Determinants of non performing loans in vietnamese banking system

The testing results are in accordance with several papers which indicated the negative relation with economic growth and positive correlation with lending interest rate and government de

UNIVERSITY OF ECONOMICS ERASMUS UNVERSITY ROTTERDAM

HO CHI MINH CITY INSTITUTE OF SOCIAL STUDIES

VIETNAM THE NETHERLANDS

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

DETERMINANTS OF NON-PERFORMING LOANS

IN VIETNAMESE BANKING SYSTEM

BY

NGUYEN THI HONG THUONG

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

HO CHI MINH CITY, DECEMBER 2017

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

DETERMINANTS OF NON-PERFORMING LOANS

IN VIETNAMESE BANKING SYSTEM

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

MASTER OF ARTS IN DEVELOPMENT ECONOMICS

By

NGUYEN THI HONG THUONG

Academic Supervisor:

A/PROF NGUYEN VAN NGAI

HO CHI MINH CITY, DECEMBER 2017

DECLARATION

I declare that the wholly and mainly contents and the work presented in this thesis (Determinants of Non-performing loans in Vietnamese Banking System) are conducted by myself The work is based on my academic knowledge as well as my review of others’ works and resources, which is always given and mentioned in the reference lists This thesis has not been previously submitted for any degree or presented to any academic board and has not been published to any sources I am hereby responsible for this thesis, the work and the results of my own original research

NGUYEN THI HONG THUONG

ACKNOWLEDGEMENT

Here I would like to show my sincere expression of gratitude to thank my supervisor, Ass Professor Nguyen Van Ngai for his dedicated guideline, understanding and supports during the making of this thesis His precious academic knowledge and ideas has motivated me for completing this thesis

Besides, I would like to express my appreciation to the lecturers and staff of the Vietnam – Netherlands Program at University of Economics Ho Chi Minh city for their willingness and priceless time to assist and give me opportunity for this thesis completion

Next, I would like to thank all of my classmates for their encouragement and their hard work, which become a good example for me to do the thesis I wish all of us will graduate at the same date

Lastly, I would like to express my gratitude to my families, my beloved group for their unlimited supports and encouragement They are the motivation for me to finish this course research project

FE: Fixed-effect estimator

GDP: Gross domestic product

NPLs: Non-performing loans

OLS: Ordinary Least Square

RE: Random-effect estimator

SBV: State Bank of Vietnam

S.GMM: the system generalized method of the moments estimator

Credit risk is one of the elements impact on the health of banking systems and performance of economic activities Non-performing loans is the general factor presents for this bank’s credit risk There are previous researches indicate the close relations between bad debts and factors from macroeconomic environment and bank specifications This is the motivations for this paper to examine both macro and micro variables of 30 Vietnamese banks from 2006 to 2016 This dynamic panel data is estimated by the System Generalized Method of Moments The regression results support the strong evidence for the impact of macro indicators on problem loans The testing results are in accordance with several papers which indicated the negative relation with economic growth and positive correlation with lending interest rate and government debts of problem loans However, due to the type of labor force, the increase of unemployment rate will lead to the increase in bad loans in Vietnam In addition, with bank-specific factors, tests of skimping hypothesis, diversification (with proxy is banks’ size) hypothesis and procyclical credit policy hypothesis have the statistical significance in Vietnam

DECLARATION i

ACKNOWLEDGEMENT ii

ABBREVIATION iii

ABSTRACT iii

CONTENTS v

APPENDIX 1

LIST OF TABLES 2

CHAPTER 1: INTRODUCTION 3

1.1 Problem statements: 3

1.3 Research objectives: 4

1.4 Research questions: 4

1.5 Structure of Research: 4

CHAPTER 2: LITERATURE REVIEWS 6

2.1 Macro-economic factors: 6

2.1.1 Theories: 6

2.1.2 Empirical review: 9

2.2 Bank-specific factors: Error! Bookmark not defined 2.2.1 Hypotheses: Error! Bookmark not defined 2.2.2 Empirical review: 14

CHAPTER 3: MODEL SPECIFICATION AND DATABASE 16

3.1 Model specification: 16

3.1.1 Econometric models: 16

3.1.2 Variable explanation: 21

3.2 Data: 25

CHAPTER 4: RESULTS AND DISCUSSIONS 26

4.1 Summary statistics: 26

4.2 Empirical results: 28

CHAPTER 5: CONCLUSIONS AND RECOMMENDATION 39

5.1 Conclusion:.……… 39

5.2 Recommendations:……… 40

5.3 Limitations: ……….41

REFERENCES 42

APPENDIX 48

APPENDIX

Appendix 1: Correlation of variables

Appendix 2: Addition estimation test with 2 lag of variables

Appendix 3: The estimated results for the regression models with separate hypotheses using system generalized method of the moments

LIST OF TABLES

Table 1: Summary statistics

Table 2: Results with Pooled OLS, FE, RE and SGMM estimations

Table 3: Estimation results of one lag variables

Table A1: Estimation without lagged variables

Table A2: Estimation with lagged variables

CHAPTER 1: INTRODUCTION 1.1 Problem statement

Both developed and emerging countries recognize the important of financial institutions Nkusu (2011) imply that the health of financial system and economy is the two-way impact It means that the performance of financiers could be improved by the economic growth On the contract, if bank crisis happens, the economy could be downturned Non-performing loans are considered as the general measure for riskiness of the banks, as well as applied to predict the bank crises Rajaraman and Visishtha (2002) indicated the investigation the causality of bad debts is important to control this risk Adebola, Wan Yusoff and Dahalan (2011) identify that one of the causalities of economic crisis in 2008, which affects not only on the U.S economy but also many countries around the world, is the problem loans Several loans in this period were issued for the segments in under standard conditions Therefore, when the economy goes down, most of them turn out bad debts The health of banking system become worse and worse after that, leads to the negative impact on economy (Nkusu, 2011)

