Determinants of non-performing loans: evidence from Southeast Asian countries

Determinants of non-performing loans: Evidence from Southeast Asian countries NGUYEN THI HONG VINH Banking University of Hochiminh City – vinhnth@buh.edu.vn NGUYEN MINH SANG Banking

Determinants of non-performing loans: Evidence from Southeast Asian countries

NGUYEN THI HONG VINH Banking University of Hochiminh City – vinhnth@buh.edu.vn

NGUYEN MINH SANG Banking University of Hochiminh City – sangnm@buh.edu.vn

Abstract

The purpose of this study is to examine the bank-specific and macroeconomic determinants

of non-performing loans using an empirical framework that incorporates the related literature and theoretical hypothesis To account for non-performing loans persistence, the paper applies the Generalized Method of Moments technique for dynamic panels which use bank-level data for Southeast Asian commercial banks over the period 2010 to 2015 The empirical results provide some evidence to affirm that both bank-level and macroeconomic factors play a role in rising the non-performing loans of Southeast Asian banks The findings indicate that the high non performing loans during these years is associated with low profitability, low credit growth, low loan to deposit, high equity and large bank size Finally, the macroeconomic determinants have the significant effect on loan quality in the anticipated ways The results also find fiscal variable has negative effect on non-performing loans and found to be significant These findings may be helpful for policy makers to design macro-prudential and fiscal policies

Keywords: non-performing loans; macroeconomic determinants; bank-specific determinants; GMM estimation

1 Introduction

Non-performing loans (NPLs) have been a limiting factor to economic stability and growth of economies It is also linked with bank failure and financial crises in both emerging markets and advanced economies In Southeast Asian area, NPLs exceeded 4.759% in 2010 and over 3% in the period 2010 to 2015 Within the region, the average NPLs ratios in the period are highest for Philippines and Thailand banks at 11.18% and

3.417% while Singapore banks have very low NPLs ratio, below 1% (Table 1) What are

the key determinants of the NPLs issue in the Southeast Asian countries? This study empirically analyses the effect of bank-specific and macroeconomic determinants of bank NPLs for this area

Table 1

The rate of non-performing loans for Southeast Asian countries, 2010-2015 (%)

Year

Indonesi

a

Cambodi

a

Philippine

s

Singapor

e Laos

Malaysi

a

Thailan

d

Vietna

m

2011 2.440 4.047 10.937 0.739

5.22

0 5.437 4.070 2.450

2012 2.177 2.946 11.428 0.739 1.540 2.526 2.992 3.393

2013 2.256 2.126 12.813 0.705 1.740 1.609 2.830 3.010

2014 2.442 2.284 14.474 1.120 1.996 1.523 2.892 2.522

2015 4.220 2.381 6.855 0.790 1.162 1.929 3.060 1.877 Averag

e 3.206 3.083 11.180 0.832 2.332 2.605 3.417 2.544 Source: Bankscope, authors’ own estimations

We contribute to existing empirical analyses in three ways First, most of the existing literature has focus on U.S or European cases (Berger and DeYoung 1997, Salas and

Sarina, 2002, Louzis et al., 2010 and Anastasiou et al., 2016) Although Southeast Asian

has become an important economic area, the Southeast Asian topic has not earned enough discussions Thus, the purpose of this paper is to examine Southeast Asian banks with the latest and a wider range of panel data that cover 204 banks from 2010 to 2015

in 8 countries Second, most studies focus mainly on the relationship between bank-specific determinants and NPLs This study discusses bank-bank-specific, macroeconomic determinants and NPLs together. Finally, dynamic panel techniques are adopted to analyze the panel data, which are designed to check the persistence of NPLs. We thus investigate the persistence of NPLs to eliminate any abnormal NPLs, and that the NPLs rates of all banks tend to converge to the same long-run average value

The rest of the paper is structured as follows Section 2 overviews previous researches

on the determinants of NPLs Section 3 provides the method that used in this research, while Section 4 describes the data that are used Empirical results are presented in Section

5 Finally, Section 6 contains concluding remarks

2 Literature review

In the literature, NPLs are affected by both bank-specific and macroeconomic determinants The bank-specific determinants are the direct result of managerial decisions, included profitability, capitalization, asset quality, and size The macroeconomic environment relates to the economic growth, inflation, unemployment, income tax, and fiscal policies

