Applying logistic model to predict the probability of default for construction enterprises in vietnam from 2014 to 2016

MINISTRY OF EDUCATION AND TRAINING THE STATE BANK OF VIET NAM BANKING UNIVERSITY OF HO CHI MINH CITY TRINH THANH DAT APPLYING LOGISTIC MODEL TO PREDICT THE PROBABILITY OF DEFAULT FOR C

MINISTRY OF EDUCATION AND TRAINING THE STATE BANK OF VIET NAM

BANKING UNIVERSITY OF HO CHI MINH CITY

TRINH THANH DAT

APPLYING LOGISTIC MODEL TO PREDICT THE PROBABILITY OF DEFAULT FOR CONSTRUCTION ENTERPRISES IN VIETNAM FROM 2014 TO 2016

GRADUATION THESIS MAJOR: FINANCE – BANKING

CODE: 7340201

INSTRUCTOR M.S TRAN KIM LONG

HO CHI MINH CITY - 2018

MINISTRY OF EDUCATION AND TRAINING THE STATE BANK OF VIET NAM

BANKING UNIVERSITY OF HO CHI MINH CITY

TRINH THANH DAT

APPLYING LOGISTIC MODEL TO PREDICT THE

PROBABILITY OF DEFAULT FOR CONSTRUCTION

ENTERPRISES IN VIETNAM FROM 2014 TO 2016

THE AUTHOR'S DECLARATION

Full name: Trinh Thanh Dat

Student class: HQ02-GE01, faculty of Banking and Finance, Banking University of Ho Chi Minh city

Student code: 030630141126

I declare that this thesis has been composed solely by myself and that it has not been submitted, in whole or in part, in any previous application for a degree Except where states otherwise by reference or acknowledgment, the work presented is entirely my own

Ho Chi Minh City, May 18, 2018

Author

Trinh Thanh Dat

THE AUTHOR'S ACKNOWLEDGEMENT

First of all, I would like to thank all lecturers at Banking University of HCMC Your enthusiastic and devoted instruction helped me to improve my logical thinking ability and knowledge

In addition, I would like to thank Mr Tran Kim Long who enthusiastically instructed and encouraged me to complete this graduation thesis

However, due to limited knowledge and practical experience and limited research time, the study cannot avoid certain shortcomings The author wishes to receive the comments

of members in the committee to complete the thesis

Ho Chi Minh City, May 18, 2018

Author

Trinh Thanh Dat

BANKING UNIVERSITY OF HO CHI MINH CITY VIETNAM

High-Quality Program of Banking and Finance

ABSTRACT

Author DAT, Thanh TRINH

Title Applying logistic model to predict the probability of default for

construction enterprises in Vietnam from 2014 to 2016

Year 2018

Language English

Instructor M.S LONG, Kim TRAN

In the current overall development of the economy, banking credit plays a very important role in the economy of every country in the world and is especially important for countries with underdeveloped financial markets like Vietnam because it is a main source

of funding for businesses However, recently, excessive credit growth, resulting in uncontrolled credit quality, has caused some consequences for the banking system such as: high credit risk, declining profit, liquidity reduced The paper focuses on building a model estimating credit risk for construction firms in Vietnam from 2014 to 2016 Based

on the results of the study, the paper provides not only an effective tool to predict the probability of default of construction companies but also comments and policy implications for commercial banks to improve the quality of credit and reduce credit risk

in the future

Key words: Credit risk, Logistic model, Basel II, Probability of default, Construction

