Collateral liquidity and loan default risks the case of vietnam

In Loan risk probability of default Collateral pledge Borrower selection effect - Lender selection effect + Risk shifting effect - Loss mitigation effect +... Hence, the borrower se

UNIVERSITY OF ECONOMICS ERASMUS UNVERSITY ROTTERDAM

HO CHI MINH CITY INSTITUTE OF SOCIAL STUDIES

VIETNAM – THE NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

COLLATERAL LIQUIDITY AND LOAN

DEFAULT RISKS: THE CASE OF VIETNAM

BY

NGUYEN LE HIEU

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

HO CHI MINH CITY, Dec 2016

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

HO CHI MINH CITY THE HAGUE

VIETNAM THE NETHERLANDS

VIETNAM - NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

COLLATERAL LIQUIDITY AND LOAN

DEFAULT RISKS: THE CASE OF VIETNAM

A thesis submitted in partial fulfilment of the requirements for the degree of

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

D E C L A R A T I O N

By these statements, I declare that the thesis titled “Collateral liquidity and loan default risks: the case

of Vietnam” is result of my own works and efforts All the contents in this thesis are my study based

on reviewing some previous papers which are clearly indicated in references In addition, this thesis has not been submitted to get any other degrees or certifications

Signature

NGUYEN LE HIEU

December 2016

A B S T R A C T

This thesis investigates the impact of the liquidity level of collaterals on the probability of default of individual loans and examines the channels through which collaterals affect default risks Following the approach of Jiménez and Saurina (2004), binominal logit model is applied on the data from individual loan accounts of a medium – size commercial bank in Vietnam The empirical results suggest the significant and negative impact of collaterals’ liquidity on loans’ probability of default, supporting the dominance of borrower selection effect and risk shifting effect over lender selection effect Moreover, the finding also implies that bank has not applied carefully and thoroughly screening process on loans that are fully secured by low liquid collaterals and therefore impaired the credit quality of loan portfolio

CONTENTS

CHAPTER 1 1

1.1 Research background and motivation 1

1.2 Research objectives and research questions 4

1.3 Research Methodologies and Data 5

1.4 Research Contribution 5

1.5 Structure of thesis 6

CHAPTER 2 7

2.1 Theoretical review of relationship between collaterals and loan risks 7

2.2 Empirical review of relationship between collaterals and loan risk 12

2.3 Theoretical framework 16

CHAPTER 3 17

3.1 Research methodology 17

3.2 Data 19

CHAPTER 4 23

4.1 Descriptive Statistics and Pre-estimation tests 23

4.2 Empirical results 25

4.3 Robustness test 30

CHAPTER 5 35

5.1 Main findings & conclusion 35

5.2 Policy implications 35

5.3 Research limitation and further research 37

REFERENCES 39

APPENDIX 42

L I S T O F T A B L E S A N D F I G U R E S

Table 1: Summary of variables 21

Table 2: Summary of loans characteristics 23

Table 3: Summary of default loans according to liquidity levels of collaterals 24

Table 4: Summary of default loans according to varied amounts of loans 24

Table 5: Summary of loans default according to rate of protection 25

Table 6: Summary of loans default according to loan time 25

Table 7: Estimation results of Logit model 26

Table 8: Estimation results of Logit model (exclude interest and loan time factors) 27

Table 9: Estimation results of second logit model for robustness test 31

Table 10: Estimation results of the third logit model for robustness test 33

Figure 1: NPL rate of Viet Nam for the period from Dec-2012 to Jun-2013 2

Figure 2: NPL of Viet Nam for the period from Jun-2014 to Dec-2015 3

Figure 3: The transmission channels of collaterals on loan risk 7

Figure 4: Screening cost prorated 9

Figure 5: House price index of HCM city from 2009 to Q3-2016 36

CHAPTER 1

I N T R O D U C T I O N

1.1 Research background and motivation

Non-performing loans (NPL) are a severe problem for the whole economy of the world since they lead to the financial crisis in East Asian countries, America and Sub-Saharan Africa (Farhan, Satta, Chaudrhy & Khalil, 2012) Therefore, finding out the main determinants of NPL plays an important role in policy making in order to prevent the future bad debts (Adebola, Wan Yusoff, & Dahala (2011) in Farhan et al (2012)) Previous studies identify that macro-economic conditions, bank and borrowers specific characteristics, loans characteristics, relationship banking1, and collaterals are key drivers of default risks and hence NPL

The relation between collateral characteristics and loan default is investigated in many studies over the world However, the findings are inconsistent among different papers, some of which show positive relationship while the others provide evidence of a negative effect

Berger, Frame and Ioannidou (2011) find a positive relationship between collaterals pledge and ex-post NPL in Bolivia for the period from 1998 to 2003 This result is supported by Jiménez and Saurina (2004) for Spain Berger and Udell (1990) in Leitner (2006) shows that borrowers who pledge collaterals tend to be worse and therefore are riskier Leitner (2006) explains that this finding is due to collaterals’ requirement of banks for riskier borrowers In contrast, John, Lynch and Puri (2003) investigate the yield difference between secured and unsecured loans in US and conclude that higher yield is decided by secured loans This result implies that borrowers who pledge collaterals are more efficient than others Kugler and Oppes (2005) investigate the impact of collaterals2 on loans risk in case of group lending in a developing country and find that collaterals are used by individuals to prevent loans default under joint borrowing

