidden layer, the output variables, training and
testing and statistical analysis of results. Then the findings are compared across a number of
dimensions, including, similarity of comparative solutions, number of correct classifications,
impact of hidden layers, and the impact of the percentage of bankrupt firms. Neural Networks
are an effective tool for supporting managerial decision making. Neural Networks formulations
used for bankruptcy have generated results that are at least as good as those generated by
discriminant analysis, logit, probit Analysis.

Ko, Blocher, Lin (1986) used Composite Rule Induction System (CRIS) to derive rules for the
prediction of corporate financial distress in Taiwan. Using step-wise regression, this study
identified five financial ratios, total liabilities/total assets, quick assets/current liabilities,
sales/fixed assets, margin/sales and cash dividend per share that could effectively predict a
financial distress. Then, it applied CRIS and Artificial Neural Network (ANN) to develop the
financial distress prediction model using these five financial ratios. The empirical results indicate
that both CRIS and ANN outperform the logistic model. Although both CRIS and ANN perform
rather well, CRIS has the advantage that the derived rules are easier for humans to learn.

A hybrid financial analysis model for business failure prediction (Huang, Tsai, Yen , Cheng,
2007), Business failure prediction can be approached differently by using machine learning
techniques, whose prediction accuracy has shown more superiors to other traditional statistical
methods. In this paper, the authors propose a hybrid financial analysis model composed of static
and trend analysis models (i.e. financial structure, credit standing, operating standing,
profitability, and short-term credit standing). The experimental results report that the proposed
model using a back-propagation neural network produces good performance of prediction
accuracy and outperforms other models including discriminant analysis, decision trees, and the
back-propagation neural network alone.

Atiya (2001) reviewed the topic of bankruptcy prediction, with emphasis on neural-network
(NN) models. Second, he developed an NN bankruptcy prediction model. Inspired by one of the
traditional credit risk models developed by Merton, he proposed novel indicators for the NN
system. He showed that the use of these indicators in addition to traditional financial ratio
indicators provides a significant improvement in the (out-of-sample) prediction accuracy
(from 81.46% to 85.5% for a three-year-ahead forecast).

SIGNIFICANCE OF THE STUDY

The ability to predict bankruptcy is critical for many users of financial statements. As discussed
such users include banks, investors, credit rating agencies, underwriters, auditors and regulators.
During a period of financial and economic crisis, the importance of using a model to predict
bankruptcy and flag warning signs as early as possible becomes increasingly important.

South Asian Academic Research Journals


http://www.saarj.com
144
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

The firm’s creditors and the owners of the firm are the two primary groups of interest when a
firm is in corporate distress. These two groups both have an extremely large importance in the
evaluation of the bankruptcy reorganization process.

The meaningful variable in determining firm’s stability and viability varies from territory to
territory. In developed economies, most of the users utilized results from the research done in
developed economies without making the certain accommodation to regional situations, which
will result in misapplication (Bhunia, Khan, Mukhuti, 2011) .This study contributes to a growing
body of literature on the relationship between financial ratios and risk of bankruptcy in various
countries. Most studies that have examined the predictive ability of the Altman Model were
performed in developed countries such as US and UK. In contrast, the study focuses on a
developing country, Sri Lanka. The financial strategy may impact the way firms are managed
which, in turn, may influence the financial ratios of firms and their importance. Additionally, in
light of the globalization and favorable macro economic variables in Sri Lanka, many foreign
investors tend to invest in the Colombo Stock Exchange.

As discussed, Majority of studies on Altman Z score model and other firm specific factors for
predicting distress prediction have been carried out in developed countries. According to the
investigations, there were very few studies in this research area in developing or emerging
economies. On the other hand, it is very hard to find a study on developing a model for
predicting distress prediction for a market like Colombo Stock Exchange. This study will tend to
fill that gap up to a certain extent. The outcomes of this study will provide a framework or
guidance for monetary policy makers, for the local as well as foreign investors, owners, debt
holders and managers of the businesses, financial analysts etc.