companies, Vietnam

INDEX

LIST OF ACRONYMS 1

LIST OF TABLES AND FIGURES 2

CHAPTER 1: INTRODUCTION 3

1.1 Research background 3

1.2 Significance of research 3

1.3 Object and scope of the study 4

1.4 Research questions 4

1.5 Research methods 4

1.6 Structure of the themes 4

SUMMARY OF CHAPTER 1 5

CHAPTER 2: LITERATURE REVIEW AND THEORETICAL FOUNDATIONS 6

2.1 Credit risk (Default risk) 6

2.1.1 Definition 6

2.1.2 Measuring credit risk 7

2.2 Probability of default (PD) 10

2.2.1 Definition 10

2.2.2 Measuring PD 10

2.3 Some previous research on measuring PD 12

2.4 Model evaluation methods 22

2.4.1 Confusion matrix 22

2.4.2 Accuracy 23

2.4.3 Sensitivity 23

2.4.4 Specificity 23

2.4.5 Precision 23

2.4.6 F1 score 24

2.5 ROC Curve 24

2.5.1 Definition and some overviews 24

2.5.2 ROC‟s construction 25

SUMMARY OF CHAPTER 2 26

CHAPTER 3: MODEL ESTABLISHMENT 27

3.1 General concept 27

3.2 Building model 27

3.2.1 Logistic model and Model selection 27

3.2.2 Collection and cleaning data 27

3.2.3 Building models 32

3.2.4 Apply the models into estimating the PD in 2016 35

3.2.5 Choosing cutoff values 35

SUMMARY OF CHAPTER 3 36

CHAPTER 4: VALIDATING MODEL‟S PERFORMANCE AND EVALUATING RESULTS 37

4.3 Evaluation indicators 42

4.4 ROC Curve (Receiver operating characteristic curve) 43

4.5 The area under curve (AUC) 44

SUMMARY OF CHAPTER 4 45

CHAPTER 5: CONCLUSIONS 46

5.1 Final result 46

5.2 Limitations 46

5.3 Recommendations 47

5.4 Future research direction 47

PREFERENCES 48

APPENDIX 1: LOGISTIC MODEL EXPLANATION 54

APPENDIX 2: CODES IN R 56

APPENDIX 3: PROBABILITY OF DEFAULT OF LOGIT MODEL 59

APPENDIX 4: PROBABILITY OF DEFAULT OF PROBIT MODEL 60

APPENDIX 5: PROBABILITY OF DEFAULT OF C LOG-LOG MODEL 61

LIST OF ACRONYMS

CAGR Compound Annual Growth Rate CRAs Credit Rating Agencies EAD Exposure at Default

IRB Approach Internal Ratings Based Approach

LEF Loan Equivalency Factor

PD Probability of Default

ROC Curve Receiver Operating Characteristic Curve

SMEs Small and Medium-sized Enterprises

LIST OF TABLES AND FIGURES

Page

Figure 2.1.Relationship between Expected and Unexpected Loss 8

Figure 2.2 The Standardized Probit, Logit and C-Log-Log Links 12

Table 2.1 List of Ratios Tested 13

Table 3.1 Number of observations and defaults per year 27

Figure 3.1 Distribution of financial ratios before handling outliers 29

Figure 3.2 Distribution of financial ratios after handling outliers 30

Table 3.2 Financial ratios in six categories 33

Table 4.1 Descriptive statistical table 35

Table 4.2 Correlation matrix 36

Table 4.3 Regression table of Logit model 37

Table 4.4 Regression table of Probit model 38

Table 4.5 Regression table of C log-log model 39

Table 4.6 Matrix confusion for logit model at cutoff of 0.01 40

Table 4.7 Matrix confusion for probit model at cutoff of 0.01 40

Table 4.8 Matrix confusion for C log-log model at cutoff of 0.01 40

Figure 4.1 Receiver Operating Characteristic Curves of three models 42

In recent times, overdue debts and non-performing loans in Vietnamese commercial banks have become a serious problem, hindering the comprehensive development of the banking industry World Bank (2014) announced that Vietnam had the highest non-performing loans compared to other Southeast Asian countries Therefore, focusing on risk management in general and credit risk management in particular is considered as a guideline to ensure that the banking system operates stably While many countries around the world have applied Basel's recommendations in credit risk management, not many commercial banks in Vietnam have a completed Internal Rating – Based Approach In addition, retail banking in Vietnam has been developing significantly recently, Wang and Kapfer (2017) said that “The retail financial services industry in Vietnam will become the top growth market in Asia Pacific this year It is expected to surge by 29% in regard to retail assets compared to 2016 and the compound annual growth rate (CAGR) of retail income in Vietnam will reach $6.5 billion by 2020” Therefore, this sector needs to have

a huge amount of credit specialists to expertise credit profiles, and this work will take a lot of time and effort because every single credit analyst could handle a certain number of credit profiles Based on the reasons stated above, the author selects the topic "Applying logistic model to predict the PD for construction enterprises in Vietnam from 2014 to

2016”

1.2 Significance of research

Practical significance: After this model is established and verified, it will be one of the most useful tools for banks before they grant credit because it allows them to avoid

mistakes in assessing firms‟ potential and risk As a result, bank decisions will be more accurate and secured For companies, the model will indicate bad signals, which helps firms‟ leaders to fix mistakes in the nick of time and improve companies‟ status

Scientific significance: First of all, this study could develop a new model that examines the predictive power of classification models and discover new factors that help commercial banks as well as financial institutions develop automatic credit rating system Finally, diversification of credit rating models with additional baseline comparison between model types and verification of previous model types.