Berger et al (2011) argue that the diversified findings about this relationship arise from the variation of data samples which include different types and characteristics of collaterals Moreover, previous papers investigate only the impact of collaterals on NPL by comparing the default risk (probability of default) between secured loans and unsecured loans To my knowledge, there is very limited work on the impact of different collateral types and characteristics on default risk Berger et al (2011) find that liquid collaterals decrease the probability of default when compared to non-liquid collaterals However, previous papers mainly focus on loans for companies/enterprises rather than individual and consumer loans

In Vietnam, bad debt has increased sharply since 2011 and still been serious until now As we can see in Figure 1 and 2, NPL ratio has risen from 4.08% in Dec-2012 up to 4.67% in April-

2013, then decreased lightly to 4.46% in Jun-2013 and kept declining to 2.58% in Jun-2016 However, this does not represent an improvement in loan quality of banks but due to banks’ switching to other asset titles in the balance sheet to hide bad debts The Viet Nam Assets Management Company (VAMC) was established on July-2013 with its main objective is bad debts purchasing by issuing special bonds for payment As of Jun-2016, VAMC purchased about 251.000 billions of bad debts from banks and only 15% of these bad debts were collected (VAMC, 2016) Hence, these purchased bad debts help to reduce NPL of banks but they were not collected in reality and still harm the whole economy Furthermore, many bad debts have been restructured but still classified as normal debts instead of bad debt in almost Vietnamese banks (this problem is permitted by the State Bank of Viet Nam) and therefore these bad debts were hidden

Figure 1: NPL rate of Viet Nam for the period from Dec-2012 to Jun-2013

Source: State Bank of Viet Nam in http://tapchitaichinh.vn/

Figure 2: NPL of Viet Nam for the period from Jun-2014 to Dec-2015

Source: State Bank of Viet Nam

Ogeisia et al (2014) argue that lending in low income countries is notoriously risky because

of information asymmetry problem which are high in developing countries United Nations Conference on Trade and Development - UNCTAD (2005) explains that high level of information asymmetry arises from weak credit information infrastructure, ineffective public

records, lack of credit management skills and underdeveloped financial intermediation, which

is worse by generally restrictive and complicated regulatory environment and a large informal cash-based economy Majority of loans in Viet Nam are collateralized loans due to information asymmetry Although Vietnamese banks determine that loan approval is always based on the payment ability of borrowers, not collaterals, but in practice, collaterals is the most important condition that make a loan to be approved Loans in Viet Nam are 100% secured loans, therefore the difference of NPL between banks depends on the various quality

of banks’ screening procedures Higher efficient banks focus on the screening quality and ask for collaterals only for increasing the borrowers’ incentive for loan repayment to avoid asset loss In contrast, smaller banks who have higher cost, may loosen the borrowers screening quality and use collateral as the premier protection from loan loss Collateral plays the most important role in NPL control in small banks in Viet Nam The information conflict between lender and borrower might be mitigated by collateral according to Berger et al (2011) and therefore mitigate loan approval for the optimists However, Manove and Padilla (1999) argue that collaterals can not help to distinguish the optimists and realist and therefore make the PD prediction base on collateral requirement is unclearly The reason is that the optimists always tend to accept the collaterals’ requirement conditions like the realists to get lower cost loans due to their confidence in the efficiency of their projects Furthermore, different characteristics of collaterals may have different impact on PD of loans due to results found by

by Berger et al (2011) as mentioned above However, liquid collaterals in their work are only Deposit and Bank guarantee and non liquid ones are the other type of assets while in Viet Nam, the most popular collaterals are real estates and vehicles which are diversified in types and therefore in liquidities Each type of them is ranked in one level of liquidity and desirability and this level is determined by banks According to these above problems, an investigation about the impact on PD of different collaterals which diversify in characteristics for the case of Viet Nam should be implemented

1.2 Research objectives and research questions

The objective of this paper is to examine the impact of collaterals’ liquidity characteristics on loans’ probability of default (PD) at the commercial banks in Vietnam

In order to achieve that research objective, this thesis aims to seek convincing answers for the following research questions:

 Do higher liquidity levels of collaterals decrease the PD of loans?

 If so, through which channels this effect is transmitted?

1.3 Research Methodologies and Data

This paper applies the logit model to examine the responses of different liquidity levels of collaterals, loans amount and ranks of protection rates of loans on PD of personals loans All predictors are categorical variables and the response takes only one of two categories at the same time: default and non - default

Data of this research is collected from internal loans account data source of a medium size Vietnamese bank Loans accounts are first generated in 3 years as 2010, 2011, 2012 and from business units placed in Ho Chi Minh city and Hanoi In order to investigate the direct impact

of collaterals on loans default, other factors that potentially affect loan’s PD are minimized by collecting loan accounts that are secured by only one collateral at one point of time during the period from 2010 to 2012 Hence, 2,295 observations are included in the research’s empirical analysis

