0 views

Uploaded by raqthesolid

prediction of default

- Introduction
- Future Predictions Article Checklist
- Literature Review
- Banking Risk Manag
- Credit Management
- Comparing Human and Computational Models
- Working Capital Management
- Fbr Preview Packet w o Questions
- Malomo et al
- Social Network Investor Sentiments for Predicting Stock Price Trends
- Abdelkader-Predicting Energy Demand Peak Using M5 Model Trees
- DRTA Prediction Log
- Defi Business and Products Introduction
- tmpD7CF.tmp
- deeoika 1.docx
- New Microsoft Word Document
- Data Mining Improves Pipeline Risk Assessment
- Nerolac_credit Application Formatted
- Ten Iterative Steps in Development and Evaluation of Environmental Models
- Japon

You are on page 1of 20

Predicting probability of default of Indian corporate bonds: logistic and Z-score model

approaches

Arindam Bandyopadhyay

Article information:

To cite this document:

Arindam Bandyopadhyay, (2006),"Predicting probability of default of Indian corporate bonds: logistic and Z-

score model approaches", The Journal of Risk Finance, Vol. 7 Iss 3 pp. 255 - 272

Permanent link to this document:

http://dx.doi.org/10.1108/15265940610664942

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

References: this document contains references to 37 other documents.

To copy this document: permissions@emeraldinsight.com

The fulltext of this document has been downloaded 2650 times since 2006*

Users who downloaded this article also downloaded:

Joseph Calandro, (2007),"Considering the utility of Altman's Z-score as a strategic assessment

and performance management tool", Strategy & Leadership, Vol. 35 Iss 5 pp. 37-43 http://

dx.doi.org/10.1108/10878570710819206

Obaid Saif H. Al Zaabi, (2011),"Potential for the application of emerging market Z-score in UAE Islamic

banks", International Journal of Islamic and Middle Eastern Finance and Management, Vol. 4 Iss 2 pp.

158-173 http://dx.doi.org/10.1108/17538391111144498

Fernando Castagnolo, Gustavo Ferro, (2014),"Models for predicting default: towards efficient forecasts",

The Journal of Risk Finance, Vol. 15 Iss 1 pp. 52-70 http://dx.doi.org/10.1108/JRF-08-2013-0057

Access to this document was granted through an Emerald subscription provided by emerald-srm:378861 []

For Authors

If you would like to write for this, or any other Emerald publication, then please use our Emerald for

Authors service information about how to choose which publication to write for and submission guidelines

are available for all. Please visit www.emeraldinsight.com/authors for more information.

About Emerald www.emeraldinsight.com

Emerald is a global publisher linking research and practice to the benefit of society. The company

manages a portfolio of more than 290 journals and over 2,350 books and book series volumes, as well as

providing an extensive range of online products and additional customer resources and services.

Emerald is both COUNTER 4 and TRANSFER compliant. The organization is a partner of the Committee

on Publication Ethics (COPE) and also works with Portico and the LOCKSS initiative for digital archive

preservation.

The current issue and full text archive of this journal is available at

www.emeraldinsight.com/1526-5943.htm

Logistics and

Predicting probability of default Z-score model

of Indian corporate bonds: logistic approaches

and Z-score model approaches

255

Arindam Bandyopadhyay

National Institute of Bank Management (NIBM), Kondhwe Khurd, Pune, India

Abstract

Purpose – This paper aims at developing an early warning signal model for predicting corporate

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

default in emerging market economy like India. At the same time, it also aims to present methods for

directly estimating corporate probability of default (PD) using financial as well as non-financial

variables.

Design/methodology/approach – Multiple Discriminate Analysis (MAD) is used for developing

Z-score models for predicting corporate bond default in India. Logistic regression model is employed to

directly estimate the probability of default.

Findings – The new Z-score model developed in this paper depicted not only a high classification

power on the estimated sample, but also exhibited a high predictive power in terms of its ability to

detect bad firms in the holdout sample. The model clearly outperforms the other two contesting models

comprising of Altman’s original and emerging market set of ratios respectively in the Indian context.

In the logit analysis, the empirical results reveal that inclusion of financial and non-financial

parameters would be useful in more accurately describing default risk.

Originality/value – Using the new Z-score model of this paper, banks, as well as investors in

emerging market like India can get early warning signals about the firm’s solvency status and might

reassess the magnitude of the default premium they require on low-grade securities. The default

probability estimate (PD) from the logistic analysis would help banks for estimation of credit risk

capital (CRC) and setting corporate pricing on a risk adjusted return basis.

Keywords India, Bonds, Modelling, Emerging markets

Paper type Research paper

Introduction

Corporate liabilities have default risk. There is always a chance that a corporate

borrower will not meet its contractual obligations and may renege from paying the

principal and the interest due. Even for the typical high-grade borrower, this risk is

there even though it may be small, perhaps 1/10 of 1 percent per year. Although these

risks do not seem large, they are in fact highly significant. They can even increase

quickly and with little warning. Further, the margins in corporate lending are very

tight, and even small miscalculations of default risks can undermine the profitability of

lending. But most importantly, many lenders are themselves borrowers, with high

levels of leverage. Unexpected realizations of default risk have destabilized,

decapitalized, and destroyed many internationally active lending institutions.

Following the release of the recent Reserve Bank of India (RBI) draft guidelines

(February 15, 2005) for the implementation of Basel II norms, the leading Indian banks The Journal of Risk Finance

Vol. 7 No. 3, 2006

are preparing to design appropriate internal credit risk models. The major motive is the pp. 255-272

incentive based approach for capital estimation for credit risk. In order to fetch the q Emerald Group Publishing Limited

1526-5943

early rewards of Basel II implementation, banks have to develop their own internal DOI 10.1108/15265940610664942

JRF models for credit risk. Internal models offer an opportunity for a bank to measure and

7,3 price counter-party risk and systemize risks inherent in lending. Prediction of default

probability (PD) for each borrower or group of borrowers is the key input for the

estimation of regulatory capital as well as economic capital for banks. It is also equally

important for the banking industry and financial institutions to discriminate the good

borrowers (non-defaulting) from the bad borrowers (defaulting). This will not only help

256 them in taking lending decisions but also practicing better pricing strategies to cover

against the counter party risk. While, internationally, considerable research has been

made to predict corporate default, very few attempts have been done for Emerging

Market like India.

The purpose of this paper is to build a robust framework that enables banks and

financial institutions in emerging market economy like India to classify a firm in the

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

default or non-default category based on the information of its financial variables. This

kind of model can serve as a useful tool for quick evaluation of the corporate risk

profile. Second, it can also be used to track the firms to check for their default status

over time. As a result such model can help banks to get an early warning signal about

the default status of its corporate clients. In this paper, we estimate an MDA model to

predict corporate default using a balanced panel data of 104 Indian corporations for the

period of 1998 to 2003. The financial ratios and other basic information are collected

from the Centre for Monitoring Indian Economy (CMIE) Prowess database. This

database is similar to the Compustat database in the USA. However, as we have

mentioned earlier, it is not enough to know a qualitative differentiation of

counter-parties for properly evaluating credit risk. One has to go one step further

and differentiate quantitatively between different counter-parties. Accordingly, we

also estimate probability of default (PD) of each firm for the same corporate portfolio

through logistic regression. Here we also explore the role of non-financial factors in

predicting default. For this purpose, we empirically examine whether the combined use

of financial and non-financial factors lead to more accurate PD estimates.