1.3 Object and scope of the study

Research object: real estate companies listed on Vietnam stock market from 2014

to 2016

Research scope: This study will use data from Financial statement (mostly from Income statement and Balance sheet) from real estate companies from 2014 to 2016 The data will be collected from www.bloomberg.com, www.vietstock.vn, www.cafef.vn and www.cophieu68.vn

Chapter 2: Literature review and theoretical foundations

Chapter 3: Model establishment

Chapter 4: Validating model‟s performance and evaluating results

Chapter 5: Conclusions

SUMMARY OF CHAPTER 1

This chapter points out clearly not only the urgency of the topic, but also the scope, the aim and the significance of the study In the next chapter, many Scientific research works, scientific papers as well as famous books which are relative to this study are review in details

CHAPTER 2: LITERATURE REVIEW AND THEORETICAL

 The market definition for default is related to financial instruments It corresponds

to principal or interest past due

 The Basel II definition considers a default event based on various alternative options such as past due 90 days on financial instruments or provisioning It can also be based on a judgmental assessment of a firm by the bank (Basel II, paragraph 452)

 The legal definition is linked with the bankruptcy of the firm It will typically depend on the legislation in various countries

However, because that the default information of companies in Vietnam is hardly accessible, companies in this study had to be assessed by „technical default‟ definition not truth default definition The Langohr (2015) defined that technical default occurs if a borrower breaches a financial obligation other than debt service payment For example, debt covenants may stipulate several balance sheet restrictions, such as minimum liquidity or solvency ratios, which the borrower has to respect in order to be allowed to reimburse a loan at maturity Were the borrower to violate these restrictions, the bank would have the right to call the loan immediately While such covenant violations constitute technical defaults, they typically do not trigger what is more normally called default, and they do not appear in the usual default statistics Nevertheless, such financial obligations will be of great interest to credit rating analysts A borrower‟s ability to honor any restrictions, and the likelihood that he will do so, clearly affects his subsequent ability to pay amounts due

on time and in full Hence an essential part of the CRAs‟ analysis is to scrutinize the

covenants of all an obligor‟s loan contracts, including those that are private or unlisted

2.1.2 Measuring credit risk

According to Stephanou and Mendoza (2005), there are several indicators measuring credit risk

The first one is Expected Loss (EL) and Unexpected Loss (UL), EL is based on three parameters:

 The likelihood that default will take place over a specified time horizon (PD or PD)

 The amount owned by the counterparty at the moment of default (exposure at default or EAD)

 The fraction of the exposure, net of any recoveries, which will be lost following a default event (loss given default or LGD)

Since PD is normally specified on a one-year basis 13, the product of these three factors is the one-year EL as follows:

EL = PD x EAD x LGD

Credit risk, in fact, arises from variations in the actual loss levels, which give rise to the so-called UL Statistically speaking, UL is simply the standard deviation of

EL (see Figure 2.1.2) As will be described later, the need for bank capital stems from the desire to cushion against loss volatility (UL) at a certain confidence level

The second one is Exposure at Default (EAD), EAD refers to the outstanding amount at the time of default In the simple case of a loan, the exposure is assumed to

be fixed for each year and can be derived from the agreed amortization plan In the case of derivatives, it requires more complex simulations to estimate the path that the underlying value of the asset may take and thus the potential future exposure that would arise In the case of lines of credits or any other type of revolving facility, EAD

levels of credit lines prior to the moment of default, as expressed in the following equation:

Exposure at Default = Current Exposure + LEF x Unutilised Portion of the

Limit

LEF is the Loan Equivalency Factor and represents the portion of the unutilized line that is expected to be drawn down before default

Figure 2.1: Relationship between Expected and Unexpected Loss

Source: Stephanou and Mendoza (2005)

The next one is Loss Given Default (LGD), LGD is the percentage of the EAD that is lost in the event of a default Its calculation requires answering the following questions:

 How much is recovered and from where (e.g collateral liquidation)?

 How long did it take to recover and what is the financial cost (i.e interest income forgone) associated with this period of time?

 How much had to be spent in the recovery process (i.e workout expenses)?

There are three main types of LGD measurement (Schuermann, 2003) The first one (workout LGD) is based on the estimation and timing of cash flows and costs from the workout process:

LGD =

The second type of LGD measurement (market LGD) can be observed from market prices and trades of defaulted bonds or loans after the actual default event In the exceptional case that a liquid distressed debt market exists, LGD can be written as:

LGD =

The third type of LGD measurement (implied market LGD) is derived from the prices of fixed income and credit derivatives products using a theoretical asset-pricing model This approach essentially „backs out‟ LGD from the credit spreads of risky (but not defaulted) bonds or from credit default swap prices, but it is limited in scope (traded debt only) and subject to methodological problems (credit spread also reflects

PD and potentially a liquidity premium, tax considerations etc.)