1.4 Research Contribution

This study provides empirical evidence about the impact of collaterals’ liquidity characteristics on the PD of personal loans in a Vietnamese bank Different from previous researches, this thesis tries to find out how PD of loans response to various liquidity levels of collaterals in cases of fully protected loans while previous studies only focus on difference in

PD between secured and unsecured loans Data of this thesis are collected from internal data

of one bank in Viet Nam and differ from other researches of which data almost were provided

by National Credit information centers of other countries or collected from questionnaire Furthermore, to my knowledge, there have been limited studies about the relationship between collateral characteristics and loan default risk has not been widely studied in Viet Nam One of the most important reasons is weak credit information infrastructure and

ineffective public records as mentioned above and therefore makes the data collection costly However, banks may have their own researches about this topic but do not publish due to information privacy

Negative responses of high liquidity level on PD are found in this paper, strongly supporting the dominance of borrower selection and risk shifting effect in this bank The higher liquidity levels of collaterals, the lower probability of default on individual loans From this result, improving the screening quality in case of lower liquid collaterals is suggested for investigated bank Moreover, status of hidden subprime loans existence is warned to policy makers Therefore, some recommendations are suggested for the State bank of Viet Nam in order to prevent severe loan loss if assets prices have devaluated

1.5 Structure of thesis

The rest of this paper is organized as follows:

Chapter 1 introduces about the background and motivation of this research Inconsistent empirical evidence about the impact of collaterals on PD is shown and the cause of this inconsistency is discussed shortly Bad debt situation in Viet Nam from 2011 until now is also presented to show the need and motivation of a research about determinants of bad debts

Chapter 2 will review the theoretical and empirical literatures about the relationship between collaterals and non-performing loans (NPL) Brief explanation of the interactions of 4 channels through which collaterals affect loan default is discussed in this part

Research methodology and data will be presented clearly in chapter 3 Brief explanation about the meaning and suitability of logit model for this research and data collection is discussed This chapter also shows the limitation of data source

Chapter 4 discusses empirical results and chapter 5 summarizes the research’s main findings from which policy implication is suggested

CHAPTER 2

T H E O R I C A L F R A M E W O R K A N D

L I T E R A T U R E R E V I E W

2.1 Theoretical review of relationship between collaterals and loan risks

There are two strands of theories that explain different effects of collateral requirement on loan risk, ex ant and ex post theory The ex-ant theory interprets the borrower selection effect and the ex-post one explains the lender selection effect, risk shifting effect and loss mitigation effect (Berger et al 2011)

Figure 3: The transmission channels of collaterals on loan risk

The ex ant theory explains the negative relationship between collateral requirement and NPL due to the borrower selection effect In this channel, higher quality borrowers tend to pledge more liquid collateral to take lower interest rates on loans thanking to lower screening cost In

Loan risk (probability of default)

Collateral

pledge

Borrower selection effect (-)

Lender selection effect (+)

Risk shifting effect (-)

Loss mitigation effect (+)

this case, decision of banks in approving loans is based on signaling which means that banks observe behavior of borrowers between secured and unsecured loans in order to classify the quality of borrowers (Japhet and Memba, 2015) And according to Berger et al (2009) in Japhet and Memba (2015), this ex ant theory is only applicable in cases of short relationship between borrowers and lenders which implies a high level of asymmetry information between two parties

The choice of pledging collateral of borrowers is based on the expectation of avoiding screening cost of banks and therefore lower interest rate The screening procedure of banks is costly and obviously this cost will be accounted in the interest rate Manove et al (1998) argues that in equilibrium, banks screen all projects, but fund the good ones only and charge

an interest rate which is equal to the cost of fund plus screening cost of that approved project and prorated share of the screening cost of all unapproved projects Banks tend to loose the screening process if being highly protected by collaterals Therefore, high-type applicants (applicants with good projects) tend to pledge collateral to remove banks from screening in order to get lower interest rate Hence, the borrower selection effect predicts a negative relation between collateral pledging and loan risk, which means that the more quantity of collaterals and the higher liquid collateral is associated with the lower probability of default However, the strength of this effect may declines due to the optimism of some borrowers Wishful thinking makes borrower’s perception biased according to Manove and Padilla (1999) Also De Bondt and Thaler (1995) in Manove and Padilla (1999, pp 325) reports that

“perhaps the most robust finding in the psychology of judgment is that people are overconfident” This report based on the summary from studies that made by behavioral economists, psychologists and sociologists In the context of optimistic borrowers, borrowers always exaggerate the efficiency of their projects and underestimate the likelihood of loans default Consequently, if collateral requirement eases the approval of projects and result in lower interest rate, optimists tend to pledge more collateral In this case, collateral requirement reduces the efficiency of economic and social welfare because of resources distribution into low quality projects (Manove and Padilla, 1999) Hence, a secured loan may imply a lower probability of default due to high type borrower and efficient project, except optimistic borrower

Figure 4: Screening cost prorated

In another point of view, secured loans may reflect low quality of borrowers and projects This perception is based on the lender selection effect Banks have advantage in project assessment and ability in distinguishing bad and/or optimistic borrowers and projects In order

to protect themselves from loan risk, banks require low quality borrowers/projects to pledge more collaterals Nevertheless, the act of screening of banks is unobservable Banks may decide to screen the efficiency of the projects or not, depending on the profit acquired from screening According to Manove et al (1998), in case of the benefit to the bank from screening (information of borrowers and projects obtained) is less than the screening cost banks will not screen the projects The choice of screening is explained in Manove et al (1998) as follows: Expected loss equation is given as: L = (1- PH)*(R – K)

where: S is screening cost, PH: Probability of success in case of good borrower and project, R: Original loan amount, K: Loan amount which is secured by collateral

