La danghuyngan sum

Object and scope of the thesis - Research object: Object of this thesis is default risk, model of default warning in Vietnamese joint stock commercial banks3. Some applied models are L

INTRODUCTION

1 Rationale

Banking system plays an important role in the economy, it is considered as

“vascular system” of the whole economy As an inseparable factor in the

activities of commercial banks in the market, risk is always contained in any

banking activity It can lead to bigger damages to the economy than other

business and the cost of reparation is huge The early warning of default helps to

avoid the bank default, to minimize the loss of depositors, the deposit insurance

companies and the economy The default of an inefficient bank can create a chain

failure in the banking system and badly affect the sustainable development of the

system Therefore, it is important for to find out soon which banks are in

financial trouble and at high default risk in order to prevent crisis in the banking

system, and to maintain the stability of the financial market and macroeconomic

The remarkable development of the system of joint stock commercial banks

in the period of 1991-1996 and the following period of 2006-2010, has contributed

significantly to the country's economic development Apart from various

achievements, joint stock commercial banks have however exposed their

shortcomings and weaknesses Many of them have become insolvent by the end of

2011 The Scheme of Credit Institutions Restructuring for the period of 2011-2015,

thus, was issued The Resolution of the 3rd Plenum (XI Session) affirmed that one

of three pilars of economic restructuring is the financial system restructuring with

the banking system at the center To accomplish a successful result, it is important to

identify and classify inefficient banks at the risk of default

So far in the world, there have existed many theories and warning models of

default crisis such as: Univariate Analysis, Multiple Discriminant Analysis (MDA),

Logit Analysis (LA), Probit Analysis (PA), etc The recent models, such as Artificial

Neural Networks (ANNs), Decision Tree (DT), Trait Recognition (TR), etc have

been applied in default warning and have promised good results The studies also

show that each method and model has its own advantages and disadvantages; each

model applied in different countries or different regions provides different variants

It depends on the economic conditions of each country and each region Many

models have been created to explain the causes as well as to forecast and to prevent

debt crisis However, the unpredictable default of banks and financial institutions at

increasing scale and impact shows that default warning models should be paid

attention to and improved Significant socio-economic changes, the unpredictability

of natural and socio-economic events make traditional and current methods no

longer appropriate in many cases From the above reasons, the author chose the thesis titled "Developing a default warning model for commercial joint stock banks

in Vietnam" to contribute to solving the issues presented by theories and reality

2 Purpose

The purposes of the thesis are as follows:

- Develop and select the system of indicators applied in the evaluation of the probability of default of joint stock commercial banks

- Develop the empirical model of risk warning in Vietnamese joint stock commercial banks

- Propose some solutions to limit the default risk of Vietnamese joint stock commercial banks

Research questions:

- In Vietnam's context, what factors could characterize bank default; what factors, indicators may affect the default risk of Joint stock commercial banks and how?

- Each bank has specific characteristics making their own probability of default How to point out the differences?

- What method and model of default warning should be applied to Vietnamese joint stock commercial banks?

- Implications for policy drawn from the model?

3 Object and scope of the thesis

- Research object: Object of this thesis is default risk, model of default

warning in Vietnamese joint stock commercial banks

- Research scope: The research has been carried out on Vietnamese

joint stock commercial banks includes 35 Joint stock commercial banks, including joint stock banks in which the State holds controlling shares such

as BIDV, MHB, Vietinbank and VCB The study has been done in the period

from 2010 to 2015

4 Research method

In order to be in line with the content, requirements and research objects, the thesis applies the quantitative analysis and qualitative analysis Some applied models are Logit Analysis with array data, neural network model and decision tree model to create a risk warning model for Vietnamese joint stock commercial banks

The data in this thesis is collected from reports of State Bank of Vietnam,

the audited financial statements of Joint stock commercial banks in the period of