The rest of the paper is structured as follows. In the next section we discuss about

the data, definitions and construction of variables and hypotheses. The third section

portrays the corporate bond default rates across different rating grades in India as well

as industry wise default rates. In the fourth section, we demonstrate the development

of the Z-score model for Indian corporations, main results, and the model validation.

The fifth section presents the results and methodology of logistic analysis to predict

corporate bond defaults. Here we also compare the significance of financial and

non-financial parameters in describing default risk. We have also tested the predictive

power of logistic model. The sixth section discusses the main conclusions.

Literature survey

In theory, corporate insolvency is indicated either by fall in the asset value or due to

liquidity shortage (i.e. falls in the ability to raise capital to finance project). Therefore,

we should expect that the ratios that reflect the cash flow structure and movement of

market value of firm’s asset to be different among defaulted and solvent firms (Wilcox,

1971; Scott, 1981). Several later studies incorporated these theoretically determined

financial characteristics in the explanation of corporate default. For example, Casey

and Bartczak (1985), Gentry et al. (1985) and Aziz et al. (1988) used cash flow variables

in their model in predicting corporate failure. Opler and Titman (1994) and Asquith

et al. (1994) report that default is primarily caused by firm-specific idiosyncratic Logistics and

factors. On the other hand, researchers like Lang and Stulz (1992) and Denis and Denis Z-score model

(1994) argue for a systematic nature of bankruptcy risk. Kranhnen and Weber (2001)

presented a normative set of generally accepted rating principles that point out the approaches

necessity of links among industry risk, business risk, financial risk, management risk,

facility risk, and probability of default. Grunert et al. (2005) analyzed credit file data

from four major German banks and found empirical evidence that the combined use of 257

financial and non-financial factors leads to a more accurate prediction of future default

events than the single use of each of these factors. Other studies look at the relation

between default and the stock market. KMV Corporation of Moody (1993) using Black

and Scholes (1973) approach made an attempt to predict default in an option pricing

context; i.e. to model when the option to default has more value than the option to

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

continue.

Because of the lack of a unifying theory, there has been an explosion of different

empirical methods used to predict business failure in different markets. Statistical

models can help banks to predict default probability to get an early warning signal

about the default status of the corporate clients. Excellent reviews of the plethora of

studies can be found in Dimitras et al. (1996) and Mossman et al. (1998). The first

approach to predicting corporate failure has been to apply a statistical classification

technique called MDA to a sample containing both failed and non-failed firms. Studies

like Beaver (1966) and Altman (1968) pioneered this approach. Beaver (1966, 1968) did

univariate analysis of a number of bankruptcy predictors and set the stage for

multivariate attempts by him and others. He found that a number of indicators could

discriminate between matched samples of failed and non-failed firms for as long as five

years prior to failure. He also developed a Z-score model by using multivariate analysis

in 1968. In the same year, Altman developed his classic multivariate insolvency

prediction model (MDA) for publicly traded manufacturing firms in the USA. The

initial sample in his study is composed of 66 corporations with 33 firms in each of the

two groups distressed and solvent. The indicator variable Z-score forecasts the

probability of a firm entering bankruptcy within a two-year period (the cut-off score is

below 1.81). In the original Z-score formula for predicting bankruptcy Altman (1968)

employed working capital/total assets ratio, retained earnings/total assets ratio,

earning before interest and taxes/total assets ratio, market value of equity/book value

of total debt ratio, and sales/total assets ratio as predictor of financial health of a

company.

In Altman et al. (1977) constructed a second-generation model with several

enhancements to the original Z-score approach. The new model, which was called

ZETA, was effective in classifying bankrupt companies up to five years prior to failure

on a sample of corporations consisting of manufacturers and retailers. The ZETA

model tests included non-linear (e.g., quadratic) as well as linear discriminate models.

The non-linear model was more accurate in the original test sample results but less

accurate and reliable in holdout or out-of-sample forecasting. Subsequently, in Altman

et al. (1995) modified his Z-score model to emerging market corporations, especially

Mexican firms that had issued Eurobonds denominated in US dollars. In this enhanced

Z-score model, he dropped sales/total assets and used book value of equity for the

fourth and final variable to make it more suitable for the private firms.

JRF Several researchers influenced by the work of Altman (1968) on the application of

7,3 discriminant analysis, explored ways to develop more reliable financial distress

prediction models. Subsequently, new analytical techniques like logit or probit models

(Martin, 1977; Ohlson, 1980; Zavgren, 1985; Lennox, 1999; Westgaard and Wijst, 2001;

Grunert et al., 2005), multidimensional scaling (Mar Molinero and Ezzamel, 1991),

artificial neural networks (Tam, 1991; Wilson and Sharda, 1994/1995), multinomial

258 logit (Johnsen and Melicher, 1994), multicriteria decision aid methodology (Zopounidis

and Dimitras, 1998), etc. have been introduced to predict corporate failure in different

markets.

As one can see from the collective literature discussed above, a very large number of

empirical bankruptcy prediction models do multiple discriminant analysis (MDA)

pioneered by the seminal works of Altman (1968, 1995) based on the accounting data.

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

However, these models, though worked successfully, are very much country specific

and may not fit well with the Indian condition. The ratios as well as the weights of

MDA would differ across countries and regions (as one can see the Altman’s original

1968 model for US manufacturing firms is different from that of Emerging Market

model of 1995 based on Mexican data). Further, Z-score model only gives prediction

about the qualitative differentiation of counterparties. If banks are interested to

directly estimate the probability of default (PD), MDA analysis is not applicable since it

does not produce such probabilities.

One possible solution for banks, which intends to estimate PD directly, is the use of

limited dependent logit model (similar to Amemiya, 1981; Maddala, 1983). Unlike the

discriminant model, the logistic model has the flexibility to incorporate both the

financial as well as non-financial factors in predicting default. While the financial

ratios capture the firm specific information, the non-financial factors help to evaluate

the link of the firm with macroeconomic factors and the capability of the firm to churn

out cash flow in the required numbers.

The information on defaulted and solvent firms is collected from CRISIL’s annual

ratings of long-term bonds issued by 542 companies from 1998 until 2004. These

companies are then matched by their asset size, year, and industry affiliation.

Following these criteria in a random selection, we finally got a sample of 52 solvent

firms and 52 defaulted firms. The defaulted group is a class of manufacturers whose

long-term bonds have been defaulted between 1998 and 2003[1]. The solvent firms are

chosen on a stratified random basis drawn from CRISIL rating database. The mean

asset size of the firms in the solvent group (Rs. 948.51 Crore) is slightly greater than

that of the defaulted group (Rs. 818.62 Crore), but matching exact asset size of the two

groups is not always necessary[2]. The financial information of these 104 companies

over the period of 1998 to 2003 is obtained from the CMIE Prowess database. In

addition to the accounting data, information on ISO certification is obtained from the

Q-Prod’s directory. In order to test the predictive accuracy of our estimated sample, a

holdout sample of another 50 (25 solvent and 25 defaulted) companies is being created

for the years 2003 and 2004. Our estimated sample of 104 firms have been classified

into 11 industry categories based on their major economic activities after matching

them with the National Industrial Classification (NIC) codes (see Table I).

