The Determinants of Loan Quality: Evidence
from Indonesian State-Owned and Commercial
Banks
Master ThesisMSc International Economics & Business
Author: Supervisor:
Rangga Kresna (s2844265) Prof. dr. J. de Haan
r.kresna@student.rug.nl jakob.de.haan@rug.nl
Co-supervisor: dr. A. A. Erumban a.a.erumban@rug.nl
2
Acknowledgments
3
Contents
I. Introduction ... 5
II. Literature Review ... 7
a. Macroeconomic factors... 7
b. Bank specific variables ... 9
III. Data and Methods ... 9
a. Data ... 9
b. Methods ... 12
IV. Empirical Results ... 15
a. The relationship between macroeconomic factors and NPL ... 19
b. The relationship between bank-specific variables and NPL ... 20
c. Robustness tests ... 20
V. Conclusions ... 22
Appendix ... 24
References ... 25
List of Tables Table 1. The variable explanations ... 11
Table 2. Summary statistics ... 12
Table 3. Correlation matrix ... 15
Table 4. Panel unit root tests ... 16
Table 5. Empirical results: Fixed Effects and Random Effects Model ... 17
Table 6. Empirical results: one step Arellano-Bond GMM ... 18
Table 7. Robustness checks: System GMM ... 21
List of Figures
4
The Determinants of Loan Quality: Evidence from Indonesian
State-Owned and Commercial Banks
Abstract
The objective of this study is to examine the relationship between macroeconomic, bank-specific variables and loan quality of Indonesian state-owned and foreign exchange commercial banks from 2011Q1 through 2016Q2. This study employs both static and dynamic panel data models. The empirical result shows that the slowdown in economic growth increases the non-performing loans (NPLs). In addition, an increase in operating costs and ratio of loans to deposits also leads to a rise of NPLs, while an increase of return on assets declines NPLs. The empirical results indicate that both macroeconomic factor and bank management quality significantly affect loan quality of banks in Indonesia.
5
I. Introduction
Nowadays, the banking industry has become one of the largest industries. As a financial intermediary, a bank aims to take deposits and makes loans to society. According to De Haan, Oosterloo, and Schoenmaker, (2009), in doing so, a bank performs two functions. First, banks minimize information and transaction costs. Acquiring information about investment projects is costly because each investor faces a high fixed cost in evaluating these projects. Banks can reduce the cost to acquire and process the information. Furthermore, banks are also able to reduce transaction costs because banks collect the deposits from plenty of small depositors for financing a large investment project. Second, banks facilitate trading, diversification, and risk management. Managing risk is highly critical, and banks can diversify their investments to manage the risk.
A bank manages four key risks, namely market risk, operational risk, liquidity risk, and credit risk (Aaron, Armstrong, & Zelmer, 2007). The price fluctuations of financial assets and liabilities can cause market risk. Most banks apply the value at risk (VAR) method to measure and manage their exposure to market risk. Liquidity risk happens when a bank cannot supply sufficient liquidity to meet demand due to their inability to liquidate assets or raise funds. Diversification of funding sources, maturities, customer types, market, currency, and regions are the method to manage liquidity risk.
Operational risk refers to the risks due to the failures of the internal, such as the system and employee and so forth, or external events. To manage operational risk, Aaron et al. (2007) suggest that a bank has to apply internal processes and controls carefully. Credit risk is the inability of borrowers to repay interest, principal, and other obligations of the loan facility. In the traditional banking business, a bank should be able to manage the mismatch of the length of the maturity of their assets and the liabilities. Normally, the bank’s liabilities are short term, while most of the bank’s assets are long term. Allen and Gale (2000) argue that banks invest in term assets because it has a higher return than the short-term assets if the long-term assets are held to maturity. However, liquidating the long-long-term assets in the middle period is costly. The loan is an earning asset that has a long maturity. A bank has to diversify its earning assets across industries and regions and carefully monitor borrowers to prevent a loss from credit risk.