After that is Unexpected Loss (UL) for a Single Loan, UL is simply the volatility in the components of EL that were described above:

UL = ( ) ( )

In order to solve this equation, it would be necessary to know the standard deviation of all three variables In the case of PD, since it reflects an underlying Bernoulli variable (i.e a variable than can only have two states – the counterparty defaults or not), its standard deviation is equal to:

(PD) = PD (1- PD)

Some practitioners also assume that the variance of EAD and LGD is zero As

a result, the UL equation for a single loan (what is often referred to as the „stand-alone UL‟) is often simplified as follows:

UL = √ ( )

Eventually, the last indicator is the PD which will be presented in details in the next section

2.2 Probability of default (PD)

2.2.1 Definition

PD is a critical ratio applied in a variety of credit risk analyses and management frameworks Under Basel II, it is a key parameter used in the calculation

of risk capital or regulatory capital for banking institution

Default Probability (DP) is a PD measures the likelihood of a borrower‟s default With a stable credit state, the chances of defaulting increase with the horizon

as the credit state and the DP changes as the credit standing migrates Basel requires the usage of annual DPs (Bessis, 2015)

Office of the Comptroller of the Currency (2012) claimed that: “PD is the risk that the borrower will be unable or unwilling to repay its debt in full or on time The risk of default is derived by analyzing the obligor‟s capacity to repay the debt in accordance with contractual terms PD is generally associated with financial characteristics such as inadequate cash flow to service debt, declining revenues or operating margins, high leverage, declining or marginal liquidity, and the inability to successfully implement a business plan In addition to these quantifiable factors, the borrower‟s willingness to repay also must be evaluated”

In general, PD in this study based mostly on Basel II, which is an event that a borrower(s) cannot make payment to the creditor within 90 days from the mature day

- Logit model, Probit model and C log-log model…

However, in this graduation thesis, Logistic model is mainly focused on as a main model in estimating the PD

Logistic regression is a statistical method is suited and usually used for testing hypothesis about relationships between a categorical dependent or an outcome variable and one or more categorical or continuous predictor or independent variables The dependent variable in logistic regression is binary or dichotomous The maximum likelihood method, which yields values for the unknown parameters, is used for estimating the least squares function Logistic regression solves such problems applying the logit transformation Logistic regression predicts the logit of Y to X Logit model could be written as:

pi

p: PD

x: financial ratios

To create the best model, we find the set of weights β to create the best fit between Pi and observations are default It means that we want Pi close to 100% for default customers and near to 0% for non-default customers This can be done by maximizing Likelihood function (this method is called Maximum Likelihood Estimation (MLE))

In addition, two other models which are Probit model and Complementary Log-Log model are also used in order to make comparison among three of them because all of them belong to one family which is called Biominal family

Under the assumption of binary response, there are two alternatives to logit model: probit model and complementary-log-log model They all follow the same form:

( ) ( )

Probit model can be written as: ( ) =

√ ∫ ( )

C log-log model can be written as: ( ) = ( )

Figure 2.2: The Standardized Probit, Logit and C-Log-Log Links

Source: Rodríguez (2007)

2.3 Some previous research on measuring PD

One of the earliest studies which belonged to Beaven (1966) focused on how important financial ratios are in conducting bankruptcy predictions In this research, Beaven used various different ratio groups such as cash flow ratios, net income ratios, debt to total asset ratios, liquid asset to total asset ratios, liquid asset to current debt ratios and turnover ratios It is used very often for decision-making processes by lenders, rating agencies, investors, regulators, management and others The research got information to calculate financial ratios from financial statements that were created based on internationally accepted reporting standards The final list of failed firms was picked from the Moody‟s Industrial Manual database included 79 failed companies form the

period 1954-1964 These corporations were from 38 different industries The classification of failed firms according to industry and asset size, and that also was the classification requirement for selection of non-failed companies After that, Beaven did the following steps Computation of Ratios, Comparison of Mean Values, Theory of ratio Analysis, Analysis of Evidence, Comparison of Mean Asset Size, Comparison with previous studies, Limitation of profile analysis, Dichotomus Classification test, Analysis

of Evidence Finally, he showed two conclusions: Not all ratios predict equally well The cash flow to total debt ratio has excellent discriminatory power in five years However, predictive power of the liquid asset ratios is much weaker The ratios do not predict failed

or non-failed firms with a same degree of success, non-failed firms can be more correctly classified than failed companies

Table 2.1: List of Ratios Tested

Group I (CASH-FLOW RATIOS)

1 Cash flow to sales

2 Cash flow to assets

3 Cash flow to net worth

4 Cash flow to debt

Group II (NET-INCOME RATIOS)

1 Net income to sales

2 Net income to assets

3 Net income to net worth

4 Net income to debt

Group III (DEBT TO TOTAL-ASSET RATIOS)

1 Current liabilities to total assets

2 Long-term liabilities to total assets

3 Current plus long-term liabilities to total

assets

4 Current plus long-term plus preferred

stock to total assets Group IV ((LIQUID-ASSET TO TOTAL-ASSET

RATIOS)

1 Cash to total assets

2 Quick assets to total assets

3 Current assets to total assets

4 Working capital to total assets

Group V ((LIQUID-ASSET TO CUR RENT DEBT RATIOS)