Logistics and

Industry dummy Industry type Number of firms

Z-score model

IND1 Food products/sugar/tea/tobacco/beverages 4 approaches

IND2 Paper 3

IND3 Textile 9

IND4 Chemical 27

IND5 Machine/electrical/computers 14 259

IND6 Metal/non-metal 23

IND7 Auto/parts 6

IND8 Power 2

IND9 Diversified 5

IND10 Service 9 Table I.

IND11 Other manufacturing 2 Industry categories of

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

prediction models should be able to relate the properties of the cash flow in

combination with the debt obligations and the movement in the asset value of the firm.

While MDA technique can successfully draw the best combination of key ratios from a

large set of financial ratios, the logistic model can capture more firm specific

idiosyncratic factors and also can relate the firm to the macro economic conditions. We

initially started with many financial ratios and finally arrive at five key ratios (in our

MDA model) that best discriminate between our sample of good and bad firms. In

order to pick up the best ratios, we looked at:

.

F and Wilk’s Lambda statistics to check the statistical significance of each

individual ratio, including determination of relative magnitude of each

independent variable (i.e. standardized values of their coefficients).

.

Within sample discriminatory power of these ratios’ best combinations.

.

Chi-square statistic as check for the overall significance of various discriminant

functions and finally.

.

Our own analytical judgment.

.

WK_TA: working capital over total assets is a measure of the net liquid assets of

the firm relative to the total capitalization. Working capital is defined as the

difference between current assets and current liabilities. Hence the ratio is a

proxy for the short-term liquidity condition of the firm.

.

CASHPROF_TA: cash profit over total assets is a measure of cash flow of the

firm. Cash profit is obtained by adding the non-cash charges such as depreciation

and amortization to the profit after tax (or net profit).

. SOLVR: solvency ratio judges the long-term solvency of a firm. Higher the

solvency ratio better is the ability of the firm to meet key term obligations and

lower will be the probability of default. The solvency ratio is computed by

dividing the firm’s total assets (net of revaluation reserves, advance tax and

miscellaneous expenses not written off) by its total borrowings plus current

liabilities and provisions minus advance payment of tax.

JRF .

OPPROF_TA: operating profit over total assets is a measure of the true

7,3 productivity of the firm’s assets. It measures the firm’s earning capability. The

higher is the ratio, the better for the company.

.

SALES_TA: the capital turnover ratio is a standard financial ratio (also used by

Altman in his original 1968 model) illustrating the sales generating ability of the

firm’s assets. It is the ratio of total sales to total assets. This ratio gives an

260 indication as to how efficiently a company is utilizing its assets. Higher is this

ratio, the better for the company.

In the logit analysis, we have included another financial variable: MVE_BVL – the

equity market value over the book value of the liabilities proxy for the firm’s asset

values. It also measures the solidity of the firm. In calculating book value of total

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

liabilities, total net worth of the firm is subtracted from the total liability of the firm.

Hence, the BVL gives the book value of total outside liabilities of the firm.

All the six ratios represent either value or income of the firms with respect to total

assets. They are all hypothesized to be either positively related to solvency or

negatively related to the firm’s default probabilities.

Following non-financial variables are taken from the existing literature about

corporate solvency:

.

Age of the firm. Age of the company since incorporation. A relatively young firm

will probably show a low retained earnings/total assets (RE/TA) ratio because it

has not had time to build up its cumulative profits (Altman, 2000). Therefore, it

may be argued that the young firm is somewhat discriminated against in this

analysis, and its chance of being classified as bankrupt is relatively higher than

that of another older firm, ceteris paribus. But, this is precisely the situation in

the real world. The incidence of failure is much higher in a firm’s earlier years

[40-50 percent of all firms that fail do so in the first five years of their existence

(Dun and Bradstreet, annual statistics)]. The age effect is thus clear: young firms

are more likely to default. We take natural log of the number of years of the firm

since incorporation as measure of firm age.

.

Group ownership. Studies covering various countries have found that firms

associated with top business groups have greater stability in the cash flows and

show better productivity as well as risk sharing than unaffiliated firms

(Gangopadhyay et al., 2001). Together with the existence of mutual debt

guarantees through group affiliation, firms may reduce the possibility of

financial distress. Some studies delineating the effect of Indian business group

affiliation on firm sales have observed that top 50 business group firms have a

better reputation advantage in the product market and are likely to export more.

They are also on average spend more advertising, marketing, distribution and

research and development (R&D), and thus have larger amount of intangible

assets (Bandyopadhyay and Das, 2005). Accordingly, we can hypothesize that

top 50 business group firms are safer firms than their non-top 50 group

counterparts.

.

ISO Quality Certification (ISOD). This dummy is taken as a product market

signal about the firm that it maintains a quality management system and is

concerned with customer expectations and satisfactions. It has been empirically

observed that ISO certified firms are successful in the product market Logistics and

(Bandyopadhyay and Das, 2005). Therefore we assume that possessing an ISO Z-score model

certificate by a firm would reduce its chance of default.

.

Control variables-industry characteristics. The industry factors affect the firms’

approaches

performance and therefore affect default as well. There are incidents of clustered

incidents of default. In order to capture the industry specific effects, our sample

firms have been classified into 11 industry dummies depending on its major 261

economic activity.

Table II presents an average one-year transition matrix of 542 corporate bond ratings

from 1995-1996 to 2004-2005. The transition probabilities captured in the matrix

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

quantify the likelihood that a company will change from one rating grade to another

within a year. The one-year average stressed probability of default (PD) of various

rating notches are reported in the last column[3]. One can see that as the credit quality

worsens (i.e. decline in the rating grades), the probability of default increases. Further,

the PD jumps sharply as soon as we move from Investment Grade Bonds to

Non-Investment Bond Grades (say from 5.17 percent for BBB to 28.93 percent for BB).

We also observe that rating stability declines as the credit quality worsens. The higher

risk in the bottom grades (mainly non-investment grades) calls for developing a

corporate default predictor model that would better capture the firm’s characteristics

and could give an early warning signal of corporate distress. In Table III, we get an

idea about corporate bond default rates by major industry groups. As expected, the

default rate varies across different industries especially in the non-investment

category. The high default rates are found in Chemical, Food Products and Tobacco

and Beverages, Machine and Electrical, Metal and Non-Metal sectors, and Paper and

Textile sectors.

Given the sample of 104 equal mutually exclusive classes of solvent and defaulted

firms over the period of 1998 to 2003, we estimate the multivariate linear discriminant

function to obtain a Z-score that will help to predict bankruptcy for new firms. The

solvent set of firms is matched with the defaulted set by asset size, industry, and year

of data. Several combinations of the variables were tried in order to estimate the best

Year 2

Year 1 AAA (%) AA (%) A (%) BBB (%) BB (%) B (%) C (%) D (%)

AA 2.54 87.57 7.93 1.05 0.60 0.15 0.00 0.15

A 0.00 4.35 79.97 9.14 3.48 0.44 0.73 1.89

BBB 0.00 0.74 5.90 67.53 14.76 2.21 3.69 5.17

BB 0.00 0.83 0.00 1.65 57.02 4.13 7.44 28.93 Table II.