Total screening cost

(S)

S of denied loans

S of Approved loans

Interest of approved loans

Good borrowers prevent this cost

by pledging collaterals Bank do not screen if fully

protected  do not pay this S

Bank will lose loans amount which is not protected (R-K) if projects fail with a probability of (1- PH) Therefore, banks will screen if S < L which means bank will pay screening cost S to avoid loss L which is larger than S in case of project failure Hence, if screening cost is small enough, bank will get profit from screening projects

If K is smaller than R, this implies larger loss in case of projects failure, the more incentive that banks have to screen the projects to prevent higher expected loss

If K ≥ R, there is no expected loss, banks will earn profit with certainty and therefore never screen In this case, banks screen the project just because of its credit policy that desires to fund good projects and borrowers (Manove et al 1998)

One more reason that gives more incentive for banks to approve secured loans but do not screen is loss mitigation effect As mentioned above, if loan is fully protected up to 100% by collaterals, banks can earn profit in certainty and have more incentive to approve loan without screening Manove and Padilla (1999) argue that banks are willing to fund inefficient projects

of enterprises and require more collaterals for these loans in order to protect banks from risks Moreover, collaterals may decrease the effort of banks in monitoring loans even though they can help to prevent PD (Cadot, 2011 in Japhet and Memba, 2015) and therefore increase PD.

Hence, in case of fully protected loans, quality of borrowers and projects may be recognized

or not and depends on the choice of screening of banks If banks screen due to the desire to fund high quality projects, secured loans may imply low quality borrowers and projects In contrast, if banks do not screen, a secured loan is not able to show the quality of borrowers The prediction of quality of borrower in this case is based on the borrower selection effect

PD will be lower if good borrowers pledge collaterals to get low interest rate, but PD will be higher if borrowers are optimists as mentioned above In cases that loans are not fully secured

by collaterals including unsecured loans (bank screen in these cases), due to lender selection effect, collateral requirement of banks implies low quality of borrowers and projects and therefore implies higher PD

Based on 3 channels through which collaterals affect loan risk, quality of borrowers and projects as well as PD are doubtful and unclearly predicted due to unobservable perception of

borrowers (optimist) and screening choice of banks However, there is still one more channel which supports the positive relationship between collateral and loan risk That is risk shifting effect which explains that in case of secured loans, borrowers tend to be careful in choosing projects to invest and shift their investment into high quality projects to prevent asset loss in cases of default (Berger et al, 2011) and therefore decreases the PD Furthermore, in the extent

of high level of moral hazard and asymmetry information, collaterals help banks to reduce the impact of bad actions of borrowers after loans are granted.

Besides, strength of each channel not only depends on loans which are protected by collaterals

or not, but also on economic characteristics (liquid, desirable) and types of collaterals based

on Berger et al (2011) Berger et al (2011) find that lender selection effect is critical for outside collaterals, risk shifting and loss mitigation effect are essential for liquid collaterals This finding makes sense The higher the liquidity and desirability of collateral is, the more effort that borrowers pay to prevent asset loss Besides, banks are easier in liquidating liquid assets to collect debt in case of loan default with lower cost than lower liquid assets

Consequently, prediction of PD that is based on secured status of loans belongs to the channel which has stronger effect In case PD of secured loans is higher than that of unsecured loans, secured loans may not belong to better quality borrowers and projects than unsecured loans because of more carefully screening process of banks in cases of unsecured loans to prevent loss when loans default and therefore only good borrowers are approved which implies lower

PD for unsecured loans Worse borrowers are only approved if they pledge collateral to secure for loans Lender selection effect is strong in this case In contrast, borrower selection effect

as well as risk shifting effect is weak in this case This is due to optimistic borrowers and loosing screening process of banks if loans are protected by collaterals and therefore banks approve low type borrowers and projects

For the case PD of secured loans is lower than unsecured loans, borrower selection effect and risk shifting effect is stronger than lender selection effect Borrowers who accept to pledge their assets to secure for loans to get lower cost are high type applicants and the effect of optimist is small here In case of unsecured loans, banks appraise quality of borrowers’ and projects’ inefficiency and therefore approve many bad applicants

In the comparison between fully protected loans, if PD of loans which are secured by higher liquid assets is lower than ones that are protected by lower liquid assets, Borrower selection effect and risk shifting effect is stronger than lender selection effect in case of high liquid collaterals For loans that are secured by high liquid collaterals, borrowers screen their projects carefully and put a lot of effort to make project to be profitable in order to prevent asset loss Hence, even though banks tend to loose their screening process because of high protection level of loans, the PD is still lower due to Risk shifting effect and Borrower selection effect In this case, if banks choose to screen loans applicants carefully so that only good borrowers and projects are approved, banks still fail in evaluating exactly the quality of projects due to moral hazard (NPL still exsits when banks screen projects) and therefore high liquid collaterals help banks to prevent ex post non performance loans more efficiency than low ones