2010-2015

5 The scientific and practical significance of the thesis

- The thesis builds the theoretical basis of the default warning model for

joint stock commercial banks

- The thesis builds and selects the system of indicators of default warning

to be applied in banking system Identification of factors and characters affecting

the default risk in Joint stock commercial banks

- Quantifying the specific characteristics of each bank which affects the

probability of default

- Develop a model of default warning for joint stock commercial banks

- Propose some solutions to reduce the default risk of joint stock

commercial banks based on the analysis of the thesis

6 Composition of thesis

Apart from the introduction, conclusions, appendices, tables and lists of

references, the content of the thesis is divided into 5 chapters as follows:

Chapter 1: Overview of bank default

Chapter 2: Theoretical background on default in commercial banks

Chapter 3: Situation of operations, default risk of Vietnamese joint stock

commercial banks in the period of 2009-2015

Chapter 4: Creation of a default warning model for commercial banks in

Vietnam

Chapter 5: Conclusions and policy recommendations

CHAPTER 1: OVERVIEW OF BANK DEFAULT

1.1 Definition of default and default in commercial banks

The definition of default, default in commercial banks, and the

consequence of bank default

1.2 Overview of international studies on default and bank default

1.2.1 Overview of representative models and studies on default

Most of the current default warning studies in the world focus on two

branches: Parametric models and methods of analysis: Univariate Analysis,

Multivariate Analysis, Logit Analysis, Probit Analysis, Survival Analysis, etc;

Non-parametric models: Artificial Neural Networks, Decision Trees, character analysis, genetic algorithms, etc

Univariate Analysis: The main content of the univariate analysis in default warning studies is: the examination of individual factors and comparison the factors between two groups of insolvent companies and non-insolvent ones In case that the financial factors show signs of difference between these two groups, they are used as predictors The studies applying univariate analysis are: studies

by FitzPatrick (1932), Smith and Winakor (1935), Merwin (1942), etc A widely-applied study was the one by Beaver published in 1966 The advantages of the univariate analysis method are: simplicity, quick and convenient application and high rate of prediction However, this method exposes three shortcomings

To avoid the disadvantages of the univariate analysis, many researchers applied the Multiple Discriminant Analysis (MDA) with Edward Atlman (1968)

as the representative author Based on the data of bankrupt enterprises in the United States, he identified the discriminant function which was widely used later Altman's MDA in 1968 became, for many years, an influential model to the studies of default warning, especially the studies prior to 1980 such as the work

of Deakin (1972), Blum (1974), Altman and Edward, Haldeman, Narayanan (1977), Norton and Smith (1979), Karel and Prakash (1987), etc Howver when the time and location of study changed, the observations in Altman’s sample didn’t represent the market anymore The estimated values were no longer appropriate Currently, many authors have developed their own discriminant function for each country and each sector

The Logit model and the Probit model appeared in the late 1970s and until the late 1980s it became more common than the MDA method in the study of default warning The Logit and Probit model focus on the probability of default/default of enterprises Logit model and Probit model can be used to evaluate the level of interpretation of independent variables Ohson (1980) used the Logit model to replace the MDA model to predict default in enterprises The

LA model was also used by Platt (1991), Smith and Lawrence (1995), Koundinya (2004), Prasad and associates (2005), etc

Odom and Sharda (1990) were the first ones using neural network (ANN)

in their default warning studies The other authors are Hawley, Johnson and Raina (1990); Boritz and Kennedy (1995); Alam and associates (2000); Celik and Karatepe (2007)

West (1985) used Logit model in combination with factor analysis to

measure and to describe the financial and operational characteristics of banks

The data was taken out from income reports, as well as reports of 1900

commercial banks in various US states The important factors identified by the

Logit model in this study are similar to the factors used in the CAMELS model

The study also showed that the combination of factor analysis and Logit Analysis

was useful for evaluate the bank performance

Nowadays witness the recent trend of application of smart technics and the

computer technologies in default warning such as neural networks, decision trees,

character analysis, etc

The most international outstanding studies on default warning are

summarized in table 1.4

Table 1.4: Summary of international studies on default, default warning

Author/ Year of

publication Approach/ Factors Main result

Beaver (1966) UDA/30 Data collected from

79 insolvent enterprises and 79 non-insolvent ones in 38 sectors

Identification of 5 factors, accuracy rate from 50% to 92%

Atlman (1968) MDA/22 Data collected from

33 enterprises for each group

Identification of 5 factors, the accuracy rate of 95% per each sample enterprise

Martin (1977) MDA; LA/25 Data collected

from American banks in the period of 1970-1976

Creation of 6 models, identification of 4 factors, the best efficiency rate at 92.3%