This study enables to identify whether reorganization process is required to restructure the firm
in a way that the firm’s financial situation will stabilize and that no other financial problems will
occur in the near future. Besides financial institutions and related organisations, individuals such
as investors, auditors and regulators will benefit from such a study, thus, emphasizing the need
for this study to focus on the Sri Lankan context. This would ensure the development of a unique
ANN model that will set the standard for effectiveness and accuracy in business failure
prediction in Sri Lanka.

OBJECTIVES

The objectives of the study are as follows.

- To develop a financial model for prediction of distressed and non-distressed companies


listed in CSE.

- To find out non linear relationship between distressed and non-distress companies listed
in CSE.

- Test the model and make suitable recommendations.

South Asian Academic Research Journals


http://www.saarj.com
145
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

RESRCH METHODOLOGY

The research methodology covers the research design describing the way of operationlztion,
measurement of the variables in the study and sample of companies selected from the CSE.
Further it discusses sampling procedures which describes what rational is to be used to select the
decided sample, types of data which is to be used, data collection and finally the way of
analyzing the collected data.

TABLE 1- OPERATION OF THE VARIABLES

Conceptual Definition Operational Definition Scale

Profitability ratios

Earnings before Interest and A measure of the productivity of the firm’s


Ratio
Tax / Total Assets assets

Earnings Per Share


A measure of income or loss per share Ratio

Net Profit Ratio A measure of net profit is being generated from


Ratio
sales

Return on Assets
A measure of net income relative to its assets Ratio

Return on Equity A measure of the firm generating return to its


Ratio
equity providers.

Financial Leverage
A measure if cumulative
Retained Earnings / Total
Profitability over its total assets. Ratio
Assets

Book value of equity / Total A measure of Book value of equity relative to


Ratio
Liabilities total liabilities

A measure of the relative size of the firm’s debt


Debt Ratio Ratio
and its ability to pay off the debt.

Debt to Equity A measure of the proportion of the debt relative


Ratio
to equity financing for the firm

South Asian Academic Research Journals


http://www.saarj.com
146
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

Liquidity Ratio
A measure of the net liquid assets of the firm
Ratio
Working Capital / Total relative to the total assets.
Assets

Current Asset / Current A measure of the ability of the firm to meet its
Ratio
Liabilities short-term financial obligations.

Operating Cash Flow Ratios


A measure of how well current liabilities are
covered by the cash flow generated from a Ratio
Cash Flow from Operation /
company's operations.
Current Liabilities

Earnings before Interest and


A measure of the Earnings before Interest and
Tax / Cash Flow from Ratio
Tax relative to cash flow from operation.
Operation

Activity Ratio
A measure of the firm’s asset utilization Ratio
Sales/Total Assets

Pi is
Probability of Financial
An estimated probability of financial health between 0
Health
and 1

RESEARCH DESIGN

As explained above, researcher selected fourteen financial ratios to predict business failures and
identify the companies which have been defined as distressed and non-distressed. Then it is
calculated the value of each ratio for the selected sample for the year 2011, 2012 and 2013
separately. The researcher constructs the model based on that information and validates the
model. As discussed, Figure 1 explains the design of research.

FIGURE 1 – RESEARCH DESIGN

Identification of distressed Develop the Check the


and non distressed business failure validity of
companies and calculating prediction the model
the financial ratios for each model
company

South Asian Academic Research Journals


http://www.saarj.com
147
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

SAMPLE AND SAMPLING PROCEDURE

The sample adopted in this study was selected from companies listed in CSE and classified as
financially distressed and non distressed firms for the period of 2011 to 2013 in accordance with
the provided definition in this research. A paired sample design technique was employed in this
study. Each financially distressed public listed company in both sub-samples was matched with a
non-financially distressed listed company, listed during the same period. The matching criteria
include same industry, closest asset size and distressed year to minimize bias in selecting the
control group or holdout sample for the development of financial prediction model.