1 Cash to current liabilities

2 Quick assets to current liabilities

3 Current ratio (current assets to current liabilities)

Group VI (TURNOVER RATIOS)

1 Cash to sales

2 Accounts receivable to sales

3 Inventory to sales

4 Quick assets to sales

5 Current assets to sales

6 Working capital to sales

7 Net worth to sales

8 Total assets to sales

9 Cash interval (cash to fund expenditures for operations)

10 Defensive interval (defensive assets to fund expenditures for operations)

11 No-credit interval (defensive assets minus current liabilities to fund expenditures for operations)

Source: Beaven (1966)

In 1968, Altman created a foundation for many default prediction models later, which is now known as Z – Score model The purpose of this paper was to attempt an assessment of this issue-the quality of ratio analysis as an analytical technique This study chose a sample comprising 66 firms in total with 33 companies for each group Group 1 included manufacturers that filed a bankruptcy petition under Chapter X of the National Bankruptcy Act in the period of 1946-1965, while the other group included the same number of non-bankrupt companies The average size of the sample companies was $ 6,4 million Set of financial ratios proposed in the literature was used in the study to as potential predictors of bankruptcy Financial data up to five years prior to bankruptcy

were taken into consideration Used financial ratios were divided into five standardized groups: liquidity, profitability, leverage, solvency and activity

*The final discriminant function is:

X4 = market value of equity / book value of total liabilities Adds market dimension that can show up security price fluctuation as a possible red flag

X5 = sales / total assets Standard measure for total asset turnover (varies greatly from industry to industry)

Z score of greater than 2.99 indicates that the entity measured is safe from bankruptcy A score of less than 1.81 means that a business is at considerable risk of bankruptcy, while scores in between are considered a red flag This approach brings together the effects of multiple items - assets, profits, and market value, which is better than using just a single ratio Over time, the Z-Score has been one of the most reliable predictors and widely used by creditors to determine the risk of bankruptcy

Altman then continued to develop Z-score into Z‟ and Z” for better application to variety of firm categories For detail:

Z' = 0.717X1 + 0.847X2 + 3.107X3 + 0.42X4 + 0.998X5, applied in de-equitized manufacturing firms

Z" = 6.56X1 + 3.26X2 + 6.72X3 + 1.05X4, applied in other cases

Eventually, Altman concluded that traditional ratio analysis is no longer an important analytical technique in the academic environment due to the relatively

unsophisticated manner in which it has been presented In order to assess its potential rigorously, a set of financial ratios was combined in a discriminant analysis approach to the problem of corporate bankruptcy prediction In addition, he supposed that the former includes business credit evaluation, internal control procedures, and investment guidelines Inherent in these applications was the assumption that signs of deterioration, detected by a ratio index, can be observed clearly enough to take profitable action A potential theoretical area of importance lies in the conceptualization of efficient portfolio selection One of the current limitations in this area was in a realistic presentation of those securities and the types of investment policies which were necessary to balance the portfolio and avoid downside risk The ideal approach was to include those securities possessing negative co-variance with other securities in the portfolio

Engelmann, Hayden, and Tasche (2003) provide an empirical comparison of the logit model and Altman‟s Z-score (discriminant analysis) on a large sample of SMEs They show a very significant out performance of the logit model in terms of rank ordering (the ROC coefficient)

Four years later, Deakin (1972) wanted to find out an alternative model for forecasting failure He used multiple discriminant analysis, which found a linear combination of ratios that discriminated between the desired groups, which should be classified The aim of this research was to re-conducted previous studies of Beaven (1966) and Altman (1968), the author searched for the linear combination of 14 ratios that Beaven used The sample consisted of 32 failed companies that experienced failure between 1964 and 1970, where by failure any form of bankruptcy, insolvency or liquidation was included The failed companies were matched with non-failed ones on the basis of industry classification, year and the asset size Financial data (financial ratios) for sampled companies were collected for the period of up to five years prior to bankruptcy Each of failed firms was matched with a non-failed firm in terms of industry classification, year of the financial information provided and asset size Deakin realized that the relative importance of the variables changes over the five years prior to bankruptcy and that almost all of the variables contribute significantly to the discriminant

ability of the function The final results indicated that the power of prediction had fallen from the third year, with the increase in error rate are 21% and 17% in fourth and fifth year respectively However, the sample size in this study was quite small