Hanweck (1977) PA/6 Data collected from 177

non-insolvent enterprises and

32 insolvent ones

Out of 6 factors, 2 factors are statistically significant The accuracy rate of 83.8%, the test pattern of 91.1%

Ohson (1980) LA/9 Data collected from 1025

insolvent enterprises and 2000 non-insolvent ones

The accuracy rate of 96.3%

Tam and Kiang

(1992)

ANN/19, MDA/19, LA/19

Data collected from 118 banks

NN network with the accuracy rate of 96.2% for training

Author/ Year of publication Approach/ Factors Main result

divided in 2 groups sample and at 85.2% for test

sample

Odom and Sharda (1993)

ANN/5; MDA/5 Data collected from 38 insolvent enterprises and 36 non-insolvent ones

Rate of 100% for training sample and 77% for test sample

associates (1996)

TR and LA Rate of 98.6% for original

sample and of 95.6% for test sample

Vander Vennet (2006)

TR model and Logit model in major banks of Russia

Accuracy rate of 91.6% for original data and and of 85.1% for test data

Ravi and Pramodh (2008)

ANN/9;12 Data from Turk banks and Spanish banks

Rate of 96.6% for Turk banks and 100% for Spanish banks

In which: UDA- Univariate Discriminant Analysis; MDA- Multiple Discriminant Analysis; ANN- Artificial Neural Network; TR- Trait Recognition; DT- Decision Tree; LA- Logit Analysis; PA- Probit Analysis

Source: Synthesis from references

1.2.2 Overview of criteria to determine default or high default risk in existing studies

1.2.3 Factors and variables in studies on default

1.3 The studies on default warning and bank default in Vietnam

The studies on default, bank default in Vietnam are summarized in Table 1.8

Table 1.8: Various studies on default, bank default in Vietnam Author/ Year of

publication

Approach/ Factors Main content/Main results

Nguyen Trong Hoa (2009)

MDA/37; LA/37 Data collected from 268 enterprises in 2007

Estimating the discriminant function, the Logit function to calculate the probability of default and ranking the observations of the five selected samples

Nguyen Quang

Dong (2009)

MDA/40 Data collected from 37 banks in 2008

Ranking the banks’ credit

Identifying two factors, the accuracy rate of 90.6% for the original data and 84.4% for the test data

Phan Hong Mai

(2012)

The model of The Vickers company

Measure the risk of default of construction companies

Identifying the cause of the increased risk of default is the poor asset management

Nguyen Viet Hung

and Ha Quynh Hoa

(2012)

Models of currency crisis forecast

Currency crisis forcast

Identifying 5 indicators reflecting the probability of economic instability

Nguyen Thi Luong

(2014)

Merton-KMV model Data collected from 380 listed companies in the period of 2011-2013

Measuring the default risk of enterprises Providing evidences for the reasonable measurement capacity of the method

Nguyen Phi Lan

(2015)

Application of structure model

Evaluating the risk of system failure

of financial institutions in Vietnam, estimating the credit loss and the risk

of failure of the banking system

Nguyen Thi Hong

Vinh (2015)

Array data model, GMM (Gaussian mixture model)

Determining micro and macro factors affecting bad debts in banks

Source: Synthesis from references Research gap: Default in joint stock commercial banks has not been fully

evaluated; the default in banks has not been monitored since the study was

carried out in only a year The existing studies identify the factors leading to the

default risk in each case, but whether are those factors leading to the default risk

in Joint stock commercial banks in a certain period? In addition, each bank owns

their individual characteristics which affect the probability of default At present,

there is no study to identify the measurement criteria The macroeconomic factors

affecting the default risk in Vietnamese banks have not been verified yet From

the research gap, the author chose the following topic: "Developing a default warning model for commercial joint stock banks in Vietnam"