6
Figure 1. The mean of non-performing loans to total gross loans of 41 Indonesian commercial banks
(Sources: Otoritas Jasa Keuangan (Financial Services Authority of Indonesia), author’s calculations)
NPLs is one of the measurements of loan quality (Clair, 1992). A bank that has high NPLs implies that the bank has a low loan quality. The previous studies found that the economic cycle strongly influences the fluctuation of the loan quality (Abid, Ouertani, & Zouari-Ghorbel, 2014; Ali & Daly, 2010; Bashir, Yu, Hussain, Wang, & Ali, 2017; Beck, Jakubik, & Piloiu, 2013; Castro, 2013; Dimitrios, Helen, & Mike, 2016; Espinoza & Prasad, 2010; Festić, Kavkler, & Repina, 2011; Filip, 2015; Gerlach, Peng, & Shu, 2005; Louzis, Vouldis, & Metaxas, 2012; Messai & Jouini, 2013; Shen & Lin, 2012; Us, 2016). However, none of the previous studies provides a specific research of the association between economic cycle and loan quality in Indonesia by employing firm-level data.
This research aims to provide an empirical study of the relationship between macroeconomic, bank-specific variables, and loan quality of Indonesian commercial banks by using firm-level data. Academically, this study will enrich the previous studies of the relationship between banks’ loan performance and the economic cycle by taking a sample of state-owned and commercial banks in Indonesia. Moreover, this paper provides insight for boards of management of Indonesian commercial banks in setting the business target and maintaining asset quality. In addition, for the intermediary financial regulator, this study will contribute to design policies and regulations to enhance financial system stability. According to De Haan et al. (2009), a financial system can stimulate economic development when it can perform its primary task by channeling the funds from one sector to another sector.
This study employs NPLs as the explained variable. The macroeconomic factors are represented by real GDP growth, real effective exchange rate, and the inflation rate. The bank-specific factors used in this study are a return on assets, the ratio of loan to deposit, and the ratio of operating expenses to operating income, and NPLs. This study employs static and dynamic panel data model to analyze the panel data of Indonesian commercial banks for the
7
period 2011Q1-2016Q2. The empirical results show that the lower in real GDP growth leads to a higher contemporaneous NPL. All bank-specific variables significantly affect NPLs. The past NPLs and return on assets have a negative relationship with NPLs, while the cost efficiency and ratio of loan to deposit have a positive correlation with NPLs. The empirical results suggest that both macroeconomic factor and bank management quality significantly affect the credit quality of banks in Indonesian.
This paper is organized as follows. Section 2 provides a literature review. Section 3 discusses data and methods that are applied in this study. Section 4 provides the empirical results. Section 5 offers the conclusion and policy implication of this study.
II. Literature Review
According to the International Monetary Fund (2006), non-performing loans are loans of which the payment of principal and interest has been due for 90 days or more. The Indonesian financial intermediary regulator, basically, adopts the same definition for NPL. In specific, the Indonesian central bank, Bank Indonesia, distinguishes five levels of loan quality, namely current, special mention, substandard, doubtful, and lost. The current loan is a loan that all principal and interest payments are fully paid according to schedule, while the special mention loan the principal and interest loan payments past are due less than 90 days. The current and special mention loans are grouped in the performing loans. The last three levels of loans, i.e. substandard, doubtful, and lost, are grouped as the non-performing loans. The sub-standard, doubtful, and lost loans are loans that the principal and interest loan payment are past due more than 90 days, 120 days, and 150 days, respectively.