Martin (1977), Ohlson (1980), and Wiginton (1980) applied logit analysis to the problem of bankruptcy prediction Ohlson (1980) took logistic regression analysis into building the bankruptcy prediction models This study relied on observations from 105 bankrupted and 2058 matched non-bankrupt industrial firms between 1970 and 1976 in building the prediction models In addition, this research was different from Altman & McGough (1974), Moyer (1977) and Altman, Haldeman & Narayaman (1977) in terms of methodology and objective Ohlson used the cumulative logistic function in transforming the value of dependent variable to the bankruptcy probability Then the obtained probability was compared with 0.5 to determine the company will be classified as the healthy company or the bankrupt one Furthermore, the data from Ohlson‟s study was collected from 10-K not from Moody‟s Manual The author assumed that there were four factors that were statistically significant to the probability of failure, they are: the size of

a company, measures of financial structure, measures of performance and measures of current liquidity Overall, Ohlson‟s model had a very good predictive capability The correct classification rate is above 95% with independent variables one and two years prior to bankruptcy Their predictive abilities were 96,12 %, 95,55 % and 92,84 % Ohlson concluded that the predictive power of any model depends on the information (financial reports) is assumed to be available In addition, predictive powers of linear transforms of a vector of ratios seem to be robust across estimation procedures Hence, significant improvements require additional predictors However, this study attempts to resolve several issues he does not adequately address First, Ohlson suffers from the lack

of theoretical determination of his model He uses a conditional logit model to classify failing and healthy firms But he selects the independent variables without benefit of theory, assuring problems similar to those observed for discriminant analysis For example, asset size is a variable with high significance in his model but is also a scale factor in other ratios

Lau (1987) presented a model which was an extended version of previous models

in corporate failure prediction such as Beaver (1966), Altman (1968), Altman, Haldeman

& Narayaman (1977), Ohlson (1980) and Zavgren (1985) in two points First of all, instead of the failing and non-failing dichotomy it represented five financial states that approximate the financial state of a company Secondly, instead of classifying the firm into one of two financial states, this study provided a probability that a company would enter each of the five predefined financial states These are financial stability state, omitting or reducing dividend payments state, technical default and default on loan payments state, protection under X or XI of Bankruptcy Act and bankruptcy and liquidation The results showed that each of the used explanatory variables assumed different values for companies in different predefined states The paired t-tests showed that for each variable the five state means are significantly different In order to construct the model multi-nominal logit analysis (MLA) was used The authors also used multiple discriminant analysis (MDA) to compare the study results The results showed that MLA outperformed MDA The author reported overall 96% prediction accuracy for the model that used one year before default data, 92% prediction accuracy for the model that used two years before default data and 90% prediction accuracy for the model that used three years before default data

Memić (2015) conducted a study with the aim to assess the PD occurrence on the banking market from Bosnia and Herzegovina In other word, the main purpose of the study is predicting credit default, or to create a prediction model that distinguishes defaulted and non-defaulted companies, based on the financial data obtained from their financial statements Its main goal is to find the best fitting model that best describes the relationship between an outcome and the set of independent variables The main mathematical concept under the logistic regression is the logit or the natural logarithm of

an odds ratio Based on the previous default prediction research, list of most frequently used financial ratios was assessed, and calculated for each defaulted and healthy company in the sample The data patterns were analyzed for the total data set and for each of the groups of companies separately (defaulted and healthy group) Two main

groups of methods were used to test the posed hypothesis and answer the research questions The financial data for this study will obtained from AFIP database for Federation of Bosnia and Herzegovina and from APIF database for Republic of Srpska data These databases consist of financial statements of all of the companies registered in Federation of Bosnia and Herzegovina and Republic of Srpska The total number of legal entities registered in AFIP database of Federation of Bosnia and Herzegovina in the year

of 2010 exceeds 20000 while in APIF database of Republic of Srpska the number of legal entities exceeds 9000 Finally, the author concluded that among four logistic regression models, some variables were more influential on the default prediction than the others Observing logistic regression models, return on assets was statistically significant in all four periods prior to default, having very high regression coefficients, or high impact on the model‟s ability to predict default In addition, best performing multiple discriminant analysis default prediction models were selected based on scored hit ratios represented by classification tables, eigenvalues, canonical correlations and Wilks‟ lambda values of created MDA models, as well as on the strength and direction of the impact of chosen predictors on default Two models were created using large corporate data sets, while the other two were created on the basis of Federation of Bosnia and Herzegovina data sets

Zavgren (1985) conducted a research with the aims to use the logit technique to develop and test a new bankruptcy model which enumerates the signs of financial ill health for a five-year period prior to failure Additionally, he would want to develop a methodology for evaluating the significance of probabilities of financial risk The models proved highly significant with reference to both the R2 and likelihood ratio tests in detecting ailing firms up to five years prior to their failure This study employs all the ratios which Pinches et al identified as most strongly related to their seven factors, with the exception of the current ratio, which is a measure of short-term liquidity The current ratio increases in proportion as the failing firm‟s unsaleable inventories accumulate, and

so it provides a misleading measure of liquidity If, however, one substitutes the current ratio with the acid test ratio, one can then ignore inventories, and derive a clearer picture The variables chosen represent an empirically determined attribute vector of the firm,

according to which one can make unprejudiced, accurate distinctions between failing and healthy firms