2.2 Empirical review of relationship between collaterals and loan risk

According to Elsas and Krahnen (2000), there are different impacts of collateral on default risk This impact is found to be positive in some papers but negatives in others and even does not exist (Berger et al 2011)

Berger et al (2011) examines the difference in PD between secured and unsecured loans, using a sample of 25,391 firm loans in Bolivia and find that collaterals increase PD and reduce loan risk premium Inderst and Mueller (2007) in Japhet and Memba (2015) supports this finding by predicting that riskier borrowers tend to pledge collaterals and therefore have higher ex post loan risk Also, Jiménez and Saurina (2004) uses binomial logit model to test this relation for 3,000,000 company loans in Spain and conclude that secured loans increase the PD when compared with unsecured lending Another finding from this paper is the significant relation between the proportion of loans amount that is secured by collateral and

PD, particularly 100% secured loans have lower PD than those that are secured from over 50% but less than 100% loans amount In contrast, Niinimaki (2010) find that “costly collateral3 turns out to have positive incentive effect whether its value is stochastic or non-stochastic” and they explain that borrowers exercise all their effort in order to prevent asset

3 Costly collateral is cost to borrower if investment project fail

loss This paper also concludes that in case of large variation of collateral value, stochastic collaterals have smaller incentive effects than non-stochastic ones This is explained as if expected value of collateral is significantly higher than present value, borrowers tend to pay more effort to repay loans because they will loss more in this case In contrast, if expected value of collateral is stochastic or unpredictable, borrower pay less effort than previous case Variation level of collateral value is considered as part of liquidity level of collaterals

No relationship between collateral and loan risk is found in Elsas and Krahnen (2000) This paper concludes there is no correlation between borrower quality and incidence of collaterals

in Germany using the data collected from top five German banks and concludes that collaterals are required primarily by lenders that have previous relation with borrowers in order to lock borrowers in this relation and therefore increase the power of bargaining of banks in the future renegotiation

The inconsistent results found in these papers may be due to the dissimilar economic conditions of different countries and samples choices Berger et al (2011) argues that different samples that have dissimilar types of collaterals may lead to different results about this relationship Furthermore, these researches only investigate the difference in probability of default between secured and unsecured loans while limited studies examine different levels of collateral liquidity Berger et al (2011) find that in cases of secured loans, the liquidity level and desirability of assets which are used as collaterals for loans are found to have impact on

PD The authors define liquidity of asset based on “the ease, cost, and time with which the secured assets can be converted into cash at fair market value in the event of default” and therefore a liquid asset is quickly transferred into cash without a significant discount on its value This research finds negatively and statistically significant relationship between liquid collateral (Deposits, Bank guarantees, Securities) and loan risk This result implies the dominance of risk shifting effect, borrower selection effect over lender selection effect Japhet and Memba (2015) collect information from banks’ loans processing and compliance departments and from entrepreneurs of SMEs of 14 commercial banks in Kisii County - Kenya through questionnaire to find out which type of collaterals are preferred by banks in order to reduce loan risk Using descriptive statistics, the paper find that motor vehicles are preferred over lands and buildings by most of banks as collaterals for loans to reduce PD The

explanation of the cause of this finding is the complicated legal procedure in pledging and liquidating lands and buildings Moreover, the process of liquidating lands and buildings costs more than that of motor vehicles One more finding that support the Risk shifting effect and loss mitigation effect in this research is associated with a very low default rate of loans which are secured by motor vehicles Hence, these findings lead to a conclusion that the more liquid and desirable collaterals are, the lower the probability of default is However, these findings may be biased due to the desirable of interviewees who are employees of banks for more simple works when handling with loans secured by motor cycles compare to loans secured by lands and buildings

Furthermore, as aforementioned, collaterals in developing countries play a more critical role than developed countries in credit market Typically in Viet Nam, almost loans are secured loans due to the critical moral hazard and asymmetry information Ogeisia et al (2014) argue that lending in low income countries is notoriously risky because of information asymmetries problem which are high in developing countries United Nations Conference on Trade and Development - UNCTAD (2005, page 119) state that the explanations for high level of information asymmetry are weak credit information infrastructure, ineffective public records, lack of credit management skills and underdeveloped financial intermediation, and this asymmetry problem was made worse by generally restrictive and complicated regulatory environment and a large informal cash-based economy Hence, the role of collateral in credit market in developing countries is very important in controlling default risk Another explanation for the higher influence level of collaterals in developing countries may be the essential meaning of assets to inhabitants The reason of this is the big gap between collaterals value (such as houses, cars) and average income per capita in low-income countries