Chapter 1 conclusion

In chapter 1, the author presents the concept of default, bank default and the consequences of bank default The author summarizes the studies of default, bank default in the world as well as in Vietnam Specifically, the author provides a full overview of the models and studies on default, from univariate discriminant analysis to non-parametric smart techniques The author presents the criteria for default or high default risk in existing studies and systematically summarizes the factors applied in those studies By analyzing the main research methods in representative studies, the advantages and disadvantages of each method, there is still no model that is superior to others Each one has its own pros and cons The number of predictors of default in those studies is diverse The more or less factors in a model do not affect the prediction rate The summary helps the author

to find the research gap and to fill that gap, the author has set the object of the thesis and carried out the study in the following chapters

CHAPTER 2: THEORETICAL BACKGROUND ON DEFAULT IN

COMMERCIAL BANKS 2.1 Criteria to determine the default risk

In banking activities, credit risk is the biggest fear for managers The quality of credit reflects in the ratio of bad debt Being a permanent problem in the operation of every bank, bad debt causes some negative effects as follows:

Bad debt reduces the bank's profitability, liquidity, credibility and lead to default The increase of bad debt in the banking system is the most urgent problem in the period of 2011-2015 Banks with a non-performing loan (NPL) ratio of 3% or higher are placed under strict supervision by the State Bank Many studies in the world have shown the negative impact of high ratio of non-performing loan on many aspects of banking operations; many studies have demonstrated the relation between high ratio of non-performing loan and bank default In addition, the commercial bank is the monetary organization whose biggest goal is profit and profit is the important indicator to evaluate the success or failure of the management and operation of the bank Profits are also needed to make up for lost loans and to set up the full provision

Therefore based on the performance of banks, more grounds will be added into

categorizing bank risks In particular, the author argues that banks of poor

credit (represented by high ratio of non-performing loans) are at high risk of

asset loss In case of medium or poor performance, banks will not be able to

make up for the loss; the high level of vulnerability consequently lead to

default

To evaluate the performance of banks, the thesis applies the DEA method

to estimate the technical efficiency of the banks, and then assess the operational

efficiency (through profitability) The banks are classified in 3 groups: (group A -

good performance, B - average performance, group C – weak performance)

On the basis of the above arguments, the author determines the default risk

as follows: the variable of default risk (Y) is assigned a value of 1 (high default

risk) if the bank has a bad debt ratio of 3% or more then it belongs to group C

when using DEA for grouping The Y is assigned a value of 0 (low default risk)

in other cases

The criteria selection to classify the defaults in this thesis is different from

other studies because of the following reasons: Many default studies in the world

use information about defaulted banks while in Vietnam, no default case as in

international practice is recorded Few studies applied the coefficient of safety as

the criteria of categorization while in Vietnam, the State Bank made a mandatory

requirement on capital adequacy ratio so all commercial banks met this criteria

2.2 Factors affecting the default risk in commercial banks

2.2.1 The macro factors affecting banking activities

• Economic development

• Legal, economic, financial and monetary policies of the State

• Competitive level

2.2.2 Micro factors affecting the default risk in JCBs

The factors in Atlman’s model (1968): 5 factors

Choudhy (2007), Pavlos Almanidis and Robin C Sickles (2012) as well as

other authors has pointed out the financial ratios in CAMELS model are the

important criteria to evaluate the performance of financial institutions,

particularly the banks; these criteria is also important to forecast the bank default

The CAMEL rating system by National Credit Union Administration (NCUA)

was created and has been applied from October 1987 as a tool to supervise the

credit institutions The factors in CAMELS (Capital adequacy, Assets,

Management Capability, Earnings, Liquidity - asset liability management, Sensitivity to market risk, especially interest rate risk)

a) Capital adequacy: 5 factors

b) Asset quality: Non-performance loan ratio/total outstanding loan; Bad

Debts / (Equity and Provisions for bad debts); Provision for bad debts / Bad debts, provision for bad debts /outstanding loans In addition, the following indicators may be considered: loan rate/earning asset; interbank deposit and

loan/earning asset

c) Management: 3 factors

d) Earnings: 13 factors

e) Liquidity: 6 factors

By analyzing the criteria in the CAMEL system, the author summarizes and expects the sign of criteria affecting the default risk in Table 2.1