Recent studies dealing with bank-specific determinants employ variables such as profitability, capital, credit growth, and size The relation between asset quality and

profitability is one of central topics in banking studies Bad management hypothesis

proposed by Berger and DeYoung (1997) suggest that the efficient banks are better at managing their credit risk This hypothesis also argues that low cost efficiency is a signal

of poor management practices, thus implying that as a result of poor loan underwriting, monitoring and control, NPLs are likely to increase Berger và DeYoung (1997) find

empirical evidence for the bad management hypothesis, suggesting that low-efficiency

causes lead to bad debt This study examine the hypothesis of US commercial banks for the period 1985-1994 and concluded that, in general, the down efficiency led to increase problem loans in the future Podpiera and Weill (2008) test the relationship between cost efficiency and NPLs in the Czech banking sector for the period 1994-2005 Beside that, Salas and Saurina (2002) and Klein (2013) examine the relationship between the lagged

of NPLs to current NPLs These findings support the bad management hypothesis that the

rising of NPLs in the past indicated bad credit risk management of banks This causes the higher NPLs in the future

According to moral hazard hypothesis, Keeton and Morris (1987) find that low

capitalization of banks leads to an increase in NPLs by examining the US commercial banks for the period 1979-1985 In order to test this hypothesis, the research variables are ROE, bank size and risk-taking of the bank represented by the variables of ROE, total assets, gross loan on total assets The study results show that NPLs are rising for banks with relatively low equity on assets This is explained by with thinly capitalized banks, their managers increase the riskiness of their loan portfolio in the moral hazard incentives The negative relationship between NPLs and capital ratios

are also found by Salas and Saurina (2002), Louzis et al., (2010) and Stolz and

Wedow (2005) Salas and Saurina (2002) investigate determinants of NPLs of

Spanish Commercial and Savings Banks for the period 1985-1997 They find a negative impact of lagged solvency ratio to NPLs which is consistent with the moral

hazard hypothesis Louzis et al (2010) also mention that there is a negative influence

of capitalization to NPLs when they examine empirically this relation for Greek banking sector. Hellmann et al (2000) and Stolz and Wedow (2005) indicate NPLs

have a positive coefficient of CAR and they explain that bank raised capital to keep

up their buffer when portfolio risk risen

Procyclical credit policy hypothesis refers to the relation between loan growth and

NPLs Accordingly, banks adopt a liberal credit during the boom of the cycle, and a tight policy in the contraction phase A number of studies find a negative relationship between

loan growth and NPLs (Louzis et al 2010; Le 2016; Jimenez, Salas, and Saurina 2006) A

number of other studies find that loan growth have a positive relationship to NPLs (Clair,

1992; Keeton 1999; Demirguc-Kunt and Detragiache 1997; and Foo et al 2010)

Size effect hypothesis mentions that there is the relationship between bank size

and asset quality Bank size is negatively related to NPLs For economies of scale, larger banks can have lower costs and undertake more screening and monitoring This helps banks to reduce credit risk arising from asymmetric information between lenders and borrowers Some studies are consistent with a positive relationship

between NPLs and bank size (e.g Louzis et al, 2010; Das and Gosh 2007, Le 2016);

while other studies find bank size is negatively related to NPLs (e.g Salas and Saurina 2002)

The determinants of NPLs should not be sought exclusively in bank-specific factors but

also are viewed in macroeconomic factors The financial accelerator theory, discussed in

Bernanke and Gertler (1989), Bernanke and Gilchrist (1999), and Kiyotaki and Moore (1997), is the most prominent theoretical framework for macro-financial linkages and credit risk This theory explains credit risk and its relationship with the cyclical fluctuations in the economy During business upturn, NPLs ratios tend to be low because high borrowers’ net worth which improve the debt servicing capacity of borrowers and the lenders assume less risk when lending to high net worth agents This leads to a loosening of lending standards and strong credit growth derived from competitive pressure and optimistic macroeconomic outlook In downturns, NPLs ratios is high because borrowers’ net worth is reduced This coupled with the decline in the value of

collaterals, engenders great caution among lenders, and lead to tightening of credit extension

Empirical studies tend to examine the macro-financial linkages and NPLs Salas and Saurina (2002) estimate a significant negative effect of GDP growth on the NPLs ratio from Spanish bank sector They conclude a quick transmission of macroeconomic

developments to the ability of economic agents to service their loans Beck et al (2015)

estimate that the most significant factors affecting NPLs are GDP growth, share prices, interest rates and the exchange rate Nkusu (2011) finds that a deterioration in the macroeconomic environment—proxied by slower economic growth, higher unemployment or falling asset prices—is associated with rising NPLs On the contrary, improving macroeconomic conditions reduce NPLs Ghosh (2006) conclude that the variables related to NPLs increases are unemployment, inflation, and public debt Fofack (2005) also notes that the NPLs can be determined by different factors e.g GDP, interest rate, exchange rate, net interest margins, interbank loans Espinoza and Prasad (2010) show that NPLs decline with growth and rise with interest rates and fiscal and external

deficits by introducing macro variables Louzis et al (2010) notes that NPLs are

significantly related to macro variables and the quality of management Messai (2013) finds that unemployment and the real interest rate influence NPLs positively