Empirical studies distinguish two sets of variables that are associated with NPLs, namely macroeconomic and bank-specific factors (e.g. Abid et al., 2014; Castro, 2013; Dimitrios et al., 2016; Espinoza & Prasad, 2010; Festić et al., 2011; Gerlach et al., 2005; Louzis et al., 2012; Messai & Jouini, 2013; Us, 2016).
a. Macroeconomic factors
Gross domestic product (GDP) growth indicates the economic growth and is a standard measure of economic progress. Messai and Jouini (2013) argued that in the economic expansion phase, borrowers are more able to pay back their debt because they have sufficient income to service their debt. So this phase of the business cycle is associated with a low level of NPLs. Loans will be provided to lower-quality borrowers if the expansion phase persists. Consequently, in a recession period, the borrowers are not able to service their debts which lead to an increase of NPLs. They found that the growth rate of GDP is negative and statistically significantly affecting the NPLs of 85 banks in Italy, Greece, and Spain for the period of 2004 through 2008. They also found that the coefficients of the unemployment rate and the real interest rate are positive and also statistically significant.
8
found that economic growth, unemployment, and the interest rate are statistically significant. The real GDP growth has a negative relationship with NPLs, which indicates that an increase of GDP growth leads to a decline of NPLs. Meanwhile, both levels of unemployment and interest rate have a positive effect on NPLs which suggests that a rise of both variables increases NPLs. Also, he includes the real effective exchange rate in his model. He found that the real effective exchange rate also has a positive impact on NPL. A depreciation of the local currency implies that goods and services produced domestically become less expensive. As a result, it strengthens competitiveness in the foreign market so that borrowers are more able to service their debts, and it decreases NPLs.
The empirical studies by Messai and Jouini (2013) and Castro (2013) support the findings of previous studies by Louzis et al. (2012) and Espinoza and Prasad (2010). Both studies employ dynamic panel models in examining the relationship between macroeconomic factors and NPLs. Using a panel data of the nine Greek banks over the period 2003 – 2009, Louzis et al. (2012) found that GDP growth rate and the real lending rates have a positive impact on the level of NPLs for consumer loans, business loans, and mortgages, while the level of unemployment is negatively associated to the NPLs. Espinoza and Prasad (2010) also found that non-oil GDP is negatively related to the NPLs of the banks in the Gulf Cooperative Council (GCC) countries in the of 2003 and 2008. In addition, more recent studies confirm the finding of the negative relationship between GDP growth and NPLs (Dimitrios et al., 2016; Us, 2016).
In contrast, Beck et al. (2013) found that the contemporaneous and lagged real GDP of 75 countries have a different impact on NPLs in the period of 2000-2010. The contemporaneous real GDP has a negative relationship with NPLs, while the lagged real GDP has a positive association with NPLs. It indicates that an increase in the current real GDP growth decreases NPLs. However, banks keep providing loans to lower-quality borrowers in a sustained expansion phase which leads to an increasing of the current NPLs. They also found that the nominal effective exchange rate has a positive impact on NPLs. They argue that a depreciation of the domestic currency leads to a decline in NPLs.
9
1994-2002. They argue that the increase inflation alleviates the real value of loan outstanding and enhance the repayment capacity of the borrower to service the debt.
b. Bank specific variables
The evolution of NPLs is not solely influenced by the macroeconomic factors, which are exogenous to the banking industry. The efficiency, the risk management, and other policy choices selected by each bank, which are endogenous to the banking sector, are also expected to influence the growth of NPLs.
Louzis et al. (2012) found that return on equity (ROE) has a negative relationship with NPLs, whereas the ratio of operating expense to operating income (CIR) has a positive effect on NPLs. They argue that a high ROE implies a better quality of management skills in lending activities. As a result, the bank’s management has sufficient skills to reduce future NPLs. They call this as ‘tight control hypothesis.' If the relationship between ROE and NPLs is positive, it reflects the ‘pro-cyclical credit policy hypothesis’ which implies that the management inflates the earning for short-term reputation reason. The consequence is an increase in the future NPLs.