Credit risk is one of the factors to evaluate the health of banking system This factors

is defined as the problem loans of banks The non-performing loan ratio of Vietnamese banking system has a significant increase from 2009 and got a peak at 2012, at 3.44% Due to the tighten monetary and lending policies of State Bank Vietnam as well as the development policies of Government, this ratio has a little decrease after that

The bad debts in banking system get the obstacles for economic growth, as well as financial system development recent years First of all, it is difficult for economic segments to approach the credit capital In the controlled period (from the end of 2010), the increase of credit was limited (approximate 15% per year) and the lending interest was high (in the range from 17% - 22% per year) In addition, in this stage, there are several M&A as well as restructured banks lead to the more tightened policies to control the stability of financial system The stuck capital flows impacted negatively on economic activities The firms were in short of capital to manufacture and investment in

order to expand production The consumption was decrease, effect on firm’s revenue and profit As a circle, the total output reduce, the economy went down Until the problem loans can resolve, the economy has to allocate the scared resources in order to support banking activities and maintain the stable of banking system This is the big obstacle which will pull down the overall economy (Nguyen Xuan Thanh, 2017)

1.4 Data and econometric model

A cross-sectional dataset is collected to support the objectives of this paper This data includes macro-economic factors, such as: economic growth, unemployment rate and lending interest rate The data for bank specification will be collected from annually audited financial statements of 30 banks from 2006 to 2016 and calculated based on the financial indexes Generalized Method of Moments (GMM) is suitable for estimating influence on banks’ problem loans of these variables with the different lagged orders

1.5 Structure of thesis

This study is organized in five chapters The first chapter is the problem statement, research question and objectives The second chapter will be review the literature,

includes theories and previous researches in order to identify the factors impact on NPLs Data collection and model specification for the study will be described in the third section Next chapter will present and interpret the results of the econometric analysis with respect to the research’s theoretical and empirical analyses, which are linked to the hypotheses of the research paper The results will show the relationship of the economic factors and the NPLs ratio of banks Finally, the conclusions could be presented in last chapter

CHAPTER 2: LITERATURE REVIEW

Credit risk is the risk in credit activity of bank when the borrowers can not complete obligations of their liabilities When this risk occurs, the banks are affected negatively The total assets, profit and capital will decrease due to the increase of loan loss provision amount The consequence is the negative effect on the economic activities due to not only the increase of banks’ exposure to economic crisis but also the restriction off the credit activities Therefore, an analysis for credit risk is necessary to maintain the stability of financial system and have the early warning of possible crisis All of them are worked for the final target: the growth of economy The factors, which impact on credit risk, are divided into two groups: systematic and unsystematic credit risk (Castro, 2013) Macro-economic factors are considers as the factors influencing the systematic credit risk On the contrast, the bank specifications are grouped as unsystematic credit risk, include financial indexes and the quality of credit management

2.1 Macro-economic factors

2.1.1 Theories:

The theoretical models of business cycle, which indicates the important role of financiers, offer the good baseline for NPL models Williamson (1987) highlights the counter-cyclicality of business failures and credit risk After that, the researches of Bernanke, Gertler and Gilchrist in 1980s and 1990s mention about the financial accelerator framework The theory of financial accelerator states that the worsening financial market conditions can amplify the negative shock to economy More broadly, the downturn period of finance and macro-economy is propagated by the disadvantage conditions in the real economy and financial markets Bernanke, Gertler and Gilchrist (1996) and Kiyotaki and Moore (1997) use the framework about “principle-agent” view

of credit market in order to rationalize the financial accelerator theoretically Their method becomes the important theoretical framework for the macro-financial linkages when modeling the interaction between NPL and macro-economy

GDP growth: When economic growth is stable or increases (i.e expansion stage

of economy), the payment of borrowers is easy to complete, and the bank credit usually meets the demand and increases over time The doubtful loans are not the most serious problem of bank’s managers On the other hand, when the economy has to face with the obstacles for growth, even downturns, the reduction of cash inflows is the trend of all segments At this time, the debt payment of firms as well as individuals becomes difficult It leads to the increase in non-performing loans in the banking system Because banks’ capitals are stuck in the recession, the capital for the economy, which is the most important for all activities, is in the shortage The consequence is the stagnation of all business, and the economy is still deeper in the crisis This is the causality of banks’ bad debts and economic growth There is a negative relation between NPL and GDP growth

The interest rate: the higher interest rate is argued to be relevant with the debt

burden due to the higher of financial obligations The asymmetric information theory can explain for this argument According to this theory, when the interest rate increases, the debtors have to face an adverse selection and the loans can be their bad choice in this scenario (Bohachova, 2008)