2.3 Theoretical backgrounds for models applied in studies on default warning 2.3.1 Logit Analysis, Logit Analysis with array data

Upon the advantages of array data, Logit Analysis and the purpose of the thesis (The default risk in Vietnamese Joint stock commercial banks in the period

of 2010-2015), the author chooses the Logit Analysis with array data The study also experiments the Artificial Neural Network, Decision Tree to classify, forecast the default risk in Joint stock commercial banks in Vietnam

2.3.2 Neural network 2.3.3 Decision tree 2.4 Data envelopment analysis (DEA) to evaluate the efficiency in Joint stock commercial banks’ activities

2.5 The frame of the research thesis

Conclusion of Chapter 2

In chapter 2, the author presents the criteria to determine the default risk, clarify the theoretical basis for the models of default risk warning in joint stock commercial banks, analyze macro and micro factors affecting the risk of bank default Based on theoretical basis, the advantages and disadvantages of some risk warning models matching the object of thesis as well as the existing data, the thesis chose to apply the Logit Analysis with array data, Neural Network, and Decision Tree to build risk warning model for Vietnamese joint stock commercial banks

CHAPTER 3: CURRENT OPERATION, DEFAULT RISK OF JOINT

STOCK COMMERCIAL BANKS IN VIETNAM DURING 2009 - 2015

In Chapter 3, the thesis analyzed the macroeconomic context and monetary

policies during the period of 2009 – 2015 Clear analysis of current operation

condition in the joint stock commercial bank system of Vietnam during 2009 –

2015 is also provided for the following aspects: structure, scale, capital adequacy

level, profitability, management efficiency, and asset quality

3.1 Macroeconomic context during 2009 - 2015

GDP growth remained quite stable with an average of 5.7% However, this

is a low number in correlation with the economy potential From 2010 to 2012,

GDP growth gradually decreased reflecting the turbulence in macroeconomic

condition From 2012 to 2015, the economy started to recover resulting in the

improving GDP growth year after year

Inflation rate sharply increased with its peak in 2011, and then gradually

decreased after that thanks to the intervention and persistent inflation control of

the Government The rate reached its lowest point in 2015

5.4

6.42 6.24

5.25 5.42

5.98 6.68

0

1

2

3

4

5

6

7

8

2009 2010 2011 2012 2013 2014 2015

GDP

GD P

Graphic 2.2: GDP growth 2009-2015 (%)

Source: Statistic bureau Import and export: Export has become one of the most important factors

pushing for economic breakthrough during the period Both import and export

growth showed strong increasing trend with approximately 25% increase

annually Today, the volume for export and import has accounted for 80% GDP

of the whole economy, reflecting its importance toward general economic

growth In 5 years from 2010 to 2014, export volume has been constantly higher

than import volume, leading to excess of export in the economy since 2012 For

the year 2014, the excess volume of export has reached approximately 2 billion

dollars

Budget income and spending: During 2009 – 2015, total income and spending of the budget has been increasing over the year and showed constant status of overspending The period 2010 – 2015 witnessed the decrease of credit growth within banking system The credit growth constantly decreased during

2010 – 2013 reflected the difficult situation for businesses

3.2 Several monetary policies during 2009-2015

• Interest policy

• Exchange rate policy

• Open market operation

• Deposit rate

• Lending rate

3.3 Operation of banking industry 3.3.1 Structure, scale, and operational scope of banks

• Ownership structure, distribution of operations in commercial bank system

• Scale, operational scope

3.3.2 Capital adequacy level and total asset scale of joint stock commercial banks

a) Equity capital and capital adequacy level of joint stock commercial banks:

Period 2005-2011

Period 2011 to 2015

Graph 3.4: Capital adequacy ratio groups

Source: Author’s calculation

Capital adequacy level of banks is guaranteed according to regulation from

the State Bank However, the ratio is still quite low compared to other countries

in the region and should be increased to counter latent risks in the coming time

Capital adequacy ratio of banks has been constantly decreasing since 2012

leading to instability within the commercial bank system

b) Total asset scale of joint stock commercial banks

From 2008 to 2011, total assets of banks have shown increasing trend with

real breakthrough within the joint stock commercial bank section Joint stock

commercial banks have actively been actively opening their branch networks

resulting in dramatic growth in raising capital and effectively exploit the capital

of citizens In 2012, the total asset scale of joint stock commercial bank section

showed the declining trend Till September 30, 2013, total assets from banking

sector reached 5,637 thousands in billion dongs Till end of July, 2015, the total

asset of the whole financial credit system reached over 6.6 million in billion

dongs, increasing 150,097 billion dongs compared to end of 2014 Despite the

dramatic growth, banks in Vietnam are still of small asset scale comparing to

other countries in the region Total assets need to be increased to sustain the

growing need of a developing economy

3.3.3 Profitability, asset management efficiency

ROE ratio of banks

Income distribution of banks

Graph 3.5: Indicators in profitability category

Source: Author’s calculation

3.3.4 Credit and deposit growth, liquidity

• The growth of credit and deposit

• Liquidity 3.3.5 Asset quality, deficit level

Graph 3.9 showed the bad debt ratio of the joint stock commercial banks in the research samples

Graph 3.9: Bad debt ratio of joint stock commercial banks in Vietnam during 2010 - 2015

Source: Author’s calculation

Bad debt ratio in Vietnam is currently at a high level comparing to other countries in the region such as Thailand (2.7%), Indonesia (2.4%) By end of 2015, bad debt ratio has been constraint at 2.9%; however, there remains a lot of

concerning issues

Several weak points of the commercial bank system in Vietnam

• Extremely high risk in banking operations

• Administrative ability and competitiveness are low

3.4 Risk of default of several typical joint stock commercial banks during the period of 2009 - 2015

Weak banks with high probability of default include SCB Bank, Tin Nghia Bank, Ficombank, Habubank, Dai Tin bank, Ocean Bank, Westernbank, Dong A Bank, and GPBank The author analyzed these weak banks to present

highlight characteristics of these banks

Conclusion for Chapter 3

In Chapter 3, the thesis presented three key points as followed:

1) Analysis of macroeconomic context as well as major monetary policies during 2009 – 2015 The global and regional economic conditions with great instability ranging from global financial crisis to sovereign debt crisis have

negatively affected Vietnam’s economy During the period of 2010 – 2012,

macroeconomic environment showed many turbulences leading to the economic

growth decline to its lowest point within 10 years Moving to the 2013 – 2015

period, the economy started to recover, however, sustainable growth has yet to be

achieved The monetary policies during this period had many major movements

strongly affecting the banking operations

2) Analysis of operations within banking system through following

indicators: structure, scale and scope of operation, capital adequacy ratio, total

asset scale, credit and deposit growth, financial efficiency and liquidity risk The

author focused on analyzing several indicators in CAMEL model, bad debt

indicator, causes and negative effects of bad debts At the same time, the research

pointed out the several weakness of commercial bank system in Vietnam

including: extremely high risk in banking operation of commercial bank system,

low capability in administration and competitiveness

3) Analysis of default risk for several typical joint stock commercial banks

and pointing out highlight characteristics of weak banks These weak banks are

those with low capability in risk management, low asset quality, high bad debt and

overdue debt rate, low profitability

CHAPTER 4: DEVELOPING DEFAULT WARNING MODEL FOR

JOINT STOCK COMMERCIAL BANKS IN VIETNAM

4.1 Research design

4.1.1 Data

The banks included in the research consist of 35 joint stock commercial

banks In detail, the number of banks over the year is indicated in table 4.1

Table 4.1: Number of banks in the research

Financial indicators used to predict default probability of banks are

calculated from index and criteria within the audited financial statements by year

end of joint stock commercial banks in Vietnam from 2010 to 2014 with the total

of 163 observations In 2015, there were 25 banks used to test out-of-sample

forecasting of the model

4.1.2 Categorization of default probability level for the banks

a) Calculation of business performance of joint stock commercial banks:

Based on the assumption that all banks try to maximize their profits, input indicators and outcome results are selected to run DEA model evaluating the business performance of joint stock commercial banks

Table 4.2: Selected Income / Outcome Variables

DEA Model (Profitability)