Bashir et al., (2017) also found a positive relationship between CIR and NPLs for panel data sets which consist of 116 Chinese banks. This finding supports the ‘bad management hypothesis' as suggested by Berger and DeYoung (1997). The low efficiency (high CIR) reflects a poor quality of bank‘s management due to their inability to monitor and control their operating expenses. A high CIR value is also related to insufficient skills of the bank’s managers in credit scoring, to assess the collateral, and to monitor the borrowers after loan disbursement. The lack of skills of the bank’s managers will leads to a rise of NPLs in the future.
Dimitrios et al. (2016) also include bank-specific factors in addition to macroeconomic factors to explain the growth of NPLs. Using the return on assets (ROA) and ROE to account for the impact of the managerial efficiency to the NPLs; they found that both variables are negative and statistically significant. It indicates that more efficient bank’s managerial skills in converting assets and equity into profit has a lower NPLs. Their findings support a previous study by Louzis et al. (2012). Moreover, using loans to deposits ratio (LTD) variable to assess the managerial risk preference, they found a positive relationship between LTD and NPLs.
III. Data and Methods
a. Data
This study uses a panel data of 41 commercial banks which consist of four state-owned banks and thirty-seven foreign exchange commercial banks in Indonesia. In June 2016, total assets of these banks were 75% of total assets of all Indonesian banks1. The banks in these groups
1
10
are allowed to make a loan in foreign currencies so that this study can examine the association between the volatility of the exchange rate and NPLs. In addition, this study only includes all banks that have a complete financial report data during the observation period. The observation period starts from the first quarter 2011 because prior to 2011, the financial reports of Sharia banks, some of which are business units of commercial banks, were reported on a consolidated basis with the commercial banks2. Consequently, the financial reports of some banks have changed significantly since 2011 because the financial reports of Sharia banks were reported separately. The period of observation ends in the second quarter of 2016 because the Bank Indonesia applied Bank Indonesia (BI) rate until July 2016. Afterward, Bank Indonesia introduced a new policy rate, namely the BI 7-Day (Reverse) Repo Rate. The BI rate was implemented in the Bank Indonesia monetary operations through movement in the Interbank Overnight (O/N) Rate. It was expected that the change in interbank rates would affect both bank deposit rate and lending rates. Meanwhile, the new BI 7-Day (Reverse) Repo Rate intend to accelerate the policy rate transmission to the money market, banking sector, and real sector. The BI 7-Day (Reverse) Repo Rate has a lower rate than BI rate because the BI 7-Day (Reverse) Repo Rate will be set weekly, whereas BI rate was set monthly. Therefore, this study restricts the period of observation only when the BI rate was applied. To sum up, this study has 41 cross-sectional units and 22 time dimension (N = 41, T = 22).
The NPL is the dependent variable in all models. NPLs are the ratio of non-performing loans to total gross loans, and it reveals the loan quality of a bank. To analyze the relationship between macroeconomic and NPL, this study includes growth of real gross domestic product (GGDP), the growth of Real Broad Effective Exchange Rate (GREER), and the inflation rate (INF) as the explanatory variables as suggested by the previous studies (Bashir et al., 2017; Beck et al., 2013; Castro, 2013; Espinoza & Prasad, 2010; Louzis et al., 2012; Messai & Jouini, 2013; Us, 2016). The objectives to include these macroeconomic variables are to examine the effect of economic growth, the volatility in the exchange rate, and the impact of inflation rate toward NPLs. The real GDP and consumer price index (CPI) data are obtained from the International Monetary Fund (IMF) database, whereas the real broad effective exchange rate is obtained from Bank of International Settlement (BIS). All the data sources are secondary sources.