To have enough income to cover the debts, the borrowers have a tendency to invest in riskier projects instead of safe projects with lower return Furthermore, banks will grow their income from credit activity due to new issued loans In addition, with outstanding loans, the banks can have more returns with the floating lending interest rate, which adjusts the increase of debt’s liabilities But banks have the role as financial intermediations, which lend to a large number of borrowers as well as borrow from a large number of depositors In some countries, despite of the high cost for fund and high risk behaviors as their culture, interest rate will be liberalized It means that high-risk creditors will be charged at higher rate in order to mitigate risks The consequence is the increase of overall risk exposure (Fofack, 2005)

At the recession stage of business cycle, the banks have to pay more interest for depositors than the returns received from borrowers This leads to the profit reduction,

even the losses Because total assets of banks include long-term fixed interest rate loans, the return is not quick enough for banks to handle their liabilities The temporary solution

is the rise of short-term lending interest rate to pay their liabilities (Mishkin, 1996) Furthermore, the increase of debt payment for borrowers will lead to the risk for banks’ loan portfolios as their ability is not guaranteed Therefore, this risk will be compensated

by the higher net interest margin (Ahmad and Ariff, 2007)

The unemployment rate: economic cycle stages have the closed correlation with

the unemployment rate So this factor is defined a determinant impacts on the credit risk According this view, the unemployment rate directly affects the income of households In addition, this rate increase will lead the decrease of social consumption, which will impact on the business production of corporates reflected in sale decline As the results, the repayment for obligations has the difficulty to complete, thus the credit risk is exacerbated (Castro, 2013) The model of Lawrence (1995), implies that low-income-segment could be charged higher rates than others due to the potential risk of unemployment and payment inability, based on the life-cycle consumption According to Rinaldi and Sanchis-Arellano (2006) results, current income and the unemployment rate, which are key elements of customer’s bankruptcy ability, are relevant with uncertainty regarding future income and the lending rates

Non-performing loans and banks’ losses can increase due to the diminished employment and corporate returns in the recession stage of economic cycle (Berge and Boye, 2007) Based on the expectation about the future flow of income and expenditure

of the debtors, the banks will decide the provision amount for their loans If the borrowers are unemployed, they have to suffer the higher costs for loan and other services from banks The capacity of these customers will be deteriorated due to the unexpected movements The result is the increase of credit risk

The Government debts (the sovereign debt hypothesis): Public debts create the

pressure on economic development to ensure the payment ability for principal and interest Therefore, when the ratio of public debt exceeds the acceptable threshold, it will

negatively affect the growth This is the causality of vulnerabilities, which are the baseline of financial crisis if having no punctual adjustment policies When the economy is downturn, the banks are careful to finance the loans The capital from banks

is tightened due to lending reduction It leads the reduction of production business as well

as the social consumption The firms’ revenue and households’ income are decrease So, the repayment for bank loans is therefore also affected accordingly, leading to bad debt ratio tends to increase

2.1.2 Empirical review

Several previous studies do estimation the impact of macro-economic on performing loans Shu (2002) indicates the change in macroeconomic factor can influence on the repayment ability of borrowers and banks’ loan portfolio when examining the banks in Hong Kong The finding of this study shows that in the expansion stage of the economic cycle, the banks have more chances to push lending activity, thus the risk can reduce

non-Salas and Saurina (2002) examine the problem loans in commercial and saving banks in Spain from 1985 to 1997 by using GMM dynamic panel estimations in order to estimate which determinants of NPL in Spanish banks The results are showed that problem loans in neither commercial banks nor saving banks have a negative relation with the growth of economy overtime

After that, the research of Jimenez and Saurina (2006) also investigate the loan loss of Spanish commercial banks from December 1984 to December 2002 By applying the Generalized Method of Moments (GMM) estimator for dynamic panel models, they support a significant evidence for the positive relationship between the interest and

problem loans This conclusion is also supported by the research of Cural, et al (2013)

for the Southeastern European banks The explanation for this relation is the top-up loans’ obligations for borrowers when the interest rate increases

Burger and Boye (2007) support an evidence for the positive and significant effect

on non-performing loans of unemployment rate in household and corporate segments

when investigating the Nordic banks from 1993 to 2005 In addition, the finding emphasizes that the strong effect of income on the capacity of debt-servicing and the volume of problem loans from household segment Therefore, when unemployment rate increases, the income, which is used to cover the borrower’s obligations, can reduce This leads the potential increase of bad debt form this segment At the same time, the income reduction will effect on individuals’ consumption, includes financial services The consequence is the lower domestic demand The next result is the go down of firm’s earnings and loan repayment ability Therefore, banks’ bad debts will increase due to the higher unemployment rate

Then, Jakubik (2007) analyses macroeconomic factors effect on the credit risk of Czech banks by applying the Merton’s methodology The author concludes that the decrease of real economic growth will lead the higher credit risk of banks as the negative impact on the loan portfolio of the reduction from the return of companies, wage growth and the increase of unemployment rate

After that, the research of Espinoza and Prasad (2010) estimates the effects of macroeconomic shocks on non-performing loans, by applying a VAR model for the data

of 80 banks in the Gulf Cooperative Council (GCC) in the period 1995-2008 The set of macro-economic variables includes non-oil growth, interest rates Their conclusion is the increase of NPL is affected by the higher interest rate, as well as the lower real non-oil GDP growth