• Total assets

• Owner’s Equity

• Operation cost

Pretax Profit

Source: Synthesis from reference materials and model design by the author

After estimation results of business performance of joint stock commercial banks from the DEA model, the thesis categorized the performance of joint stock commercial banks into 3 categories

c) Identification of default probability for joint stock commercial banks

For this indicator, the thesis calculated bad debt ratio of banks over the years from their published financial statements and combined these with the categories defined in section a) as well as the analysis of non-financial information in order to identify the default probability of banks Of the observations in category C, there are 39 observations with bad debt ratio from 3% and up The rest 70 observations of category C have bad debt ratio lower than 3% Analysis of these observations showed that, despite low business performance, these banks all have credit quality indicators satisfying the requirements from the State Bank and capital adequacy ratio CAR higher than 10%

Thus, banks are considered having high default probability if they belong to category C and have bad debt ratio from 3% and up – equivalent to status Y = 1

as opposed to Y = 0 in other status Results showed 39 out of 163 observations belonging to the high default probability group (Y = 1), accounting for 23.92% of all observations The rest of observations fallen under low default probability group (Y = 0) equal 124 observations, accounting for 76.08% of total

4.1.3 System of indicators affecting default probability

• Three indicators of macro element group: Gross domestic product; inflation rate; credit growth

• Indicators of micro element group: The author selected and built a total

of 39 indicators These were initially categorized into 7 groups: Profitability (11

indicators); Deficit index (3 indicators); Asset management (4 indicators); Asset

quality (7 indicators); Adequacy index (4 indicators); Sustainable growth index

(4 indicators); and Liquidity (6 indicators)

4.1.4 Statistical analysis

The author conducted correlation analysis to identify indicators within

groups having ability to recognize level of risk This resulted in 18 variables

Then, the thesis applied correlation analysis between any two variables within

the groups

4.2 The Logistic model with array data

From the set of 18 variables in Table 4.9 and 3 macro variables in Table

4.8, the research used the entry method in which predictors are entered one at a

time into the Logistic regression model with array data Hausman test resulted in

the adequacy of fixed effect model

After estimation results were out for

^

β of β, the next step was to estimate i

α Each bank has their characteristics represented by αi With a data set

expanding 5 years from 2010 to 2015, the calculation of αiled to solving quintic,

quartic or cubic equations depending on the number of years we have data for

each bank Matlab software is used to handle this work

Model result:

i

p

ln( ) -1.2955*RG DP 1.0346 * d3 2.014 * e11 3.0769 * l3

Source: The author’s calculation

in which p is the probability of observation being in the high default risk

group

+ Variable RGDP has negative effect on p at significance level of 6%,

while variable e11 has similar effect at significance level of 1%

+ Variable d3 has positive effect on p at significance level of 1% while

variable l3 has similar effect at significance level of 6%

Calculation of default probability and testing of the model efficiency shoed

the ratio of correct categorization for the Logistic model is at 87.71%

Other intercept values showed characteristics of each bank affecting its

default probability indicate that four banks coded 22, 14, 19, and 7 in the research

have high intercept values or high latent default risk compared to others

4.3 Neural network model

The data used in neural network model consists of 163 observations These are randomly categorized into 3 sub-sets including: i) Training set of 115 observations; ii) Validation set of 24 observations; iii) Test set of 24 observations

To identify the optimal number of neurons in hidden layer, the author used the iterative process to review the number of neurons until finding the smallest mean squared error (MSE) The author’s neural network structure consists of 21 input nodes corresponding to variables in Table 4.9 and Table 4.8, 10 nodes in hidden layer, and 2 output nodes The classification performance of neural network on the samples is presented in Table 4.19

Table 4.19: Neural Network Performance

Samples Training set Validation set Test set

Source: The author’s calculation

Applied on the data set of 114 observations used to run Logistic regression model with array data and fixed effect, the accuracy of classification for ANN model

is 92.98% compared to 87.71% of Logistic model Moreover, class I error of ANN model is also lower than Logistic model