B 0.00 0.00 0.00 7.41 0.00 55.56 7.41 29.63 Average one year

C 0.00 0.00 0.00 2.33 0.00 0.00 51.16 46.51 transition matrix (years

D 0.00 0.00 0.31 0.31 0.92 0.00 0.00 98.46 1995-1996 to 2004-2005)

JRF

Figures in percent

7,3 Industry IG NIG ALL

Chemical 0.88 32.00 3.98

Diverse 3.70 18.75 7.14

262 Food products/sugar/tea/tobacco/beverages 1.37 55.56 7.41

Machine/electrical/computers 2.75 37.14 7.51

Metal/non-metal 3.11 35.48 6.60

Table III. Other manufacturing 0.00 0.00 0.00

Industry wide average Paper 0.00 33.33 5.66

PD for IG and NIG and Power 0.00 0.00 0.00

pooled, 1995-1996: Service 0.22 27.78 1.25

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

assumption is that variance-covariance matrices of the two groups are statistically

identical[4]. The weights of the discriminant function are the difference of the mean

vectors of the explanatory variables for the solvent and defaulted groups. In the MDA

model our purpose is two folds: The first one is to look for predictors (financial ratios)

that lead to lowest misclassification rates within the sample and the second one is to

get improved prediction accuracy in an un-estimated holdout sample.

The discriminant analysis model involves linear combinations of the following

form:

Z ¼ b0 þ b1 X 1 þ b2 X 2 þ b3 X 3 þ . . . þ bk X k

where D ¼ discriminant score, bs ¼ discriminant coefficients or weights, and X s ¼

predictors or independent variables. The coefficients, or weights (b), are estimated so

that groups differ as much as possible on the values of the discriminant function. This

occurs when the ratio of between-group sum of squares to within-group of sum of

squares for the discriminant scores is at a maximum. Any other linear combination of

the predictors will result in a smaller ratio. The test statistics used for carrying out this

analysis is F and Wilk’s Lambda. Intuitively, a small Lambda signifies that a small

proportion of the total variance of the constituent variables is being accounted for by

within groups’ dispersions while a larger proportion of the total variance is explained

by the squared deviation of the between group means from their pooled mean. This

would be translating to a high value of the F-statistic, which means a greater chance

for the null of equal means to be rejected.

Based on the above methodology, three reduced form single equation of the original

discriminant equations and their summary results are reported in Table IV. Model 1 is

the rework of Altman’s original 1968 Z-score model. It comprises of variables that have

been used by Altman for analyzing corporate default chance in US market. The

coefficients of the variables have been re-estimated using the above data. Model 2

comprises of a set of four variables and is the revised form of Altman’s Emerging

Logistics and

Percent of correct

classification Z-score model

(within sample) approaches

Model Linear discriminant equation Good Bad

(1968) 8.158RE_ TA þ 3.73PBIT_TA 263

þ 0.037MVE_BVL þ 1.602

SALES_TA

Model 2: re-worked Z ¼ 2 1.096 þ 2.893WK_TA þ 88.2 75.9

emerging market (1995) 1.197RE_TA þ 11.711PBIT _TA

þ 0.042MVE_BVL

Model 3: new Z-score model Z ¼ 2 3.337 þ 0.736WK_TA 85.2 91

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

þ 0.864SOLVR þ Three alternative

7.554OPPROF_TA þ discriminant models for

1.544SALES_TA Indian firms

Market Score Model (1995). Model 3 of Table V is ours. This new Z-score model

comprises of five ratios. Two of these ratios are same to those in Model 1 namely

working capital to total assets (WK_TA) and sales over total assets (SALES_TA).

Three new variables are cash profits to total assets (CASHPROF_TA), solvency ratio

(SOLVR), and operating profit over total assets (OPPROF_TA). The discussion on the

expected signs of these ratios has already been done in variable definitions section.

Although Model 1 and Model 2 exhibit a reasonable high degree of classification

power, Model 3 (which is ours) has the best ability to classify the current sample of

good and bad firms. However, the robustness of Model 3 needs to be established by a

set of diagnostic tests.

The first set of tests pertains to checking the difference of the means of the two

groups, both individually and also as a whole for the entire function. From Table VI,

the magnitude of the Wilk’s Lamda and F-statistic of the individual variables (used in

Model 3) suggests that given the data, the likelihood of the means of the solvent and

defaulted groups to be equal is highly unlikely. Hence, the null hypothesis of the

equality of means with respect the same variance co-variance matrix for both the

Solvent Defaulted

(DEF ¼ 0) (DEF ¼ 1) Wilk’s Lambda for F-stat. for difference

Mean Std dev. Mean Std dev. difference in mean in mean

CASHPROF_TA 0.099 0.073 20.027 0.092 0.633 347.33 *

SOLVR 2.32 1.304 1.27 0.37 0.763 185.26 *

OPPROF_TA 0.093 0.075 20.032 0.084 0.622 363.56 *

SALES_TA 1.06 0.6 0.57 0.33 0.796 152.81 *

Notes: Total number of observations: 624. F-statistic and Wilk’s Lambda are used for discriminating

the solvent group from the defaulted group. The higher value of F and lower value of Wilk’s Lambda

indicate greater chance for the null of equal means of the two groups to be rejected. * denotes Table V.

significant at 1 percent or better Group statistics

JRF groups is rejected at 1 per cent or better level of significance. The result also fits well

7,3 for the discriminant function as a whole. The overall chi-square of the discriminant

function is 388.8 with degrees of freedom 5 (with probability. x2 ¼ 0:00) indicating a

very high overall significance of the model. Many other variables are also being tested;

however the variables used in the third model have the best combination with highest

level of discriminatory power.

264 The Z-score obtained in Model 3 may however suffer from the misclassification cost.

Misclassification may arise due to type I and type II errors. Type I error occurs when

the model incorrectly classifies a “bad” firm as “good”. Type II error arises when the

model identifies a “good” firm as “bad”. Obviously, type I error is more costly for bank

than the type II error. Therefore, it is necessary to estimate posterior probabilities to set

a benchmark to make a correct decision about the firm’s default status. For this, we

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

24:757ðOPPROF_TAÞ þ 5:354ðSALES_TAÞ

212:311ðOPPROF_TAÞ þ 3:81ðSALES_TAÞ

Using the above two equations, we obtain two scores for the same firm. In the next

step, the final Z-score obtained in Model 3 denotes a reduced form representation of the

above two discriminant equations (i.e. the difference between solvent and defaulted

score):

Z ¼ Zsolvent 2 Zdefault

or:

þ7:554ðOPPROF_TAÞ þ 1:544ðSALES_TAÞ

obtained is positive (as for him, Prob. (solvent) . Prob. (defaulted)). Similarly, the firm

with a negative Z-score is classified as one liable to default within a year horizon.

Altman (1968) (%) market (1995) (%) model (%)

Table VI.

Classification power of Defaulted no. 80 84 92

the model for the holdout correct (Type I)

dample of 25 corporations Solvent no. correct 88 84 96

for the year 2004 (Type II)

Z-score model validation Logistics and

In order to judge the correct prediction power of the discriminant function, the model Z-score model

needs to be tested with a sample that has not been used for estimation. The holdout

sample validation perhaps constitutes one of the best tests to validate the discriminant approaches

function. Further, the model should also be able to predict the default much before the

occurrence of the incident. Since we are using a balanced panel data over the period

1998 to 2003, the Z-score model should be able to capture the dynamics of default 265

incidents.