Nkusu (2011) uses the single-equatio panel regressions for the sample of 26 developed countries from 1998 to 2009 His data set is the macro and financial indices, include economic growth, unemployment rate, inflation, interest rate and the price variation of housing and stock The author estimated with many methods, such as: OLS model, panel-corrected standard error (PCSE) models, lagged dependent variables, fixed effects and one-step GMM The regression results indicate that the increase of NPL is affected by the downturn of macro-economy, which is measured clearly by the lower rate

of economic growth as well as employment

Zribi and Boujelbène (2011) have the same conclusion about the inverse relation between GDP growth and bank credit risk when analyzing the bank credit risk in Tunisia from

1995 to 2008

Beside, Louzis, et al (2012) approaches NPL of each loan categories in Greek

banks by using a set of macroeconomic factors, include the real rate of GDP growth, the unemployment rate and the real interest rate The result indicates that the injured debts have relationship not only with this set of variables but also with the bank’s management qualify By using government debts factor in order to formulate the sovereign debt hypothesis, which is based on the findings of Reinhart and Rogoff (2010) and Perotti (1996), the authors support a strong evidence for this hypothesis

Reinhart and Rogoff (2010) use OLS with robust errors and fractional logit to estimate the relation of bank crisis and debts Their findings indicate that bank crises are affected by the external financial obligations In addition, banking crisis usually goes with the sovereign debts crisis

In addition, Messai and Jouini (2013) apply data of 85 banks in three countries: Italia, Greece, and Spain in the duration 2004-2008 The effects of macro determinants on loan losses are estimated by variables: real growth rate, unemployment rate, real interest rate The regression results are consistent with previous studies The conclusions indicate that the NPLs have related negatively with real GDP growth and employment rate but positively with real interest rate

Recently, Chaibi and Ftiti (2015) conclude that both French and German banks increase the problem loans when the unemployment rate rising By using the growth of GDP and unemployment rate, they find that the credit risk in French banks is more sensitive to the economic environment than in Germany

2.2 Bank-specific factors

2.2.1 Hypotheses:

Three prominent hypotheses are investigated by Berger and DeYoung (1997) when the authors take into account the relationship between non-performing loans of the bank and its cost- inefficiency

“Bad management” hypothesis: The cost efficiency of banks is expected as the

obviously significant factor impacts on non-performing loans of banks This indicator is considered as the index to appraise the quality of management They are assumed that the bad management could be caused by the poor skills in credit section, such as: scoring, loan approval, loan monitor, etc The banks have to spend more and more cost on operating but the risk management could not be controlled efficiently Therefore, the NPL ratio of banks could increase due to the cost inefficiency

“Skimping” hypothesis: according this hypothesis, there is the trade-off between

operation expense allocation and future problem loans Skimping on operation costs, which devote to underwriting and controlling loans, could have cost efficiency in short-run when lower operation costs still support the quantity of loans However, the bank could be faced to the reduction of cost efficiency when non-performing loans become higher due to its less effort to maintain the quality of loan in long run

“Moral hazard” hypothesis: one of the solutions to increase bank’s profit is

increasing their loan portfolio The bank with lower capital usually serves risky segments Their performance could be better in short-term but NPL will grow in the future

Louzis et al (2011) added three hypotheses for the impact of bank- specific factors on non-performing loans They supply more respects to investigate whether other bank characteristics (different from bank’s cost efficiency) can impact on its bad debts

“Bad management II” hypothesis: bad performance in the past could predict the

increase of future NPLs According this view, bank’s performance in the past is another proxy to the measure of management ability

“Diversification” hypothesis: this theory states that if the portfolio could be

diversified, the firm could reduce the risk and maintain the revenue Banks are also a type

of corporate Therefore, their profit could be table or increase if they could have the diversification in operation

This hypothesis could be examined by bank’s size or its multiple income sources

It is said that with the large size, the banks have many opportunities to diversify their portfolio They could not be depended on credit sector as a majority operation Therefore, they could control the problem loans but maintain the stable profit

Another proxy for diversification of banks is income sources According this view, the non-interest income of bank is higher, the diversification in operation of bank is better Credit section is not only the main income if bank So, bank could control their loan efficiently This could lead the lower NPL

“Too big to fail” hypothesis: In this view, the too big to fail banks could be

expected the protection from Government in case of its failure Therefore, these banks could have tendency to increase their leverage to gain higher profit The majority capital, which finances banks’ assets, is their liabilities such as: customer deposits, borrowed money from Government and other financiers, bonds, etc This leads the result that the bank could lend more money to maintain the higher profit The bank size is bigger, the higher its pressure of debts payment So, its standard for borrowers could be lower and riskier in order to have more income The consequence is the increase of problem loans in the future

“Procyclical credit policy” hypothesis: According this view, the policy could be

decided due to bank’s optimal profit expectation as well as other targets such as it reputation So, the banks’ profitability could be desirable in the market in short-term as the effect of managers to manipulate earnings However, this index could be considered

as a negative net present value of credit extension period As below declaration, these policies are built up for short-term target After this blooming development, the credit policies could be tightened This leads the denial with positive net present value in this

stage And the bank’s problem loans are more serious in downturn time Overall, the test

of this hypothesis could reflect the liberal magnitude of bank’s credit policy, due to comparing the performance with the increase of NPL in future growth

2.2.2 Empirical review:

Berger and Young (1997) use the Granger-causality models to examine which of four hypotheses, include bad luck, bad management, skimping and moral hazard, is in accordance with the data of U.S commercial banks in the period 1985 -1994 The estimations give a strong support for bad management hypothesis, when the result indicated that higher of cost inefficiency can lead the rising of non-performing loans In addition, the possibility of skimping hypothesis is investigated in individual banks The moral hypothesis is also supported in their research