The secondary data collected from annual reports of listed companies in Colombo Stock
Exchange. Accordingly, the sample consists of eighty two (82) distressed and non-distressed
companies from 2011 to 2013. Following criteria should be satisfied to identify as distressed
firms and non distressed firms (Sori & Jalil, 2009).

DISTRESSED FIRMS

The companies had been incurring losses for three consecutive years or more.

NON DISTRESSED FIRMS

The companies had been incurring profits which are in same industry comparative to the selected
distressed firms.

Then, it will be matched each distressed firm by another non-distressed firm in the same
industry. This produces a matched paired sample.

The sample is divided into two sub-samples in Artificial Neural Network analysis, the analysis
sample which is used as an estimation of Artificial Neural Network analysis and the holdout
sample used for validation purposes.

DATA ANALYSIS

The test methodology involves computing the financial ratios selected in the study for both the
distressed and non-distressed samples. For each firm in the distressed sample, will be computed
the financial ratios for each of the three years before distress. Then, the same financial ratios for
the same years are computed for the control sample firms as well.

ANN model is constructed for the business failure prediction of distressed firms listed in CSE.
Moreover ANN is used to identify the non liner relationship of variables of business failures
(Fadlalla & Lin, 2001; Leary, 1998; Boritz, Kennedy, Sun, 2007). ANN analysis is employed by
using DataEngine4 software package.

ARTIFICIAL NEURAL NETWORK (ANN)

ANN is a collection of simple processors connected together. Each processor can only perform a
very straightforward mathematical task, but a large network of them has much greater

South Asian Academic Research Journals


http://www.saarj.com
148
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

capabilities and can do many things. Artificial neural networks are computerized intelligence
systems that simulate the inductive power and behavior of the human brain. They have the ability
to generalize and see through noise and distortion, to abstract essential characteristics in the
presence of irrelevant data, and to provide a high degree of robustness and fault tolerance.

NETWORK MODEL

The network models have distinguishing characteristics, such as type of network topology (for
example, recurrent or non-recurrent), number of hidden layers in the network, and the learning
rules of the network. Topology describes the structure of the neural network, that is, how layers
of neurons (also called processing elements) are organized and connected. Those neurons that
receive input are called the input layer, and those that generate output are called the output layer.
Both the input and the output layers interact with the outside world, is called a hidden layer. A
model has only one input layer and one output layer, but may have any number of hidden layers.

NEURAL COMPUTING

Neural computing is a computer system that consists of a network of interconnected units called
Artificial Neurons (ANs). ANs are organized in layers inside the network. The first layer is the
input layer, and the last is the output layer. Hidden layers exist between the input and output
layers, and there can be several hidden layers for complex applications. Computer programs
process the training sample to identify the relationships between input and output data. Neural
computing is more adaptive to the real world situation because it is not subject to distribution
constraints. This advantage makes neural computing an appealing tool for developing prediction
models because the variance-covariance matrices of failed and non-failed firms are often not
equal, and financial data seldom follow the multivariate normal distribution, each of which is a
violation of the MDA assumptions (Coats and Fant, 1993).

TRAINING AND TESTING THE DATA

The objective of the training process is to autonomously learn the relationship between the
output and patterns in the input, and to incrementally capture that knowledge in a unique
structure of hidden nodes and connection weights which produce correct categorizations. This
process of working toward accurate mappings is called "convergence”. According to a
mathematical theorem proved by Kolmogorov (1950s), restated for back propagation neural net
works by Hecht Nielsen (1980s) and for Cascor neural networks by Lacher (1990), the network
will always eventually figure out how to make perfect mappings of the data on which the NN is
being trained. (Coats and Fant, 1993; Gunawardana, 2009)