Table VI shows that when the Model 3 tested on a holdout sample of 50 companies (25

defaulted and 25 solvent) for the period 2004, it can correctly classify 92 percent of the

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

defaulted firms and 96 percent of the solvent firms. However, both the type I and type

II error rates are much higher in case of Model 1 (20 percent and 12 percent

respectively) and Model 2 (16 percent each). Therefore, Model 3 has the best holdout

sample predictive accuracy.

Having enough confirmed the predictive power of our new Z-score model; we now

check its long run predictive ability. The question we try to answer is: how far into the

future the model predicts accurately? Accordingly, we examine the overall

effectiveness of the third model for a longer period of time prior to the occurrence of

actual default. This has important strategic meaning for the banks, because the

corporation may default on bank loan much before its bond instrument publicly rated

as D. Therefore, earlier the model can identify signs of stress, lesser are the costs

involved in taking the required steps.

To test the long run accuracy of our Z-score model, we have created holdout sample

of another 37 defaulted Indian corporations and collected their financial data. These

corporations got D ratings from CRISIL between the years 1996-2005. Next, we

examine the prediction power of our model back in time from 0, 1, 2, 3, 4, 5, 6 years

prior to default. As can be seen from the Table VII that the predictive ability of the

model to identify defaulting firm falls from 88 percent one year prior to default to 45

percent as one goes back six years prior to the occurrence of default. We also compare

Altman (1968) emerging market (1995) Z-score model

Year prior to default (%) (%) (%)

0 79 79 88

1 59 73 88

2 55 50 68

3 – – 57 Table VII.

4 – – 56 Relative comparison of

5 – – 45 classification and

6 – – 45 predictive accuracy of

three discriminant

Notes: Using a holdout sample of 37 Indian corporate bonds defaulted between the year 1996-2005. models: early warning

Also using 0 as the cutoff score signal (time dimension)

JRF the long run predictive power of our model with Model 1 and Model 2. Our model (i.e.

7,3 Model 3) clearly out performs the other two models even if one goes back two years

prior to default (with 68 percent accuracy). Moreover, the type I accuracy rate of Model

3 is pretty high (88 percent) on data from one financial statement prior to default on

outstanding bonds.

Finally, we examine 148 distressed manufacturing firms who reported bankrupted by

Board for Industrial and Financial Reconstruction (BIFR) of India in year 2004. As

Table VIII shows, our Z-score model is still robust and can also forecast corporate

failure up to five years prior to distress. The accuracy rate is very high until two years

prior to distress (87 percent). We also compare the predictive accuracy power of our

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

model with Model 1 and Model 2. It is evident from Table VIII that our model performs

much better than the other two previous models.

We have done Logistic regression analysis to investigate the relationship between

binary or ordinal response probability and explanatory variables. The method fits

linear logistic regression model for binary or ordinal response data by the method of

maximum likelihood. Like discriminant analysis, this technique weights the

independent variables and assigns a Z-score in a form of failure probability to each

company in a sample. Discriminant analysis and logit analysis have different

assumptions concerning the relationships between the independent variables. While

linear discriminant analysis is based on linear combination of independent variables,

logit analysis uses the logistic cumulative probability function in predicting default.

The logit equation we have estimated takes the following form:

PD ¼ FðZ Þ

¼ 1

1þe 2Z

¼ 1=1 þ e 2ðb0 þb1 X 1 þb2 X 2 þb3 X 3 þ...þbk XÞ

Altman (1968) market (1995) Z-score model

Year prior to failure (%) (%) (%)

0 94 95 97

1 95.3 95.3 96.3

Table VIII. 2 82.5 85 87

Relative comparison of 3 73 73 78

bankruptcy prediction 4 62 68 73.3

power of three 5 55 56 68.1

discriminant models: Notes: Using a holdout sample of 148 Indian Manufacturing Firms reported bankrupted by Board for

early warning signal Industrial and Financial Reconstruction (BIFR) of India in the year 2004. Also using 0 as the cut-off

(time dimension) score

We have applied Maximum Likelihood Estimation (MLE) procedure for estimation of Logistics and

the parameters. Z-score model

In the logit regression, our purpose is to evaluate the role of balance sheet variables

as well as the non-financial variables in predicting corporate bond default and to arrive approaches

at an estimate of probability of default for a firm using them.

Before we discuss our results, let us first look at the descriptive statistics of the

variables used in the logistic regressions. Table IX, gives us some descriptive statistics 267

about the sample of firms used in the logistic analysis. It is evident from the descriptive

statistics table that all the financial ratios for solvent group of firms on average look

relatively better than their defaulted counterparts. As far as the non-financial parameters

are concerned, the greater percentage of solvent firms possess ISO certificate than the

defaulted firm. Similarly, defaulted firms are on average younger than the defaulted

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

firms. The solvent firms also mostly belong to the top 50 business groups than the

defaulted ones. The difference is also statistically significant as evident from the

t-statistics reported in column 8[5]. Furthermore, standard deviations of financial ratios

are very high for defaulted companies in comparison to the solvent firms.

We define the dependent variable by coding an indicator binary variable with a 1

(default) or a 0 (non-default). The same balanced panel data set of 104 companies (52

defaulted firms and 52 solvent firms) over the period 1998 to 2003 has been used to run

the logit model. In a stepwise logistic regression method, we finally obtain three sets of

models. In Model 1 and 2, we test financial as well as non-financial parameters that

have been discussed earlier in the variable definition section. Model 3 only explains the

role of financial factors on the probability of default by a firm. The results are

summarized in columns 2, 3, and 4 of Table X.

Model 3 (Table X) shows that the financial ratios (SOLVR, CASHPROF_TA,

WK_TA, SALES_TA and MVE_BVL) are negatively significant (5 percent or

better) on default probability. The results of Model 3 are consistent with our

theoretical expectation, which is discussed in the data and variable section. In

Model 1 and Model 2, we test the explanatory power of both the financial as well

as non-financial parameters. The financial ratios have the same expected signs as

we have found in Model 1 and 2. As far as non-financial parameters are

concerned, the age parameter LN(AGE) is negatively significant, implying that

Mean Std. dev. DEF ¼ 0 Std. dev. DEF ¼ 1 Std. dev. for difference

CASHPROF_TA 0.036 0.105 0.1 0.07 20.03 0.09 19.1 * * *

WK_TA 0.06 0.29 0.19 0.17 20.07 0.33 12.17 * * *

SALES_TA 0.81 0.55 1.04 0.61 0.57 0.33 11.81 * * *

MVE_BVL 1.303 4.65 2.63 6.58 0.155 0.35 6.53 * * * Table IX.