Podpiera and Weeill (2008) examine whether bad luck or bad management impacts on the bank failures in Czech banks from 1994 to 2005 By extending the Granger-causality models, which were developed by Berger and DeYoung 91997), the authors also apply GMM dynamic panel estimations in their research Their regression results support a strong evidence for bad management hypothesis According to this view, the cost-efficiency and bank’s problem loans are the negative relation The result concludes that the banks try to improve on cost-efficiency can lead to the decrease of problem loans as well as precede bank failures

After that, Karim and Hassan (2010) using the Tobit models also support the bad management hypothesis when they research the problem loans in Singapore and Malaysian banks

Although support the hypothesis about “bad management” but the results of

Louzis, et al (2012) cannot support “moral hazard” hypothesis in Greek banking system,

due to the small number of bank

When investigating the factors impact on non-performing loans in Eurozone banking system by different GMM estimation, Makri, Tsagkanos and Bellas (2013) show the negative and significant relation between NPL ratios with banks’ performance, which

is measured by the index of return-on-assets and return-on-equity Their findings reconfirm that the deterioration of profitability index can increase the bad debts The

results are consistent with the research of Louiz, et al (2012), also support the strong

evidence for the bad management II hypothesis

Salas and Saurina (2002), Hu, et al (2004) and Rajan and Dhal (2003) have the

same empirical evidence to support the diversification hypothesis when using proxy is bank size Their results indicated that the bigger bank the more diversification

opportunities However, the study of Louzis, et al (2012) cannot find the empirical

evidence to support this hypothesis, either by proxy of bank size nor by the proxy of interest income ratio They explain that the bank size could not be present the diversification fully, or that is the counter-tendencies since the bigger banks have a higher degree of risk-taking leads to higher NPLs Furthermore, their result consist with Stiroh (2004) in rejection hypothesis when apply proxy of income This consequence could be from the “potential dark sides of diversification” It means that NPLs could increase if bank could not have either the experienced managers or comparative advantages

non-Mattana, Petroni and Rossi (2014) support “too-big-to-fail” hypothesis by

examining in European banks via ROA index However, the results either of Louzis, et

al (2012) or Boyd and Gerler (1994), Ennis and Malek (2005) cannot have an empirical

evidence to support this hypothesis

Berger and Udell (2002) cite the speech of Alan Greenspan – old chairman of Federal Reserve “the worst loans are made at the top of the business cycle” However, the

findings of Louiz, et al (2012) cannot support this hypothesis

CHAPTER 3: MODEL SPECIFICATION AND DATABASE

3.1 Model specification

3.1.1 Econometric models

3.1.1.1 Dynamic panel data estimator

In the literature review, the non-performing loans are impacted by its ratio in the last year In the previous papers, the authors research the effect of the non-performing loans one year ago on its situation at the present (i.e how could the ratio of NPL in t-1 influence on NPL in t) Therefore, this study also uses the variable about NPL ratio which has lag of first order to estimate the current problem debts ratio

Based on the literature review, non-performing loans are affected by themselves in the past, especially by the nearest values Therefore, the dependent variable with one time lag is added into the right-hand side of the model So, dynamic panel data is built up The general formula of dynamic panel data approach is

𝑌𝑖𝑡 = 𝛼1𝑌𝑖𝑡−1+ 𝛼2𝑙𝑎𝑔𝑋𝑖𝑡+ 𝛾𝑖+∪𝑖𝑡; |𝛼1| < 1, i = 1, …, N; t= 1, …, T (1)

where the subscripts i and t denote the cross sectional and time dimension of the panel

sample respectively, 𝑌𝑖𝑡 is the change in the NPLs, 𝛼2𝑙𝑎𝑔 is the lag of multiple vectors,

𝑋𝑖𝑡 is the matric of vector of independent variables other than 𝑌𝑖𝑡−1, 𝛾𝑖 are the unobserved effects of bank specific and ∪𝑖𝑡 is the error term The use of Generalized Method of Moments (GMM) created by Arellano and Bond (1991) and amended by Arellano and Bover (1995) and Blundell and Bond (1998) is applied to estimate Eq (1) The first difference transformation of Equation (1) is calculated by Equation (1) at year t minuses Equation (1) at year t-1 This formula not only is consistent with the GMM estimation of Arellano and Bond, but also eliminates the impact of bank-specific factor:

∆𝑌𝑖𝑡 = 𝛼1∆𝑌𝑖𝑡−1+ 𝛼2𝑙𝑎𝑔𝑋𝑖𝑡+ ∆ ∪𝑖𝑡 (2) where ∆ is the first difference calculation ∆𝑌𝑖𝑡−1 presents for the lag of dependent variable Due to the correlation between the lagged explained variable and error term, the estimation result could be discrepant Nevertheless, 𝑌𝑖𝑡−2 has the correlation with ∆𝑌𝑖𝑡−1but independence with ∆ ∪𝑖𝑡 for t = 3, …, T, is an instrument variable of Equation (2)

regression in order to prove that ∆ ∪𝑖𝑡 are not serially correlated The dependent variable could be lagged two or more but has to meet the moment conditions:

𝐸[𝑌𝑖𝑡−𝑠∆ ∪𝑖𝑡] = 0, with t ≥ 3 and s ≥ 2 (3) Nonetheless, the correlation of the independent variables and residual also causes the bias results To resolve the endogeneity, there is an assumption about independences between error term and all values of explanatory variables, as the equation:

𝐸[𝑋𝑖𝑡−𝑠∆ ∪𝑖𝑡] = 0, with t ≥ 3 and s is not limited (4) The two-way causality is the limitation for the strictly exogenous presumption For example, if t has smaller value than s, the value of 𝐸[𝑋𝑖𝑡−𝑠∆ ∪𝑖𝑡] is not equal 0 With a set

of fixed independent variables which is fragile exogeneity, the valid instruments are the value of 𝑋𝑖𝑡 at present and lagged time, as below function:

𝐸[𝑋𝑖𝑡−𝑠∆ ∪𝑖𝑡] = 0, with t ≥ 3 and s ≥ 2 (5) Equation (3), (4) and (5) describe the statistically independent limitations They are foundation of the one-step GMM regression, following the assumption about the independence and homoscedasticity of residuals (both cross sectional and over time), consistence of parameter estimates Arellano and Bond (1991) estimate the residuals by the two-step GMM regression The result will be a consistent variance–covariance matrix

of the moment conditions This estimator can enforce the bias in standard errors statistics) due to its dependence on the estimated residuals Bond (2002); Bond and Windmeijer (2002), Windmeijer (2005) indicated that is the reason of unreliable asymptotic statistical conclusion Arellano and Bond (1991); Blundell and Bond (1998) re-confirm this inference by the relatively small cross section dimension data samples The specification test of Sagan, which distribution is asymptotical as chi-square, will utilize to examine the variables’ overall validation, based on the assumption about valid moment conditions After that, testing the null hypothesis that the difference of error terms does not having the second order autocorrelation will give the outcome for the serially uncorrelated errors (∪𝑖𝑡)fundamental assumption If the result is rejection this assumption, it means that the error terms exist the serial correlation However, this estimation is not consistent with GMM methods

(t-3.1.1.2 Econometric model

In the baseline model, the below equation is built up from Equation (1), which is

followed as Louzis, et al (2012):

∆𝑁𝑃𝐿𝑖𝑡 = 𝛼1∆𝑁𝑃𝐿𝑖𝑡−1+ ∑1𝑗=0𝛼2𝑗∆𝐺𝐷𝑃𝑡−𝑗+ ∑1𝑗=0𝛼3𝑗∆𝑈𝑁𝑡−𝑗 + ∑1𝑗=0𝛼4𝑗∆𝐿𝐼𝑅𝑡−𝑗 +

𝛾𝑖+∪𝑖𝑡; |𝛼1| < 1, i = 1, …, 30; t= 1, …, 11 (6)

In this equation, ∆𝑁𝑃𝐿𝑖𝑡 is the difference of the prolem loans ratio, ∆𝐺𝐷𝑃𝑡, ∆𝑈𝑁𝑡and ∆𝐿𝐼𝑅𝑡 are the change of GDP growth rate, unemployment rate and the lending interest rate respectively

In order to test thee “sovereign debt hypothesis”, variable of debt will add into Equation (6) as below:

∆𝑁𝑃𝐿𝑖𝑡 = 𝛼1∆𝑁𝑃𝐿𝑖𝑡−1+ ∑2𝑗=1𝛼2𝑗∆𝐺𝐷𝑃𝑡−𝑗+ ∑2𝑗=1𝛼3𝑗∆𝑈𝑁𝑡−𝑗 + ∑2𝑗=1𝛼4𝑗∆𝐿𝐼𝑅𝑡−𝑗 +

∑2𝑗=1𝛼5𝑗𝐴𝑖𝑡−𝑗 + 𝛾𝑖+∪𝑖𝑡 (8)

In the Equation (8), 𝐴𝑖𝑡 presents for variables about bank’s specifications as we explained in the previous section The dynamic of regressors year by year will be controlled due to the lag at the forth order of 𝐴𝑖𝑡 in this regression (Berger and DeYoung, 1997) At the current level, non-performing loan rate is assumed not being impacted by

the bank-specific variables (Louzis, et al., 2012) The accounting characteristics, the

latency time of management’s decisions changes as well as data variation in balance sheets of banks are the foundation for this assumption

The coefficients will be calculated in the long term in order to estimate the the cumulative effect on bad debt ratio at the current level of each regressor, as the below:

𝛼5𝑙𝑜𝑛𝑔 𝑟𝑢𝑛 = ∑2𝑗=1𝛼5𝑗⁄(1 − 𝛼1) (9)

The variance of the coefficients in the long run is calculated as Stuart and Ord (1998):

𝑣𝑎𝑟(𝛼5𝑙𝑜𝑛𝑔 𝑟𝑢𝑛) = (∑ 𝛼4𝑗

2 𝑗=1 )2(1−𝛼 1 )2 [𝑣𝑎𝑟(∑ 𝛼5𝑗

2

∑4𝑗=1𝛼5𝑗2 − 2𝑐𝑜𝑣((∑ 𝛼5𝑗

2 𝑗=1 ),(1−𝛼 1 )) (∑4𝑗=1𝛼5𝑗)(1−𝛼 1 ) +𝑣𝑎𝑟(𝛼1 )

(1−𝛼1) 2](10) Where 𝑣𝑎𝑟(∑2𝑗=1𝛼5𝑗) = ∑2𝑗=1𝑣𝑎𝑟(𝛼5𝑗)+ 2 ∑𝑗≠1𝑐𝑜𝑣(𝛼5𝑗, 𝛼5𝑙)