In conclusion, impact of collaterals characteristics on PD is inconsistent between previous studies due to different samples and researched countries Almost researches study difference

in PD which is affected by collaterals requirement between secured and unsecured loans Few papers investigate this difference between loans that are fully secured by collaterals which differ in liquidity levels Comparison of PD which is affected by collaterals between secured and unsecured loans is significantly different compared with that of fully protected loans In Viet Nam, fully protected bad debts are still increasing sharply in the last period even though

collaterals requirement is the most important condition in loans approval The question about the role of collaterals in controlling for these bad debts which are fully secured has not yet been widely investigated Almost loans in Viet Nam are fully secured by many types of collaterals which are different each other in liquidity levels Moreover, as mentioned above, home, residential lands, cars are main assets to be requested by banks for collaterals due to their easy transfer into cash with low cost This argument is supported by Japhet and Memba (2015) and support for different decision of bank in loans approval in cases of different liquidity levels of collaterals Banks will request higher liquid collaterals in cases of low quality borrowers due to lender selection effect and loss mitigation effect and therefore these loans have higher PD However, valuable assets are essential meaning to inhabitants in low income country as Viet Nam due to big gap between collaterals value and average income per capita Hence, in low income countries, people tend to spend many efforts to prevent their assets loss, especially in case of highly desirable collaterals Besides, people who afford to own an asset usually be considered as having income generation ability and therefore have capacity in loans repayment Borrower selection effect and risk shifting effect are dominant in this case and imply lower PD of loans protected by higher liquidity level collaterals From these arguments, PD in cases of higher liquidity collaterals is doubtful and base on the dominance of which effect channel The dominance of which effect channel in turn bases on the screening choice as well as screening quality of bank As mention above, banks do not have incentive to screen in case of fully protected and therefore PD of loans are depend mainly on borrower selection and risk shifting effect which imply lower PD for higher liquidity collaterals From these above arguments and from the theoretical and empirical review about the relationship between collaterals characteristics and PD of loans, the following hypotheses are developed:

H1: Secured loans with higher liquidity level of collaterals have lower PD than those with lower liquidity level of collaterals

H2: Borrower selection effect and risk shifting effect are transmission mechanisms through which collaterals affect loan risk default

Lender selection effect

Risk shifting effect

Loss mitigation effect

Dependent variable is dummy variable yit which take value of 1 if loan i generated in year t is default and 0 if not A loan is default if it is downgraded to the rank 3-5 or is overdue more than 90 days during loan time according to 493/2005/QĐ-NHNN, and 18/2007/QĐ-NHNN regulations that were promulgated by the State Bank of Viet Nam These documents regulated the debt classification and provision that applied for financial institutions in Viet Nam This NPL criteria is used in Jiménez and Saurina (2004) and consistent with the regulations of the State bank of Viet Nam

Predictors of the model include variables that represent liquidity ranks of collaterals, interest rates, loans times, protected levels of loans, loans sizes and ownership status of collaterals Jiménez and Saurina (2004) find negative relationship between loans duration, size of loans and probability of default which means the longer the loan duration is, the lower the PD of that loan is and the same as in case of size of loans The model is controlled for professions of borrowers, timely factors and regions that loans are generated in order to cover the difference

in lending policies between time and other temporal differences Lending standards differ from time As mentioned above, Berger et al (2011) find that lender selection effect is critical for outside collaterals which are owned by the third parties such as house of owners of the company If owners of company lose their own private assets in case of loan default, they tend

to consider investment projects more carefully This may suggests a difference in PD between

collaterals which are owned by borrowers and those are owned by the third parties who almost are members of borrower’s family

Variable that presents the liquidity ranks of collaterals is ordinal variable This variable takes value from 1 to 6 that represent for 6 liquidity levels of collaterals from low to high Liquidity levels of collaterals are collected from internal loans accounts database of investigated bank Loan sizes are divided into 3 levels from low to high: equal or less than 500 millions dong, more than 500 millions dong to 1 billion dong, more than 1 billion dong 2 dummy variables are added to the model to control for 3 levels of loans sizes Interest rates are changed every 3 months (this term is defined in loans contracts), therefore average interest rate of loans are calculated in order to have the relative consistent number in comparing with the fact Loans times are measured in months

Because linear model is not adequate for dummy dependent variable which presents default status of loan due to constant variance assumption, logit model is applied to predict the probability of default

Specific model

An individual borrower will default if the utility that borrower expects to obtain when default

is greater than that he/she would obtain if not default Call y*it is the difference in the utility mentioned above and y*it is non-observable As argued above, a borrower will default if y*it

>0 which means borrower get higher utility in case of default

y*it takes the form: y* it = α + x’ it β + z’ t γ + w’ i Ω + hcm + ε it

x’it: Liquidity levels of collaterals This is the main explanation variable

z’t: set of other explanatory variables such as: interest rates, loans times, protected levels of loans, loans sizes, ownership

dummy variable for the year in that each loan is created

w’ i : control variables for time, profession and regions factors

Call yit is decision in default of borrower i who get a loan in year t yit is an endogenous variable and dichotomous, where yit = 1 if the loan is delinquent (y*it>0) and 0 otherwise (y*it<0)

According to the aforementioned relationship, obtain:

𝑙𝑜𝑔𝑖𝑡 = log Prob default

1−Prob default = log odds = F (α + x’it β + z’t γ + w’i Ω + hcm + ε it)

The odds ratios are obtained from the logit estimation by comparing two odds Probability of default is calculated as 𝑒

𝑙𝑜𝑔𝑖𝑡 1+𝑒 𝑙𝑜𝑔𝑖𝑡

Outcome of the model now is the prediction of the changes in PD compared to Probability of not default of a loan when its predictors vary, especially liquidity levels of collaterals Marginal effect is calculated as MG = Pr(y = 1|x, xk = 1) – Pr(y=1|x, xk = 0)