ISOD 0.57 0.49 0.69 0.46 0.46 0.50 5.99 * * * Descriptive statistics for

Dtop50grp 0.37 0.48 0.46 0.50 0.29 0.45 4.53 * * * logit model: comparison

LN(AGE) 3.22 0.78 3.46 0.77 2.99 0.73 7.78 * * * between defaulted group

No. of observations 624 312 312 and solvent group

JRF

Variables Model 1 Model 2 Model 3

7,3

DEF Coefficients Coefficients Coefficients

SOLVR 2 1.23 * * * 2 1.47 * * * 21.78 * * *

CASHPROF_TA 2 12.91 * * * 2 13.49 * * * 2 11.74 * * *

WK_TA 2 10.35 * * * 2 9.02 * * * 24.54 * * *

268 SALES_TA 2 1.67 * * * 2 1.79 * * * 21.32 * * *

MVE_BVL 2 1.74 * * * 2 1.56 * * * 21.33 * * *

ISOD 2 1.27 * * * 2 1.88 * * * –

Dtop50grp – 2 0.805 * * * –

LN(AGE) 2 1.66 * * – –

IND1 2 1.114 Dropped –

IND2 Dropped Dropped –

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

IND4 1.27 2.69 * * * –

IND5 2.48 * * 3.51 * * * –

IND6 0.86 2.92 * * * –

IND7 0.46 2.08 –

IND8 Dropped Dropped –

IND9 1.49 1.57 –

IND10 2 3.52 * * * 2 1.89 –

IND11 Dropped Dropped –

Intercept 11.46 * * * 5.24 * * * 5.5 * * *

Number of Obs. 518 518 558

LR x2 statistics 492.93 (14) 469.47 (14) 437.69 (5)

Prob.. x2 0.00 0.00 0.00

Pseudo R 2 0.70 0.66 0.57

Table X. Notes: The dependent variable DEF is a default dummy; DEF=1, if the company’s long term bond is

Logit model: prediction of defaulted in any year between the year 1998 to 2003 and DEF =0, if there is no default. The model 1

default events with and model 2 use all financial and non-financial factors. Model 3 uses only financial parameters. * * *

different factor types denotes significant at 5 percent or better; * * denotes significant at 5-10 percent

younger firms are more risky than the older firms. It is more likely that matured

firms have established a reputation with credit institutes and private investors that

alleviates the asymmetric information problems because an extended period of

scrutiny would permit a better understanding of the economic viability of the firm.

In the case of a liquidity crunch, an older firm could rely on such a relationship to

obtain additional lines of credit or favorable grace periods and can avoid a

corporate default event. On the other hand, young firms have less time to solidify

a relationship with its creditors and private investors hence increasing the chance

of financial distress during a credit crunch.

Similarly, the likelihood of default is less if the firm belongs to the top 50 business

group. Likewise, the ISO dummy (ISOD) has negative influence on the probability of

default (DEF), indicating that the firms that maintain a quality management system have

less chance of default. The industry dummies are significantly different from zero

suggesting that we cannot reject the presence of industry effects on firm’s default status.

Now, let’s compare the diagnostic tests of these models. As reported in the lower panel

of Table X, Pseudo R 2 is highest (0.70) in Model 1 in comparison to Model 2 (0.66) and

Model 3 (0.57)[6]. The chi-square statistics is also highest in case of Model 1. We also

checked the predictive power of the logistic models by using ROC graphs in Figure 1 and

calculate the area under the ROC curve based on the model estimates by logit[7]. One can Logistics and

clearly see from Figure 1 that Model 1 has the highest within sample prediction power of Z-score model

97.2 percent in comparison to Model 2 and Model 3. Further, we have performed a

chi-squared test to summarize the predictive accuracy of these three models into a approaches

summary statistic. The chi-squared test yielded a significance probability of 0.001

suggesting that there is a significant difference in the areas under the three ROC curves.

Hence, it is evident from our results that inclusion of non-financial factors along with the 269

financial factors improves the default-forecasting ability of the model. The results of

Model 1 indicate a strong relationship between default and the financial and non-financial

variables. The Model 1 can be directly used for finding PDs in credit-risk models[8].

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

One may again argue that model performance may be driven by type I and type II error

rates. Consequently, Model 1 has been tested in the same holdout sample that we used in

MDA analysis. The 2003 and 2004 holdout sample consists of equal number of 25 cases of

solvent and 25 cases of defaulted firms. As evident from the results reported in Table XI,

model 1 clearly shows good capacity to discriminate between defaulted and solvent firms.

Figure 1.

Comparison of ROC

curves for three models

Predicted group

Defaulted 47 3 50 Classification power of

(94%) (6%) (100%) the logistic Model 1 for

Solvent 8 42 50 the holdout sample of the

(16%) (84%) (100%) years 2003 and 2004

JRF Conclusions

7,3 Using a sample of 104 listed corporations from CRISIL, we have developed a Z-score

model for Indian firms that can accurately predict bond default one year in advance.

The model not only has a high classification power within sample (91 percent), but also

exhibited a high predictive power in terms of its ability to detect bad firms in the

two-holdout samples (with 92 percent and 88 percent accuracy rates). Moreover, the

270 model also can predict corporate bankruptcy in two years prior to financial distress

with an accuracy rate of 97 percent and 96.3 percent respectively. The new Z-score

model of this paper outperforms the other two contesting models comprising of

Altman, 1968 and emerging market score 1995 set of ratios respectively. Using our

Z-score model, banks as well as investors can get early warning signals about the firm

and might reassess the magnitude of the default premium they require on low-grade

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

securities.

In the logit analysis, we link the firm’s performance with the macro economic

environment. The logit results show that PD is a decreasing function of cash profit

over total assets, working capital to assets, total sales relative to total assets, solidity,

solvency ratio, firm age, ISO certification and top 50 group affiliation. Further, industry

affiliation of a firm is also an important factor for explaining its default status and also

needs to be taken into account. From our empirical analysis we find that inclusion of

both financial and non-financial factors leads to more accurate default prediction than

the single use of accounting ratios.

Notes

1. CRISIL defines default as a credit event where the underlying corporate has missed

payments (a single day’s delay or a shortfall of even a single rupee) on a rated instrument in

terms of the promised repayment schedule. CRISIL’s rating does not factor in any post

default recovery.

2. A paired t-test on the mean asset difference between the two groups had shown statistically

insignificant results.

3. The probability of default (PD) per rating grade counts the average percentage of bond in

this rating grade in the course of one year.

4. It is empirically observed fact that the linear discriminant model has a higher holdout

sample predictive power compared to the quadratic discriminant model (Altman, 1993).

Moreover, the former is more amenable for interpretation compared to the latter.

5. A Wilcoxon rank-sum test showed that all the financial and non-financial parameters are

significantly better for solvent group (at 1 percent or better level) than the defaulted group.

6. Pseudo R 2 is a likelihood ratio index, which is analogous to the R 2 in a conventional

regression model. Here Pseudo R 2 ¼ 1 2 Lmax =L0 , where L0 is the initial value of likelihood

function and Lmax is the highest value.

7. Receiver Operating Characteristic Curve (ROC) quantifies the accuracy of diagnostic tests to

discriminate between defaulted firms and solvent firms using each value of the logit score as

a possible cutoff point. The analysis uses the ROC curves of the sensitivity (percentage of

true defaulted outcomes correctly specified) vs. 1-specificity (percentage of false defaulted

outcomes correctly specified) of the diagnostic test. This calculates the area under the ROC

curve based on the model estimated by logistic regression predictions. The greater the area

under the ROC curve, the better the predictive power of the model. Therefore, a steeper curve

from the diagonal line indicates a more powerful model.

8. From Table results of Model 1, one can estimate the probability of default (PD) by using the

1

Logistics and

following equation: PDi ¼ 1þexpð2z , zi ¼ a þ bX i , where a is the intercept and b represents

iÞ

the parameters that may explain default incidents. Z-score model

approaches

References

Altman, E.I. (1968), “Financial ratios, discriminant analysis and the prediction of corporate 271

bankruptcy”, Journal of Finance, September, pp. 189-209.

Altman, E. (1993), Corporate Financial Distress and Bankruptcy, 2nd ed., John Wiley and Sons,

New York, NY.