Zavgren supposed that Models which generated a probability of failure as a cardinal measure of risk proved to be more useful than the dichotomous classification usually obtained from discriminant analysis models The latter turns out to be too stringent a partition of the outcome space for most decision settings The models estimated here were found to be highly significant (at greater than the 99% confidence level) in distinguishing between failing and healthy firms over the five-year period The amount of information over the subsequent five-year period increases by an average of 18 per cent for the failed firms and 16 per cent for the non-failed firms Classification and prediction error rates were also evaluated for the models They were found to compare favorably with other models, especially for prediction ability when the stringency of the inter-temporal generalizability test is considered The significance of the coefficients for each of the variables in the models was traced for each of the five years The pattern of significance was found to be highly congruent with an expectation The efficiency ratios were found to have the most significance over the long run, which indicated that efficiency in the utilization of assets is difficult to modify over the short run Financial ratios can provide highly significant measures for evaluating bankruptcy risk In addition, the pattern of significance of the coefficients in these models indicates that these variables would be important for helping a manager or analyst to assess risk

Basel II is the second version of the Basel Convention, which sets out the general principles and banking rules of the Basel Committee on Banking Supervision The Basel

II Basel Convention, presented as a proposed set of rules, would bring about a series of compliance challenges for banks around the world Basel II allows firms to use one of two broad approaches to the calculation of capital:

• Standardized Approach: uses supervisory risk weights to calculate capital based primarily on the asset classes

• Internal Ratings Based Approach: Banks are allowed to use their internal estimates of borrower creditworthiness to assess credit risk in their portfolios, subject

to strict methodological and disclosure standards Under the IRB approach, banks estimate each borrower‟s creditworthiness and translate the results into estimates of a potential future loss amount

According to Stephanou and Mendoza (2005), Basel II consists of a broad set of supervisory standards to improve risk management practices, which are structured along three mutually reinforcing elements or pillars:

· Pillar 1, which addresses minimum requirements for credit and operational risks

· Pillar 2, which provides guidance on the supervisory oversight process

· Pillar 3, which requires banks to publicly disclose key information on their risk profile and capitalization as a means of encouraging market discipline

In 2011, Engelmann and Rauhmeier co-published a book called The Basel II Risk Parameters The first and second chapter were written by Hayden and Porath introduced and distinguished many statistical methods from each other such as discriminant analysis, regression analyst, probit and logit models, panel models, Hazard models, neutral networks and decision trees However, these authors decided to choose logit model as the model for their research because of some of its advantages which will be explained on the next chapter In the second chapter, Hayden guided how to predict PD of 2283 firms from

1995 to 1999 and introduced how to process the data such as deal with outliers and how

to run test of Linearity of Assumption and some knowledge that relative to my study

In Viet Nam, in some recent years, credit rating models have received a rising concern among researchers Many articles were published: Nguyen (2004) interpreted the use of an internal database system to assess credit risk, thereby identifying a minimum capital adequacy ratio, on basis of Basel II Additionally, Tran (2003) proposed a solution

to improve the internal rating model that are currently used by Vietnamese commercial banks Finally, Tram (2002) with the introduction of a method to use financial indicators

to classify loan applicants

Thomas, Edelman and Crook (2002) defined that Credit scoring is the set of decision models and their underlying techniques that aid lenders in the granting of consumer credit These techniques decide who will get credit, how much credit they

should get, and what operational strategies will enhance the profitability of the borrowers

to the lenders

Credit-scoring techniques assess the risk in lending to a particular consumer One sometimes hears it said that credit scoring assesses the creditworthiness of the consumer, but this is an unfortunate turn of phrase Creditworthiness is not an attribute of individuals like height or weight or even income It is an assessment by a lender of a borrower and reflects the circumstances of both and the lender's view of the likely future economic scenarios Thus, some lenders will assess an individual as creditworthy and others will not One of the longer-term dangers of credit scoring is that this may cease to

be the case, and there will be those who can get credit from all lenders and those who cannot Describing someone as uncreditworthy causes offense It is better for the lender

to state the reality, which is that the proposition of lending to this consumer represents a risk that the lender is not willing to take According to Bessis (2015), Credit scoring uses techniques for finding the criteria that best discriminate between defaulters and non-defaulters Scoring does not rely on conceptual models, but on statistical fits of “scoring functions” These functions provide a score derived from observable attributes, which is a number Scoring applies best for retail portfolios

2.4 Model evaluation methods

There are 4 parameters that are:

 True Positives (TP) - These are the correctly predicted positive values

which means that the value of actual class is yes and the value of predicted class is also yes