3.2 Data

Data for this thesis is collected from the internal loan accounts of a medium size bank in Viet Nam Differentiating from many previous papers which focus on enterprises loans, this paper aims at individual borrowers Data is collected in only from one bank due to the lack and difficulty in collecting data about loans accounts in all banks in Viet Nam

The filtering process for data collection from individual loan accounts is based on the following criteria: (i) loans were first generated in the years 2010, 2011, 2012 and matured before the day 03/31/2016; (ii) loans have short and medium terms (more than 12 months to

60 months) but duration of loans time is at least one year, (iii) loan amount is at least 100 million VND, (iv) each loan account is secured by only one asset and this asset is used as a collateral for that loan account at each point of time, (v) locations where loans are applied and approved are Ho Chi Minh city and Ha Noi

The data sample includes 2,295 observations which satisfy the above filtered criteria In this sample, majority of professions of borrowers are workers, employees of enterprises, businessman in service sector and traders, therefore borrowers’ profession is grouped into nonbusiness, service and trade Loans of borrowers whose professions are belong to group 2 (service) and 3 (trade) will compare to those belong to group 1 in the impact on PD

Collaterals are classified based on the internal policy’s ranking of bank about collaterals (Regulation number 1782/2011/QĐ-TGĐ and 591/2012/QĐ-TGĐ) Accordingly, the collaterals are ranked into 11 levels based on the liquidity of collaterals and discounted risk level of collateral value Highest level is 11 that is the most liquid and desirable collateral Data collected do not include loans that are secured by deposits issued by investigated bank and therefore rank 11 collaterals do not exist in sample Collaterals of which levels are from 2

to 10 are collected However, level 2, 3, 4 will be combined in 1 group because of few observations in level 2 and 4 Level 9, 10 are also combines as the same reason Consequently, there are 6 levels of liquidity of collaterals in sample More detail about collaterals classification is below:

There are 4 ranks corresponding to 11 levels of collateral liquidities according to above regulations of investigated bank:

Rank A (A1, A2, A3): most easily transfer into cash and lowest discounted risk

Rank B (B1, B2): easily transfer into cash and low discounted risk

Rank C (C1, C2): easily transfer into cash but high discounted risk or low discounted risk but hard to transfer into cash

Rank D (D1, D2) and rank E (E1, E2): Higher risk in liquidity and discounted than C These two types of collaterals appear rarely in individual loans of investigated bank

Detail assets correspond with each rank presented in Appendix 2

Table 1: Summary of variables

Explanation variables

This is ordinal variable that takes value from 1 to 6, corresponding to increasingly liquidity levels of collaterals

Owner of collateral is the borrower or family members

of the borrower This is a dummy variable which is equal 1 if the owner is not borrower and 0 otherwise

3 Amount1 Loan amount at loan origination is less than 500

millions dong

4 Amount2 Loan amount at loan origination is more than 500

millions dong to 1 billion dong

5 Amount3 Loan amount at loan origination is more than 500

8 Middle-term Dummy variable which takes value of 1 if loan time is

more than 12 months, 0 if not

loans amount on collaterals value varies from equal and higher than 50% to 75% and 0 otherwise

11 Protect100

A dummy variable taking the value of 1 if ratio of loans amount on collaterals value varies from equal and higher than 70% and 0 otherwise

Control variables

Three dummy variables are constructed to represent three different groups of lenders’ professions: Trade-profession, Service-profession and Nonbusiness4

2 hcm = 1 if loans are generated in HCM city; = 0 if in Ha

Noi

3 year Equal 1 if loans are first generated in 2010; equal 2 if

that in 2011; equal 3 if that in 2012

Firstly, the model is the prediction about PD based of all explanatory variables The result shows the insignificant impact of interest and loan time on PD Moreover, interest and loan time variables have strong correlations with each other and with other variables (high VIF indicator for interest factor and loan time) and therefore are omitted from the model Interaction variables of HCM and two profession variables Trade and Service which represents loans that are generated in Ho Chi Minh city and borrowers are participating in Trade, Service industry respectively are added into the model to guarantee the fit, suitable and meaningful of the model Linktest is applied to test for the suitable specification of dependent variables and prove that the model is fit and suitable due to the insignificant meaning of the prediction square (hatsq has 12.8% p.value) Wald test also implemented to assure that there are differences between collateral rankings, loans amount groups and dissimilar protected rate Finally, the liquidrank variable is substituted by 5 dummy variables which represent for

5 ranks of collateral liquidities and predict PD again for robustness

4 Nonbusiness: borrowers who classified in this group are workers, employees in government agencies and

companies This variable is considered as base variable and therefore be dropped out from regression model in order

to prevent multicollinearity problem

CHAPTER 4

E M P I R I C A L R E S U L T S &

D I S C U S S I O N

4.1 Descriptive Statistics and Pre-estimation tests

Table 2: Summary of loans characteristics

to 26% This number is suitable with the medium interest rate in credit market in the period 2010-2012 Loan sizes which are less than 1,000 millions dong dominate in sample when mean value of loans amount is 678.7 millions dong Majority of loans in sample are short term loans (Mean of loans duration is 25.39)