Altman, E.I. (2000), “Predicting financial distress of companies: revisiting the Z-score and ZETA

models”, working paper, Stern School of Business, New York University, New York, NY.

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

Altman, E.I., Haldemann, R.G. and Narayan, P. (1977), “ZETATM analysis: a new model to

identify bankruptcy risk of corporations”, Journal of Banking and Finance, Vol. 1,

pp. 29-54June.

Altman, E.I., Hartzell, J. and Peck, M. (1995), Emerging Markets Corporate Bonds: A Scoring

System, Salomon Brothers, New York, NY.

Amemiya, T. (1981), “Qualitative response models: a survey”, Journal of Economic Literature,

Vol. 19, pp. 1483-536.

Asquith, P., Gertner, R. and Scharfstein, D. (1994), “Anatomy of financial distress: an

examination of junk bond issuers”, Quarterly Journal of Economics, pp. 625-58.

Aziz, A., Emanuel, D. and Lawson, G. (1988), “Bankruptcy prediction: an investigation of cash

flow based models”, Journal of Management Studies, Vol. 25, pp. 35-51.

Bandyopadhyay, A. and Das, S.K. (2005), “The linkage between the firm’s financing decisions

and real market performance: a panel study of Indian corporate sector”, Journal of

Economics and Business, Vol. 57 No. 4, pp. 288-316.

Beaver, W.H. (1966), “Financial ratios as predictors of failure”, Journal of Accounting Research,

Vol. 4, pp. 71-111.

Beaver, W.H. (1968), “Market prices, financial ratios and the prediction of failure”, Journal of

Accounting Research, Vol. 6 No. 2, pp. 179-92.

Black, F. and Scholes, M. (1973), “The pricing of options and corporate liabilities”, Journal of

Political Economy, Vol. 81 No. 3, pp. 637-54.

Casey, C. and Bartczak, N. (1985), “Using operating cash flow to predict financial distress: some

extensions”, Journal of Accounting Research, Vol. 23 No. 1, pp. 384-401.

Denis, D.J. and Denis, D.K. (1994), “Causes of financial distress following leveraged

recapitalizations”, Journal of Financial Economics, Vol. 37 No. 2, pp. 129-57.

Dimitras, A.I., Zanakis, S.H. and Zopounidis, C. (1996), “A survey of business failures with an

emphasis on prediction methods and industrial applications”, European Journal of

Operational Research, Vol. 90, pp. 487-513.

Gangopadhyay, S., Lensink, R. and Molen, V. (2001), “Business groups, financing constraints and

investment: the case of India”, mimeo.

Gentry, J.A., Newbold, P. and Whitford, D.T. (1985), “Classifying bankrupt firms with funds flow

components”, Journal of Accounting Research, Vol. 23 No. 1, pp. 146-59.

Grunert, J., Norden, L. and Weber, M. (2005), “The role of non-financial factors in internal credit

ratings”, Journal of Banking and Finance, Vol. 29, pp. 509-31.

JRF Johnsen, T. and Melicher, R.W. (1994), “Predicting corporate bankruptcy and financial distress:

information value added by multinational logit models”, Journal of Economics and

7,3 Business, Vol. 46, pp. 269-86.

KMV (1993), “Credit monitor overview”, mimeo, KMV Corporation, San Francisco, CA.

Kranhnen, J.P. and Weber, M. (2001), “Generally accepted rating principles: a primer”, Journal of

Banking and Finance, Vol. 25, pp. 3-23.

272 Lang, L.H.P. and Stulz, R.M. (1992), “Contagion and competitive intra-industry effects of

bankruptcy announcements”, Journal of Financial Economics, Vol. 32 No. 1, pp. 45-60.

Lennox, C. (1999), “Identifying failing companies: a re-evaluation of the logit, probit, and DA

approaches”, Journal of Economics and Business, Vol. 51, pp. 347-64.

Maddala, G.S. (1983), Limited Dependent and Qualitative Variables in Econometrics, Cambridge

University Press, Cambridge.

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

Mar Molinero, C. and Ezzamel, M. (1991), “Multidimensional scaling applied to corporate failure”,

OMEGA, Vol. 19 No. 4, pp. 259-74.

Martin, D. (1977), “Early warning of bank failure: a logit regression approach”, Journal of

Banking and Finance, Vol. 1, pp. 249-76.

Mossman, C.E., Bell, G.G., Swartz, L.M. and Turtle, H. (1998), “An empirical comparison of

bankruptcy models”, The Financial Review, Vol. 33, pp. 35-54.

Ohlson, J.A. (1980), “Financial ratios and the probabilistic prediction of bankruptcy”, Journal of

Accounting Research, Vol. 18 No. 1, pp. 109-31.

Opler, T.C. and Titman, S. (1994), “Financial distress and corporate performance”, Journal of

Finance, Vol. 49, pp. 1015-40.

Scott, J. (1981), “The probability of bankruptcy: a comparison of empirical predictions and

theoretical models”, Journal of Banking and Finance, Vol. 5, pp. 317-44.

Tam, K. (1991), “Neural network models and prediction of bank bankruptcy”, OMEGA, Vol. 19

No. 5, pp. 429-45.

Westgaard, S. and Wijst, N.V. (2001), “Default probabilities in a corporate bank portfolio: a

logistic model approach”, European Journal of Operational Research, Vol. 135, pp. 338-49.

Wilcox, J.W. (1971), “A simple theory of financial ratios as predictors of failure”, Journal of

Accounting Research, Vol. 9 No. 2, pp. 389-95.

Wilson, R.L. and Sharda, R. (1994/1995), “Bankruptcy prediction using neural networks”,

Decision Support Systems, Vol. 11, pp. 545-57.

Zavgren, C.V. (1985), “Assessing the vulnerability to failure of American industrial firms: a

logistic analysis”, Journal of Business Finance and Accounting, Vol. 12 No. 1, pp. 19-45.

Zopounidis, C. and Dimitras, A.I. (1998), Multicriteria Decision Aid Methods for the Prediction of

Business Failure, Kluwer Academic Publishers, Dordrecht.

Further reading

Eisenbeis, R.A. (1977), “Pitfalls in the application of discriminant analysis in business, finance

and economics”, Journal of Finance, Vol. 42, pp. 875-900.

Corresponding author

Arindam Bandyopadhyay can be contacted at: arindam@nibmindia.org

Or visit our web site for further details: www.emeraldinsight.com/reprints

This article has been cited by:

1. A.M. Hafizi, Shahida Shahimi, B.A.M. Hafizuddin-Syah, M. Badrul Hakimi Daud. 2015. Road to default:

Tracoma holding Berhad Bai Bithaman Ajil Debt securities (BaIDS). Emerald Emerging Markets Case

Studies 5:5, 1-11. [Abstract] [Full Text] [PDF]

2. Mohammad M. Mousavi, Jamal Ouenniche, Bing Xu. 2015. Performance evaluation of bankruptcy

prediction models: An orientation-free super-efficiency DEA-based framework. International Review of

Financial Analysis . [CrossRef]

3. Khushbu Agrawal, Yogesh Maheshwari. 2014. Default risk modelling using macroeconomic variables.

Journal of Indian Business Research 6:4, 270-285. [Abstract] [Full Text] [PDF]

4. Tomasz Korol. 2013. Early warning models against bankruptcy risk for Central European and Latin

American enterprises. Economic Modelling 31, 22-30. [CrossRef]