Initially, the network is organized into two layers of nodes. The input layer presents the variety
of patterns. In this research experiments, for example, each pattern represents a different firm's
set of fourteen financial ratios for a given year of operation, i.e., fourteen input nodes. The output
layer is a one-node response (either 0 for distressed or +1 for non-distressed) associated with a
given input pattern. It is the association of pattern and response that the network attempts to learn
through its internal nodes. The actual number of hidden nodes and the connection weights are
determined by the neural net. The trained data set is tested by the developed neural network and
verifies the error rate to setup the most fitted neural network model.
South Asian Academic Research Journals
http://www.saarj.com
149
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

BACK PROPRGATION LEARING ALGRORITHM

It is very difficult to assign an appropriate weight for a classification task, especially when there
is little information about the population distribution. A general solution is to let the network
learn the task by training it with examples. A typical learning algorithm will search through the
space of W for a set of weights offering the best fit with the given examples. Notable learning
algorithms are the perception convergence procedure and the back-propagation algorithm (Coats
and Fant, 1993; Atiya, 2001). The back-propagation learning algorithm, designed to train a feed-
forward network, overcomes some of perception’s limitations by making it possible to train a
multiple-layer network. It is an effective learning technique that is capable of exploiting the
regularities and exceptions in the training sample.

RESULTS AND DISCUSSION

This chapter discusses the data presentation, results and the analysis of the results in relation to
the development of business failure prediction model. Analysis was carried DataEngine4
software package with Artificial Neural Network. Artificial Neural Network model was
constructed for the business failure prediction of distressed firms listed in CSE. Moreover ANN
is used to identify the non liner relationship of variables of business failures. Finally, business
failure prediction model constructed in ANN was validated.

THE ARCHITECTURE OF ARTIFICIAL NEURAL NETWORK MODEL

This network architecture consists of fourteen neurons in the input layer with one hidden layer
and two neurons in the output layer which represent the distressed and non-distressed firms. The
learning method is the back-propagation with weight initialization -0.1 and 0.1. Learning
parameters of NN shows in table 2.

TABLE 2 – LEARNING PARAMETERS OF NN

Layer Learning rate Momentum Weight decay

1st Hidden 0.1 0.1 0.9999

Output 0.1 0.1 0.9999

This network has been chosen as the most suitable network for generalization due to its Small
Root Mean Square Error (RMSE) and a high percentage in good patterns as compared to other
different architecture networks. After selecting the best possible architecture, the network was
trained to reduce the error between the neural network output and the target output. The aim of
training is to find a set of connection weights that will minimize the mean squared error
forecasting error in the shortest possible training time. Training data is a process to minimize the
RMSE between actual and estimated output values with a set of suitable connection weights.

South Asian Academic Research Journals


http://www.saarj.com
150
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

Error back-propagation training algorithm is used to minimize the output error by updating the
weights. The training process weights and biases are modified and converge towards values
representing a solution of the problem. Training stops when the RMSE achieves an equal or less
than value of 0.01 or the percentage of good patterns achieved is close to 100%. The networks
become unstable and oscillation occurs when the learning rate is higher than 0.10 and the
momentum rate is between 0.85 and 1.0. The RMSE becomes bigger when the higher learning
rate is added to the network due to that network being unable to learn or store the knowledge as
the learning rate is too fast.

TABLE 3- THE RESULT OF ANN ANALYSIS

Error Types 2011 2012 2013

Epoch 2,000,000 500,000 100,000

Maximum Training Error 0.0025 0.0006 0.0010

RMS Training Error 0.0080 0.0001 0.0003

Maximum Test Error 0.0143 0.0004 0.0194

RMS Test Error 0.0048 0.0001 0.0003

Table 3 shows that the RMS training error and RMS test error are 0.0080 and 0.0048
respectively for the year 2011. Further it indicates 0.0001 RMS training error and 0.0001 RMS
test error for the year 2012. RMS training error and RMS test error are 0.0003 and 0.0003
respectively for the year 2013. The fourteen neurons in the input layer with one hidden layer and
two neurons in the output layer of that network topology which has produced lowest training and
testing errors for all three years. Therefore both the RMS Training and Testing errors are
minimized and closed to zero and accordingly the best fitted network is obtained. Accordingly
figure 2, 3 and 4 shows how learning curves of the model behave by minimizing and making
closed to zero both RMS Training and Testing errors for the year 2011, 2012 and 2013.