The analysis of cumulative impact of these lagged regressors can be accurate and robust statistical due to the variance estimation in equation (9) In addition, reviewing the standard errors in long term can detect the multi-collinearity between the lagged variables, which could be misled in the lags of each regressors in statistical significance (Berger and Deyoung, 1997) For that reason, these hypotheses could be examined based

on the long-run coefficients The general hypothesis test is:

H0: 𝛼5𝑙𝑜𝑛𝑔 𝑟𝑢𝑛 = 0

H1: 𝛼5𝑙𝑜𝑛𝑔 𝑟𝑢𝑛 > 0; or H1: 𝛼5𝑙𝑜𝑛𝑔 𝑟𝑢𝑛 < 0

Implementing which H1 depends on which hypothesis is tested

Among these hypotheses of impact of bank’s specifications, the “too-big-to fail” hypothesis is special, due to the conditions for testing The interaction terms between the size and the leverage will be added in the regression model in order to expand understanding of the relationship between the NLP rate and leverage ratio in different sizes The corresponding specification test in econometric will be:

∆𝑁𝑃𝐿𝑖𝑡 = 𝛼1∆𝑁𝑃𝐿𝑖𝑡−1+ ∑2𝑗=1𝛼2𝑗∆𝐺𝐷𝑃𝑡−𝑗+ ∑2𝑗=1𝛼3𝑗∆𝑈𝑁𝑡−𝑗 + ∑2𝑗=1𝛼4𝑗∆𝐿𝐼𝑅𝑡−𝑗 +

𝛼5𝑆𝑖𝑧𝑒𝑖𝑡+ ∑2𝑗=1𝛼6𝑗𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒 𝑟𝑎𝑡𝑖𝑜𝑖𝑡−𝑗 + ∑2𝑗=1𝛼7𝑆𝑖𝑧𝑒𝑖𝑡× 𝐿𝑒𝑣𝑒𝑟𝑎𝑔𝑒 𝑟𝑎𝑡𝑖𝑜𝑖𝑡−𝑗 +

𝛾𝑖+∪𝑖𝑡 (11)

Louzis, et al (2012) compute the marginal effect of leverage on NPLs conditional

on the banks’ size in the long run by deriving leverage ratio (LR) in Equation (11), as below:

𝛼6𝑙𝑜𝑛𝑔−𝑟𝑢𝑛+ 𝛼7𝑙𝑜𝑛𝑔−𝑟𝑢𝑛𝑆𝑖𝑧𝑒 = ∑𝑛𝑗=1𝛼6𝑗⁄(1 − 𝛼1)+ [∑𝑛𝑗=1𝛼7𝑗⁄(1 − 𝛼1)] × 𝑆𝑖𝑧𝑒 (12)

The corresponding variance is computed as Brambor et al (2006) and Shehzad et

al (2010) as Equation (10) They also imply that simple parameters t-statistics should not

be the baseline any statistic conclusion of the multiplicative terms Following Louiz et al (2012), the long run marginal effect of this bank-specific factor is the statistic significant for this hypothesis assessment:

reality, Louiz, et al (2011) supposes micro-economic variables will be a week of

exogeneity Following Bobba and Coveiloo (2007), this assumption can solve the endogeneity between error terms and future shocks in non-performing loans, although there can have the autocorrelation between the errors in the current and past level Therefore, equation (4) and (5) are the conditions for the instruments of macro and micro regressors respectively

3.1.2 Variable explanation

3.1.2.1 Dependent variable

The dependent variable of this paper is change of non-performing loan (∆NPLit) of banks, which present for the credit risk NPLs are the loans which are overdue interest and/ or principle more than 90 days In Vietnam, State Bank of Vietnam (SBV) divides debt into 5 groups:

 Group 1: outstanding balance has overdue less than 10 days

 Group 2: outstanding balance has overdue from 10 days to 90 days

 Group 3: outstanding balance has overdue from 91 days to 180 days

 Group 4: outstanding balance has overdue from 181 days to 360 days

 Group 5: outstanding balance has overdue more than 180 days

Followed SBV regulations, bad debts are loans belong to group 3, 4 and 5 The NPL ratio is calculated by total NPLs divide total outstanding balance of the bank at the end of year This formula is ∆NPLit = 𝑁𝑃𝐿𝑖𝑡− 𝑁𝑃𝐿𝑖𝑡−1 where i and t denote the bank

and time series in panel data

The 1 lag of this variable will be applied in the model as an exogenous variable to investigate whether the current non-performing loans can be impacted by its history positively

3.1.2.2 Macro-economic variables

 GDP growth (∆𝐺𝐷𝑃):

Due to the finding of Williamson (1987), the business failures and credit risk are counter-cycle, the impact of economic growth is expected having the negative sign with the NPL ratio The common variable of the economic cycle is GDP The lower GDP growth leads the lower earnings of companies, income of household, higher level of unemployment rate, thus the quality of loan portfolio will be deteriorated

There are several researches indicate the negative relation between GDP growth and non-performing loans Jakubik (2007) explains the decrease of GDP rate will lead the higher unemployment rate and lower corporates’s earning, thus the credit risk can

increase potentially Castrol (2013) supports evidence about the increase of problem loans in the recession stage of economic cycle as the consequence of the credit expansion

in the blooming period

This variable (∆𝐺𝐷𝑃) presents the change of economic growth, will be applied in the model The formula for this factor is: 𝐺𝐷𝑃𝑡 − 𝐺𝐷𝑃𝑡−𝑗, where j denotes as the lag of

this factor and in the range from 1 to 2 This variable is applied to investigate whether the economic growth could have the negative impact on the credit risk of banks