4.4 Decision tree model

The author experimentally constructed decision tree model to forecast default probability for joint stock commercial banks using the data set of 163 observations Independent variables for decision tree consist of 21 variables in Table 4.9 and 4.8 The author used J48 algorithm on Weka software version 3.6.9

to create decision tree The algorithm in decision tree returned 5 positive

indicators for the classification

The accuracy level of classification using decision tree model on the 163-observation data set is 96.93%, which is a considerably high number With the data set of 114 observations (previously used in Logistic regression model), the decision tree model returned 95.61% accuracy rate

The author has summarized and compared classification results returned by the three models (including Logistic model, neural network model, and decision tree model) on different data sets and concluded that using neural network or

decision tree model will raise the classification accuracy

Conclusions for Chapter 4

In Chapter 4, the author has experimentally constructed default warning

model for joint stock commercial banks in Vietnam with following details:

calculation of business performance for banks, classification into performance

groups, and identification of default probability of banks The thesis also

constructed 39 financial indicators categorized into 7 groups to use in default

prediction

The Logistic model with array data and fixed effect has shown variables

with direct effect on default probability of banks including overdue debt/ debt

payable, net interest margin, net loan / total deposit Variable GDP reflects the

growth of the economy and represent macro elements with negative effect on

default probability of banks

The author also experimented with ANN and DT models in default

prediction for joint stock commercial banks Results showed these two models

had higher classification accuracy comparing to the Logistic model Decision tree

model showed 5 indicators used to alert of default risk in banks

CHAPTER 5: CONCLUSIONS AND POLICY

RECOMMENDATIONS 5.1 Obtained results

a) The Logistic model results with array data and fixed impact

From the initial 42 variables, then 18 variables, the final Logistic model

has only 4 variables left During estimation, the author has tested the selection of

fixed impact versus random impact, and the fixed impact model has been

selected Thus, the impact of variables RGDP, d3, e11, l3 toward default

probability is fixed impact This also means that the above-mentioned variables

affecting the default probability of banks are of similar trend, fixed within the

studied period, and sharing similarity between banks

+ Estimation result from variable RGDP model has β = − $1 1.29 This means

the RGDP variable has negative effect on default probability and Logistic model

result proves the impact of macroeconomic conditions, specifically the growth of

gross domestic product, on default probability of joint stock commercial banks in

Vietnam

+ Variable d3 = Overdue debt / Debt payable in the model has $

2 1 0 3

which is higher than 0 This variable indicates the deficit level of banks, and the

higher its value, the less safe the banks are This variable has positive effect on

default probability

+ Variable e11 = (Receivable from interest – Payable from loan interest) / Earning assets, has β = − $3 2 0 1 4 Net interest margin shows the difference between revenue from earning interest and cost from paying interest The author expected that variable e11 had negative effect on default probability Regression test result showed a negative value as expected

+ Variable l3 = Net loan / Total deposit with β = $3 3 0 7 This ratio represents the liquidity of banks and its higher value means the banks’ lower liquidity Bankrupt probability is under positive effect of this indicator

+ The intercept αi of banks are calculated in the model These indicate the differences and characteristics of banks which affecting their default probability According to calculation, there are four banks coded 22, 14, 19, and 7 with high intercept values, meaning high default probability The three banks coded 10, 21, and 6 have lowest intercept values, meaning lower default probability under the same condition of variables

b) Marginal effect of variables toward default probability p: Based on the

value of the coefficients on Logistic model, the author calculated the marginal effect of these variables on default probability

c) Summary and comparison of results of classification models:

The author compares the effectiveness of models with the data set of 114 observations in 2015 In different data sets, neural network model and decision tree model both have higher classification effectiveness compared to Logit model Especially, the number of observations wrongly classified by all models are minimal at 1 observation, thus, the combination of three models will bring higher classification effectiveness

5.2 Classification of joint stock commercial banks

From the results of Logistic model with array data, fixed effect in section 4.2 and the regulation of bank classification standard according to Decision 06/2008 of the State Bank; the author has classified banks into 4 categories The author then compares the research classification results with the actual classification results from the State Bank

5.3 Several proposals and implied policies

After establishing and testing several default prediction models for the joint stock commercial banks in Vietnam, the author proposes several actions as