South Asian Academic Research Journals


http://www.saarj.com
151
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

FIGURE 2 - LEANING CURVE OF 2011

FIGURE 3 - LEANING CURVE OF 2012

South Asian Academic Research Journals


http://www.saarj.com
152
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

FIGURE 4 - LEARNING CURVE OF 2013

VALIDATION OF THE ANN MODEL

Holdout samples were used to test the robustness of the neural network model developed for
business failure prediction. Table 4 shows results of forecasts made by the trained networks for
holdout sample.

TABLE 4 - THE ESTIMATION RESULTS OF ANN MODEL

2011 2012 2013

Correctly Classified as Distressed and Non- 80% 90% 90%


Distressed

Type I Error 40% 20% 20%

Type II Error - - -

As per the table 4, the model correctly predicted 80% while Type I Error and Type II Error are
40 % and 0% respectively for the year 2011. Accordingly, the model correctly predicted 90%
while Type I Error and Type II Error are 20% and 0% respectively for the year 2012.

South Asian Academic Research Journals


http://www.saarj.com
153
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

Furthermore, the model correctly predicted 90% while Type I Error and Type II Error are 20%
and 0% respectively for the year 2013.

FINDINGS AND CONCLUSION

It is important to understand the early warning indicators and implications of corporate financial
distress. If the stakeholder can predict a company is on its way to a financial distress, they can
take a necessary action in time. Similarly, it is vitally important for an auditor to be able to assess
whether or not a company is a going concern in preparing the audit report. The significant
consequence of corporate financial distress has generated a lot of research interest and numerous
methods have been applied to develop prediction models. Accordingly, this study was conducted
with the intention of construction of business failure prediction model.

Financial distress in companies can lead to problems that can reduce the efficiency of
management. As maximizing firm value and maximizing shareholder value cease to be
equivalent managers who are responsible to shareholders might try to transfer value from
creditors to shareholders. The result is a conflict of interest between bondholders (creditors) and
shareholders. As a firm's liquidation value slips below its debt, it is the shareholder's interest for
the company to invest in risky projects which increase the probability of the firm's value to rise
over debt.

Failure is not an impulsive outcome and it grows constantly in stages. There are unique
characteristics of failure in firm’s financial levels prior achieving total failures. An impulsive
protective effort could be accommodated if the company is foreseen to be proceeding in the
direction of potential bankruptcy and this can help allay the financial distress to all investors and
abate the costs of bankruptcy. It is clear that notwithstanding the tremendous amount of research
that has gone into this topic around the world; the predicament of prediction can by no means be
absolutely interpreted. This is because prediction is not an actual science and at best, purely a
calculated estimate (Bhunia, Khan, Mukhuti , 2011).

The study was performed to develop a ANN model for prediction of distressed and non-
distressed companies listed in CSE, to find out non linear relationship of variables in business
failure prediction between distressed and non-distress companies listed in CSE and test the
model and make suitable recommendations. To achieve above mention objectives the researcher
identify distress and non-distress companies listed in CSE for the period of 2011 – 2013.
Accordingly, the researcher developed ANN model to predict business failure prediction.
Moreover ANN model was facilitated to identify the non linear relationship of business failure
prediction variables and to validate the model.

The network has been chosen as the most suitable network for generalization due to its RMSE
and a high percentage in good patterns as compared to other different architecture networks. As
per the results of ANN analysis, the RMS training error and RMS test error are 0.0080 and
0.0048 respectively for the year 2011. Further it indicates 0.0001 RMS training error and 0.0001
RMS test error for the year 2012. RMS training error and RMS test error are 0.0003 and 0.0003
respectively for the year 2013. The fourteen neurons in the input layer with one hidden layer and
two neurons in the output layer of that network topology which has produced lowest training and

South Asian Academic Research Journals


http://www.saarj.com
154
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

testing errors for all three years. Therefore both the RMS Training and Testing errors are
minimized and closed to zero and accordingly the best fitted network is obtained.