 Lending interest rate: (∆𝐿𝐼𝑅)

The influence of interest rate on credit risk is proved by several researches, such

as: Jakubik (2007), Nkusu (2011), Louzis, et al (2012), Castro (2013) and Chaibi and

Ftiti (2015) The general explanation is the increase of interest rate leads the debt burden both individuals and corporate Therefore the credit risk is predicted to be worsen due to the weakness of repayment capacity

According the view of theories about asymmetric information, Minskin (1996) concludes the higher interest rate will lead the bad selection of borrowers when they tend

to invest the riskier projects for more earnings in order to cover the liabilities This paper applies the change of the lending interest rate (∆𝐿𝐼𝑅) to present for the interest rate factor The formula is: 𝐿𝐼𝑅𝑡 − 𝐿𝐼𝑅𝑡−𝑗 where j denotes as the lag of this factor and in the

range from 1 to 2 This variable will help to examine if the interest rate and credit risk could have the positive relation

 Unemployment rate (∆𝑈𝑁)

The economic conditions can be considered by the unemployment rate This rate impresses the loan repayment ability of borrowers, thus to predict the potential of problem loans

The previous researches identify the negative effect of unfavorable increase in the unemployment rate on the income of households To remain the capable of debt- servicing, they can reduce consumption or savings However, a decrease in the market

production cost (such as wages) would be stable or increase Therefore, firms’ capability

to repay their loans is impacted negatively

This paper applies the change of the variable (∆𝑈𝑁) to estimate their effect The formula for the factor is: 𝑈𝑁𝑡 − 𝑈𝑁𝑡−𝑗where j denotes as the lag of this factor and in the

range from 1 to 2 Therefore, the unemployment rate may increase the risk of credit defaults and it is hypothesized that unemployment rate has a positive impact on bank credit risk

 Government debts (∆𝐺𝐵)

One of independent variables is the ratio of government debt in GDP from 2006 to

2016 This factor could be used in order to estimate the impact of the Public debts on problem loans The testing result was expected to having a positive relation as sovereign debt hypothesis

3.1.2.3 Bank –specific variable:

 Return-on-equity (ROEit):

This financial index is the ratio of profit and total equity of bank i in year t The

doubtful loans and ROE of banks are expected to having the negative relation as the bad management (version II) and positive sign in procyclical credit policy hypothesis mentioned

 Solvency ratio (SRit):

This variable is applied in the model to examining the moral hazard hypothesis This variable, which reflects the strength of banks capital, is calculated by the owned capital in total assets of banks It is generally considered that higher level of capital allows bank to absorb shocks that may appear in the credit market Berger and DeYoung (1997) find evidence supporting the ‘moral hazard’ hypothesis, implying impact of bank capital to NPLs, which is a negative association The explanation for the hypothesis lies

on the role of banks’ managers, who decide to accept the high riskiness rate in their loan portfolio even though their banks’ capitals are thin On the other hand, Curak et al (2013) propose that even the bank are higher capitalized, it could encourage banks to take more

risk in lending activities resulting in loan losses because of moral hazard behavior of bank managers Thus, the negative sign in this test is expectation results

 Inefficiency ratio (IEit):

The operation efficiency of banks is calculated by formula as: 𝐼𝐸𝑖𝑡 =

𝑂𝑝𝑒𝑟𝑎𝑡𝑖𝑜𝑛 𝑒𝑥𝑝𝑒𝑛𝑠𝑒𝑠 𝑖𝑡

𝑂𝑝𝑒𝑟𝑎𝑡𝑖𝑜𝑛 𝐼𝑛𝑐𝑜𝑚𝑒𝑖𝑡 The result is expected that if the bank could not operate efficiently within year, its NPL could increase at the end of the year, as the hypothesis about bad management and skimping stated Therefore, the positive sign is expected in bad management hypothesis testing, otherwise for skimping hypothesis

 Bank’s size (Sizeit):

The diversification hypothesis is tested by variable about size of the bank (Size) as

the formula: 𝑆𝑖𝑧𝑒𝑖𝑡 = 𝑇𝑜𝑡𝑎𝑙𝐴𝑠𝑠𝑒𝑡𝑠𝑖𝑡

∑30𝑖=1𝑇𝑜𝑡𝑎𝑙 𝐴𝑠𝑠𝑒𝑡𝑠 𝑖𝑡 This hypothesis indicated that the small banks have less possibility to diversify their portfolio than big competitions Therefore, these banks could have more NPL ration than other group Expectation about the testing result

is a negative sign of coefficient

 Non-interest income ratio (NIit):

This variable, is a proxy for the bank diversification, is also described by the portion of non-interest income in total income (NI) It means that the majority in total income is from interest income, the bank has to push the lending activity to maintain the growth rate of earnings The potential problem loans will increase in the credit expansion stage So the estimation result is expected a negative relation between non-performing loans and percentage of non-interest income

 Bank’s leverage ratio (LRit)

This variable is calculated by the proportion of total assets which are financed by total liabilities The expectation is the positive sign, which means that this ratio is high will increase the non-performing loans of the banks