3 Methodology

Following the earlier literature discussion (e.g Salas and Sarina, 2002, Merkl and

Stolz 2005, Louzis et al., 2010 and Anastasiou et al., 2016 on banking and

macroeconomic related studies), a dynamic approach is adopted in order to account for the time persistence in the NPLs structure The relationships between determinants and

NPLs can be specified as follows:

𝑁𝑃𝐿$% = 𝛼𝑁𝑃𝐿$%()+ 𝛽𝑀$%+ 𝜋) 𝐹$%+ 𝜀),$%, 𝛼 ≤ 1 (3.1)

where t and i denote time period and banks, respectively, 𝜀),4,5,6,$% = 𝜂%+ 𝜐$% and

𝜂$% is an unobserved bank-specific effect, 𝜐$% is the idiosyncratic error term To test for

the persistence of NPLs, we use lagged NPLs (i.e., NPL t -1 ) as an explanatory variable

and we expect a positive and significant sign The vector of explanatory variables includes bank-specific variables (F), included the profitability proxied by the ratio of equity on total assets, the capital and solvency presented by the ratio of equity on total assets and the

ratio of loan to deposit, loan proxied by percentage change in gross loan, and macroeconomic factor (M) included GDP, inflation, Government budget balance (% GDP) and income tax (% GDP) Previous researches show that bank-specific characteristic variables are likely to be potentially endogenous (Athanasoglou et al 2008) and some other independent variables are not strictly exogenous

This paper applies the two-step dynamic panel data approach suggested by Arellano and Bover (1995) and Blundell and Bond (2000) and also uses dynamic panel GMM technique to address potential endogeneity, heteroskedasticity, and autocorrelation problems in the data (Doytch and Uctum, 2011) The dynamic panel data model provides for a more flexible variance-covariance structure under the moment conditions The GMM approach is better than traditional OLS in examining financial variable movements Driffill et al (1998) indicate that a conventional OLS analysis of the actual change in the short rate on the relevant lagged term spread yields coefficients with some wrong signs and wrong size The research also follow Windmeijer’s (2005) finite-sample correction to report standard errors of the two-step estimation, without which those standard errors tend to be severely downward biased By using GMM estimation, it allows for instrumenting of the endogenous variables and provides consistent estimates We use the lags of right hand side variables in the equations as instruments The two-step estimation

is used because it is asymptotically more efficient than the one-step estimation for the presence of heteroskedasticity and serial correlation (Blundell and Bond 2000) In this estimation, the Hansen J-test is used to test the validity of instrument sets and the Arellano-Bond test is applied to check the absence of second-order serial correlation in the first differenced residuals

Table 2 lists the variables used in this study The NPLs variable is represented by NPLs

to gross loan The macroeconomic variables consist of the real GDP annual growth rate (GDP); inflation calculated as the average change in the CPI (INF); Government budget balance as % of GDP (FISCAL); Income tax as % of GDP (TAXC); and unemployment rate

Table 2

Summary of explanatory variables

Classification Variable Descriptions

Bank-level

Variables

(Bankscope)

NPL Ratio of non-performing loan to total loans ROE Ratio of net income after tax to average equity ETA Ratio of equity on total assets

LTD Ratio of loan to customer deposit LGR Percentage change in gross loan provided to

non-bank sectors

TA Logarithm of bank’s total asset Macroeconomic

Variables

( IMF - IFS)

GDP Real GDP annual growth rate INF Inflation, average consumer price (percentage

change) FISCAL Government budget balance as % of GDP TAXC Income tax as % of GDP

UNEMP Unemployment rate

4 Descriptions of variables and data sources

Models are estimated on an annual panel dataset of 204 commercial banks in eight Southeast Asian countries (Singapore, Malaysia, Indonesia, Philippines, Thailand, Vietnam, Cambodia, and Laos) from 2010 to 2015 The bank-level data are extracted from BankScope, and it consists of 903 observations The macroeconomic data come from IMF – IFS website

Table 3 reports the summary of statistics for the maximum, minimum, average and standard deviation of the variables used to estimate determinants of non-performing loans The statistics are calculated from yearly data in which all variables are expressed

in percentage From these figures, the NPLs ratio is from 0.00% to 101.22%, and the return on equity is from 86.751% to 82.786% show the difference in profitability of different banks Besides that, the loan to deposit is very large with 102.8579% This shows that the Southeast Asian banks still depend on lending activities