Holdout samples were used to test the robustness of the neural network models developed for
business failure prediction. As per the results of forecasts made by the trained networks for
holdout sample, the model correctly predicted 80% while Type I Error and Type II Error are 40
% and 0% respectively for the year 2011. Accordingly, the model correctly predicted 90% while
Type I Error and Type II Error are 20% and 0% respectively for the year 2012. Furthermore, the
model correctly predicted 90% while Type I Error and Type II Error are 20% and 0%
respectively for the year 2013. It is found that artificial neural network model predict business
failures of firms listed in CSE with high level of accuracy by identifying non liner relationships
of variables of business failure prediction.

REFERENCES

Altman E.I (1968), Financial ratios, discriminant analysis and the prediction of corporate
bankruptcy, Journal of Finance.

Altman, E. I. (1982), Accounting Implications of Failure Prediction Models. Journal of


Accounting, Auditing, and Finance: 4-31.

Atiya A.F. (2001) Bankruptcy Prediction for Credit Risk Using Neural Networks: A Survey and
New Results, IEEE Transactions on Neural Networks, VOL. 12, No. 4.

Bhunia A., Khan S.I.U., Mukhuti S. (2011) Prediction of Financial Distress - A Case Study of
Indian Companies, Fakir Chand College, University of Calcutta, Diamond Harbour, University
of Gour Banga, West Bengal, India.

Boritz J.E., Kennedy D.B., Sun J.Y., (2007) Predicting Business Failures in Canada, School of
Accountancy, University of Waterloo, Canada

Coats P.K. and Fant L.F. (1993), Recognizing Financial Distress Patterns Using a Neural
Network Tool ,College of Business, Florida State University, Tallahassee, Florida.

Corporate Affairs Division, (2011), Regulation of Non Performing Listed Companies,


Consultation Paper 07, Securities and Exchange Commission of Sri Lanka

Fadlalla A. ,Lin C.H., (2001) An Analysis of the Applications of Neural Networks in Finance,
Department of Computer and Information Science Cleveland State University, Cleveland State
University.

Gunawardana K.D. (2009) An introduction to Artifiial Neureal Networks for Accounting and
Finance Modeling, University of Sri Jayewardenepura, Sri Lanka.

Huang S.M., Tsai C.F., Yen D.C., Cheng Y.L. ( 2007) A hybrid financial analysis model for
business failure prediction, Department of Accounting and Information Technology, National

South Asian Academic Research Journals


http://www.saarj.com
155
ISSN:2249-7137 Vol. 4, Issue 9, Sept. 2014 Impact Factor: SJIF 2013=5.099

Chung Cheng University, Taiwan, Department of Decision Sciences and Management


Information Systems, Miami University, USA.

Ko L.J., Blocher E.J., Lin P.P. (1986), Prediction of Corporate Financial Distress: An
Application of the Composite Rule Induction System, The International Journal of Digital
Accounting Research, Vol. 1, No. 1, pp. 69-85, Durham Technical Community College. USA.

Leary D.E.O, (1998), Using Neural Networks to Predict Corporate Failure, University of
Southern California, USA.

Sori Z.M. and Jalil H.A. (2009), Financial Ratios, Discriminant Analysis and the Prediction of
Corporate Distress, Faculty of Economics and Management, Universitiy of Putra Malaysia
Faculty of Economics and Management, Universitiy of Putra Malaysia.

Sprengers M.A. (2005) Bankruptcy Prediction using Classification and Regression Trees Faculty
of Economics, Erasmus University Rotterdam.

South Asian Academic Research Journals


http://www.saarj.com
156

View publication stats

You might also like