Downloaded by Sung Kyun Kwan University At 21:47 04 November 2015 (PT)

5. Michael Alexeev, Jounghyeon Kim. 2012. Bankruptcy and institutions. Economics Letters 117, 676-678.

[CrossRef]

6. Nor Balkish Zakaria, Mohamad Azwan Md Isa, Rabiatul Alawiyah Zainal Abidin. 2012. The Construct

of Sukuk, Rating and Default Risk. Procedia - Social and Behavioral Sciences 65, 662-667. [CrossRef]

7. Xing Zhong, Yi Peng, Gang KouA dynamic self-adoptive genetic algorithm for personal credit risk

assessment 711-716. [CrossRef]

8. Arindam Bandyopadhyay. 2007. Mapping corporate drift towards default. The Journal of Risk Finance 8:1,

46-55. [Abstract] [Full Text] [PDF]

- IntroductionUploaded byDubey Sonam
- Future Predictions Article ChecklistUploaded byLuis Antonio Triana Muñoz
- Literature ReviewUploaded byneerjam
- Banking Risk ManagUploaded bySimona Vornicelu
- Credit ManagementUploaded bypreet20
- Comparing Human and Computational ModelsUploaded byGary
- Working Capital ManagementUploaded byIman Nurakhmad Fajar
- Fbr Preview Packet w o QuestionsUploaded byAshima Mishra
- Malomo et alUploaded bybvomidiji
- Social Network Investor Sentiments for Predicting Stock Price TrendsUploaded bySandeep Ranjan
- Abdelkader-Predicting Energy Demand Peak Using M5 Model TreesUploaded byAnonymous PsEz5kGVae
- DRTA Prediction LogUploaded byAnonymous Env0Wed
- Defi Business and Products IntroductionUploaded byou fan
- tmpD7CF.tmpUploaded byFrontiers
- deeoika 1.docxUploaded byTanmoy Chakraborty
- New Microsoft Word DocumentUploaded byPolash Wub
- Data Mining Improves Pipeline Risk AssessmentUploaded bySharon Freeman
- Nerolac_credit Application FormattedUploaded byrahulrelan
- Ten Iterative Steps in Development and Evaluation of Environmental ModelsUploaded byDario Daddà Turres
- JaponUploaded byMaría Sendino
- 1-s2.0-S0048969714016027-main.pdfUploaded byChidiebere Kalu
- nclevels-wp.pdfUploaded byNoushad P Hamsa
- Measuring RiskUploaded byRitika Jain
- assessment rubric mouldy bread investigationUploaded byapi-238560466
- STAT4607Uploaded byAlex Rush
- Soft Tissues AdaptabilityUploaded byMargarita Lopez Martinez
- 06_Forecasting With Limited Data_Combining ARIMA and Diffusion ModelsUploaded bydokter Hewan Adhona
- Tools for ConservationUploaded byDave Plakorus
- tmpCB04.tmpUploaded byFrontiers
- 4. English - IJEL - The Effect of English Fluency - Rita Sutjiati JohanUploaded byTJPRC Publications

- Are Ratings Consistent With Default Probabilities Emerging BanksUploaded byraqthesolid
- Secured Lending and Default Risk Equilibrium Analysis, Policy Implications and Empirical ResultsUploaded byraqthesolid
- Fahlenbrach and Stulz (2011) “Bank CEO incentives and the credit crisis.pdfUploaded byAnonymous WFjMFHQ
- SensitivitiesUploaded byswanyamawi16
- A Survey of Systemic Risk AnalyticsUploaded byraqthesolid
- Sovereign Default, Domestic Banks, And Financial InstitutionsUploaded byraqthesolid
- A Risk Assessment Model for BanksUploaded byraqthesolid
- Estimating ProbabilitiEs of DEfault for German ServicesUploaded byraqthesolid
- The Empirical Relationship Between Average Asset Correlation, Firm Probability of Default and Asset SizeUploaded byraqthesolid
- bank_insolvency__bad_luck__bad_policy_or_104582Uploaded byMyra Ellen
- distancedefault.pdfUploaded byraqthesolid
- distancedefault.pdfUploaded byraqthesolid
- distancedefault.pdfUploaded byraqthesolid
- Writing Research Paper(陆林)Uploaded byraqthesolid
- An Empirical Comparison of Default Risk Forecasts From Alternative PhilosophiesUploaded byraqthesolid
- How to design a coverUploaded byraqthesolid
- CritiqueUploaded byraqthesolid
- Research IdeasUploaded byraqthesolid
- 1-s2.0-S0304405X11000420-mainUploaded byraqthesolid
- VariablesUploaded byraqthesolid
- Name of JournalsUploaded byraqthesolid
- SKKU HandbookUploaded byHa
- JCR Impact Factors 2013Uploaded byaksasingh
- Stock Exchange ResultsssUploaded byraqthesolid
- Country Related DataUploaded byraqthesolid
- Financial ManagementUploaded byraqthesolid
- Chalte Ho To Cheen Ko Chaliay - Ibne InshaUploaded byusman_reus
- Signalling ModelUploaded byraqthesolid
- SAS9.3_Win_Wrkstn-201306Uploaded bysimon hopkins

- SwedishUploaded bycgabriela12
- Guidelines for Writing a Project ProposalUploaded byJehadAbuSalim
- (Public Management and Change Series) Rosemary O'Leary, David M. Van Slyke, Soonhee Kim-The Future of Public Administration Around the World_ the Minnowbrook Perspective-Georgetown University Press (2Uploaded byUlfa M. Rusli
- paul bryant resume 1 21 14Uploaded byapi-268708599
- VoIPBlasterUploaded bytrunks1703
- Structure of a Critical ReviewUploaded byPogonici Ștefan
- Shuaa Capital UAE Vision BookUploaded byBRR_DAG
- harison capital programUploaded byapi-263749875
- YHT VS CAUploaded byCandypop
- Explain and assess Nietzsche’s claim that ‘the world is will to power and nothing besides’.docxUploaded byPatricia Villya
- Barcoo Independent 221010Uploaded bybarcooindependent
- fur in fashionUploaded byВалерияЧербу
- Thomas PaineUploaded bypapiyaparazzi
- 6. COCOFED vs. Republic.docxUploaded byJay Kent Roiles
- Zimbabwe ReportUploaded by1MilnZimVoices
- Hack Yourself First FinalUploaded bykopi
- Poles Living in Ireland and their Quality of LifeUploaded byMichał Nowosielski
- Staad Sample Output FileUploaded byAllan Bautista
- StartupDocsUploaded byzika007
- Sdmp PresentationUploaded bydied_dead
- cfo experience presentation vcurrentUploaded byapi-276673700
- zachary - civil and criminal law of ancient romeUploaded byapi-313966244
- Umali v. Estanislao (1992, 209 SCRA 446)Uploaded byKT
- PEACE_LESSONS_FINAL.pdfUploaded byProanio Segovia Catherine Elisa
- AFGP-2002-000103-TransUploaded bydibs59
- IndexUploaded byCharlyn Acojedo Malimata
- NTSE the Making of a Global WorldUploaded byvidushi1121
- B2B IntelUploaded byani872
- [Martial Arts] (eBook PDF) - Principles of WarUploaded byPruthvish Patel
- The Legacy of Shaw and McKayUploaded byNap Gonzales