 True Negatives (TN) - These are the correctly predicted negative values

which means that the value of actual class is no and value of predicted class

is also no

 False Positives (FP) – These are the incorrectly predicted positive values

which means that when actual class is no and predicted class is yes

 False Negatives (FN) – These are the incorrectly predicted negative values

which means that when actual class is yes but predicted class in no

2.4.2 Accuracy

Accuracy =

Accuracy is the most visual performance measurement in validating the performance of a model The higher this indicator is the more accurate this model

is but only when the dataset is symmetric that means the positives and negatives are almost the same Therefore, I have to look at some other measurements

This indicator illustrates how much this model predicts correctly in percentage to all estimated-defaulted cases The higher this ratio is, the more accurate this model

F-2.5 ROC Curve

2.5.1 Definition and some overviews

“In statistics, a ROC curve, is a graphical plot that illustrates the diagnostic ability of a binary classifier system as its discrimination threshold is varied The Total Operating Characteristic (TOC) expands on the idea of ROC by showing the total information in the two-by-two contingency table for each threshold” (Pontius et al, 2014) ROC gives only two bits of relative information for each threshold, thus the TOC gives strictly more information than the ROC

ROC analysis provides tools to select possibly optimal models and to discard suboptimal ones independently from (and prior to specifying) the cost context or the class distribution ROC analysis is related in a direct and natural way to cost/benefit analysis of diagnostic decision making

The ROC is also known as a relative operating characteristic curve, because it

is a comparison of two operating characteristics (TPR and FPR) as the criterion changes (Swets, 1996)

The diagnostic performance of a test, or the accuracy of a test to discriminate diseased cases from normal cases is evaluated using Receiver Operating Characteristic (ROC) curve analysis (Metz, 1978; Zweig & Campbell, 1993) ROC

curves can also be used to compare the diagnostic performance of two or more laboratory or diagnostic tests (Griner et al., 1981) When you consider the results of a particular test in two populations, one population with a disease, the other population without the disease, you will rarely observe a perfect separation between the two groups Indeed, the distribution of the test results will overlap, as shown in the following figure

An overview of possible applications of the ROC curves is given by Swets (1988) Sobehart and Keenan (2001) introduce the ROC concept to internal rating model validation and focus on the calculation and interpretation of the ROC measure Engelmann, Hayden and Tasche (2003) show that AR is a linear transformation of AUROC; their work complements the work of Sobehart and Keenan (2001) with more statistical analysis of the ROC Satchell and Xia (2007) further explored the statistical properties of the ROC Curve and its summary indices, especially under a number of rating score distribution assumptions

If the line expressing the relationship between Sensitivity and Specificity is a straight line with 45 degrees, it is the worst predicting model because Sensitivity increases 1 unit and Specificity decreases 1 unit too Therefore, it is just a simple

tradeoff no gain at all According to Satchell and Xia (2007), ROC curve can also be

used to analyze the accuracy ratios A ROC curve demonstrates several things:

 It shows the tradeoff between sensitivity and specificity (any increase in sensitivity will be accompanied by a decrease in specificity)

 The closer the curve follows the left-hand border and then the top border of the ROC space, the more accurate the test

 The area under the curve is a measure of accuracy

SUMMARY OF CHAPTER 2

In this chapter, all of the theoretical foundations, definitions and scientific studies which are relative to the research are reviewed and presented deeply This compromises credit risk definition and the approach to measure it; PD definition and how to measure it; and finally, some Biominal models and financial ratios in creating estimating PD model In the following chapter, the process of constructing the model will be presented step by step

CHAPTER 3: MODEL ESTABLISHMENT

3.1 General concept

(1) The author uses a data set which includes variables (financial ratios) of construction companies in Vietnam in 2014 and their status in 2015 to build up a predicting default model by applying logit model in R

(2) Applying the model: put the variables from 2015 to get the results that show construction firms‟ status in 2016 After that, the author compares the results with the real status in 2016 and make final conclusion

3.2 Building model

3.2.1 Logistic model and Model selection

Logistic model is chosen because of some reasons (Engelmann & Rauhmeier, 2011):

 The output from the logit model can be directly shows the results whether these firms default or not in percentage

 The model allows an easy check as to whether the empirical dependence between the potential explanatory variables and default risk is economically meaningful

3.2.2 Collection and cleaning data

The samples selected for the study include information from 127 construction firms listed on the Hanoi Stock Exchange, Ho Chi Minh Stock Exchange and Upcom from 2014 to 2016 Data is collected from financial statement, balance sheet and performance statement from www.bloomberg.com, www.cafef.vn, www.vietstock.vn and www.cophieu68.vn I collect data (financial statements) of various construction firms in Vietnam from 2014 to 2016 because of some reasons:

 According to Finlay (2012), The data used for model construction should also be as similar as possible to the data that will exist when the completed model is put into service – which usually means that the