Table 3

Descriptive statistics of variables

Source: Bankscope, authors’ own estimations

Because our panel is unbalanced, we employ the unit root test by Augmented Dickey-Fuller (ADF) Fisher type test The null hypothesis shows that all panels contain a unit root The results are presented in Table 4 All of our variables are found to be stationary Correlation coefficients among all our variables are found not to exceed 0.382

Table 4

Unit root test

Fisher type ADF p-values Fisher type ADF Statistics

Source: Bankscope, own estimations

5 Empirical results

The estimation results are presented in Tables 5, reporting the respective impacts of determinants on NPLs Various specifications of Eq 3.1 are examined Specification 1 shows estimated parameters of NPLs, which is subjected to bank-specific characteristics suggested by the literature The lagged of the bank-specific variables are added to specification 2 Specification 3 and 4 respectively show the impact of macroeconomic variables and the lagged of these variables Specification 5 presents the results of both specific-bank variables and macroeconomic variables, and the lagged of these variables

are then included to specification 6

Table 5

GMM estimation results for Southeast Asian area NPLs, 2010 - 2015

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

NPL it-1

0.6797***

(0.0051)

0.5826***

(0.0043)

0.6487***

(0.0063)

0.8719***

(0.0203)

0.6544***

(0.0021)

0.6335*** (0.0021)

ROE

-0.0198*

(0.0074)

-0.0937***

(0.0143)

-0.0120***

(0.0020)

-0.0561*** (0.0043)

ROE it-1

-0.0393**

(0.0137)

0.0526*** (0.0067)

ETA

0.0466***

(0.0084)

0.0357 0.0359

0.0635***

(0.0039)

0.1123*** (0.0117)

ETA it-1

-0.0766 (0.0360)

-0.0395** (0.0132)

LTD

-0.0052***

(0.0012)

-0.0132***

(0.0017)

-0.0059***

(0.0006)

-0.0081*** (0.0025)

LTD it-1

0.0066***

(0.0018)

0.0021 (0.0016)

LGR

-0.0616***

(0.0045

-0.0366***

(0.0057)

-0.0360***

(0.0007)

-0.0394*** (0.0018)

GGL it-1

-0.0100***

(0.0017)

0.0136*** (0.0007)

LnTA

-0.2858***

(0.0619)

0.06935 (0.7024)

-0.2284***

(0.0364)

-0.455 (0.2886)

LnTA it-1

-0.2836 (0.6857)

0.1834 (0.2880)

Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

GDP

-0.1467***

(0.0409)

-0.0964**

(0.0340)

-0.0810***

(0.0149)

-0.1418*** (0.0248)

GDP it-1

0.1162***

(0.0331)

0.0555*** (0.017)

INF

0.0564*

(0.0216)

0.0503**

(0.0175)

0.0051 (0.0080)

-0.0257 (0.0113)

INF it-1

0.1279***

(0.0279)

-0.0380 (0.0167)

FISCAL

-0.1935***

(0.0598)

0.0564 (-0.0550)

-0.0691***

(0.0049)

-0.0837** (0.0270)

FISCAL it-1

-0.0608 (0.5840)

0.047 (0.0219)

TAXC

-0.1013 (0.0497)

-0.0140 (0.0856)

0.1240***

(0.01293)

-0.0495 (-.0336)

TAXC it-1

-0.0917 (0.0770)

0.2263*** (0.0370)

UNEMP

0.0814 (0.0636)

0.4181*

(0.1556)

0.3371***

(0.02568)

0.4559*** (0.0852)

UNEMP it-1

-0.4541**

(0.1533)

-0.0870 (0.0789)

Constant

8.5471***

(1.4580)

9.2280***

(2.2121)

2.0878**

(0.7301)

2.1810***

(0.4632)

3.9292***

(-0.8075)

4.0543*** (1.1536)

Pro>chi2 0.000 0.000 0.000 0.000 0.000 0.000

Hansen test 0.101 0.210 0.391 0.118 0.358 0.442

Source: Bankscope, own estimations

***, **, * * and ** denote significance at the 10 %, 5 % and 1% levels, respectively 5% và 10% Standard errors in parentheses

The estimation results in Table 5 confirm that both bank-level and macroeconomic factors play a role in affecting the NPLs of Southeast Asian banks Our finding shows that the highly significant coefficient value of the NPLs persistence The coefficient’s size of the lagged NPL ranges between 0.5826 to 0.8719, thus suggesting that a shock to NPLs is likely to have a prolong effect on the Southeast Asian banking system