Table 4, 5, 6, 7 will present statistic of NPL corresponding with explanation variables

Most collaterals in sample are belong to rank 5 (or liquidity level 5) with 1,166 observations, equivalent 50.81% of the total number of loans accounts in sample Types of collateral in sample are home and residential lands, agricultural lands, apartments, cars These are regular assets that usually pledged by individuals As can be seen from Table 4, default rate is high

for liquidity level 1, 2, 3 with the value of 0.35, 0.39, 0.33 respectively This rate is lower for level 4, 5, 6 (0.18, 0.12, 0.14 respectively)

Table 3: Summary of default loans according to liquidity levels of collaterals

of loans) Default rate

Table 4: Summary of default loans according to varied amounts of loans

Table 5: Summary of loans default according to rate of protection

Total Default rate

Table 6: Summary of loans default according to loan time

on PD which is different from many previous papers such as Louzis, Vouldis and Metaxas, (2011) and Farhan et al (2012) Furthermore, interest factor suffers from serious multicollinearity problem and therefore be dropped out of model

Results of these two regressions after dropping interest rate variable out are presented in table

7

Table 7: Estimation results of Logit model

Loan time (months)

Dummy loan time for middle term

*** p<0.01, ** p<0.05, * p<0.1 Although almost main factors are significant at 99% level of confidence, multicollinearity problem still exists due to loan time factor Besides, different from Jiménez and Saurina (2004) which find negative relationship between loan time and PD, this paper finds an insignificant relationship between them Hence, loan time factor in turn is excluded from model (hcm also is dropped out due to high VIF indicator) To ensure the suitability and meaningful of the model, 2 interaction variables between hcm and professions are added into model The last model satisfies the specification error test (linktest) Almost factors are significant at 99% confident level Final results are shown in below table

Table 8: Estimation results of Logit model (exclude interest and loan time factors)

These above findings support the dominance of borrower selection effect and risk shifting effect over lender selection effect in examined sample In these individual loans cases, loans that are secured by high liquidity level collaterals may imply good projects and borrowers and therefore a lower ex post default risk These good borrowers tend to pledge collaterals to be easily approved for their loans applicants with lower interest Also, they choose investment projects more seriously in order to prevent asset loss in case of default Moreover, these findings also support for the low quality of screening process of banks For the loans that secured by lower liquidity level collateral, higher PD implies bad screening of banks of these loans and therefore lead to the context in which loans approved for low quality borrowers and projects are protected by low liquid collateral As mentioned above, banks have incentive to loose their screening process due to all loans are fully protected These findings also show that there is not much difference in screening quality between loans that are secured by different liquidity levels of collaterals Hence, the PD of loans depends primarily on borrower selection effect and risk shifting effect and in turn these effects depend on the liquidity of collaterals These finding answer for the 2 research questions that higher liquidity level of collaterals lead to lower PD of loans through borrower selection and risk mitigation effect Furthermore, these results are consistent with findings in Berger et al 2011 for the case of Bolivia

Regarding to the other explanation variables, loan maturity or loan time is found to have insignificant impact on PD as mentioned above This finding may imply indifferent screening quality of banks between middle term and short term loans, ceteris paribus Another implication of this finding is that even though the liquidity level of collateral of a loan is low, long term of loan still not be screened more carefully by this bank In contrast, loan amount has significant impact on PD Compared to loans which amount is 500 millions dong or less, loans in group 2 (amount from 500 millions dong to 1 billion dong) and group 3 (more than 1 billion dong) have higher probability of default, other factors unchanged The odd ratios of amount2 and amount3 variables are 1.67 and 3.63 respectively at 99% confident level Again, this result continuously supports indifferent and bad quality of screening between various amounts of loans The investigating bank did not examine more careful the quality of borrowers and projects when loans amounts of these applicants are high and therefore lead to higher PD despite of the liquidity level of collaterals they pledged This finding is still contrary to Jiménez and Saurina (2004) which find significant and decreasing relationship between loan size and PD and explain that banks examine more carefully in cases of bigger loans size However, because this paper researches only for full protected loans, these above findings support for the screening choice theory of financial institutions in Manove, Padilla, Pagano (1998) which argues that banks do not have incentive to screen projects if loans are fully protected by collaterals Hence, in the researching bank, probability of default may be mainly depends on borrower selection effect and risk shifting effect This bank make loans approval decision through observing the type collaterals that borrowers use to pledged If borrowers pledge high level of liquidity collaterals which predict good quality of borrowers and projects and therefore lower the PD, bank tends to easily approve these loans One more finding that support for risk shifting effect is the significant impact of protection rate of loans

on PD Loans that have the ratio of loans amounts to collaterals value from 50% to 75% have the odd ratio 2.3 which means the probability of default of these loans is 2.3 times the default probability of loans of which the ratio of loans amount to collaterals value is lower or equal 50% And in cases of this ratio higher than 75%, odd ratio is 7.8 This finding is reinforced by Jiménez and Saurina (2004) concluding that fully-secured loans have lower PD than loans that are secured from 50% to less than 100% Hence, the higher the ratio of collaterals value to loans amount, the lower the PD is This result also supports the borrower selection and risk