In addition to macroeconomic factors, as the explanatory variables, there are three bank-specific factors included in this study, i.e. return on assets (ROA), loan to deposit ratio (LDR), and the ratio of operating expense to operating income (CIR) as suggested by previous studies (Bashir et al., 2017; Dimitrios et al., 2016; Louzis et al., 2012). The objectives to include these three bank-specific variables are to examine the quality of bank’s
Authority of Indonesia) and Indonesian Banking Statistics Vol. 14 No.07 (Otoritas Jasa Keuangan, 2016). The top 4 biggest Indonesian commercial banks based on assets, PT BANK MANDIRI (PERSERO), Tbk, PT BANK RAKYAT INDONESIA (PERSERO), Tbk, PT BANK CENTRAL ASIA, Tbk, and PT BANK NEGARA INDONESIA (PERSERO), Tbk, were among these 41 banks. These 4 biggest banks have branches throughout Indonesia.
2
11
management skills in converting assets into profit (reflected by ROA), the effect of loan expansion on credit quality (indicated by the LDR) and the effect of cost efficiency on loan quality (reflected by CIR). The NPL, ROA, LDR, and CIR data are obtained from Otoritas Jasa Keuangan (Financial Services Authority of Indonesia). Table 1 provides the explanation of all variables used in this study.
Table 1. The variable explanations
Symbol Description Source
Dependent variable NPL The ratio of nonperformance loans over
total loans (in percent)
Otoritas Jasa Keuangan (Financial Services Authority of Indonesia) (cfs.ojk.go.id/cfs/) Independent variable
SOB Dummy 1 for state-owned banks, otherwise 0
TBK Dummy 1 for listed banks, otherwise 0
Country level GGDP Quarter-on-quarter change in real gross
domestic product (in logs)
International Monetary Fund / International Financial Statistics
(data.imf.org) GREER Quarter-on-quarter change in Real
Broad Effective Exchange Rate (in logs)
Bank of International Settlements (bis.org/statistics/eer)
INF Quarter-on-quarter change in the consumer price index (in logs)
International Monetary Fund / International Financial Statistics
(data.imf.org) Bank level
CIR Operating expenses to operating income ratio (in percent)
Otoritas Jasa Keuangan (Financial Services Authority of Indonesia) (cfs.ojk.go.id/cfs/) ROA Return on assets (in percent) Otoritas Jasa Keuangan (Financial Services
Authority of Indonesia) (cfs.ojk.go.id/cfs/) LDR Loan to deposit ratio (in percent) Otoritas Jasa Keuangan (Financial Services
Authority of Indonesia) (cfs.ojk.go.id/cfs/)
12
ordinary least square (OLS) estimator still can be applied in spite of the data is not normally distributed with its mean and variance.
Table 2. Summary statistics
Variable Mean SD Skewness Kurtosis p50 Max Min
NPL 2.00 1.41 0.60 2.76 1.96 6.36 0.00 GGDP 0.01 0.02 -0.38 1.47 0.03 0.04 -0.02 GREER 0.00 0.04 -0.24 3.16 -0.01 0.06 -0.10 INF 0.01 0.01 1.24 5.37 0.01 0.04 0.00 ROA 1.77 1.12 0.73 3.41 1.61 5.25 -1.03 CIR 84.19 9.49 -0.46 2.85 85.16 108.05 59.93 LDR 84.95 13.91 0.45 5.78 85.36 150.82 39.86 b. Methods
Panel data are observations of the same cross-sectional units (individuals, firms, countries, etc.) over time. A balanced panel is a set of data that has the same number of period observations for each cross-sectional unit. In a panel data model, the error term of the regression includes the unobserved individual-specific characteristics or heterogeneity. There are two ways to recognize the existence of individual components. First, the individual errors in different time periods are allowed to be correlated. Second, it is assumed that all individuals have the same coefficients. There are different types of panel data sets, namely long and narrow, short and wide, and long and wide. If the time dimension is long, but the number cross-sectional units are small, it is called as long and narrow. The period of observation is short, but it has a large number of cross-sectional units, it is called as short and wide. Long and wide imply that the period observations and number of cross-sectional units are long and broad.
Hill, Griffiths, and Lim (2012) argue that in short and wide panels, the coefficients cannot be assumed to be different for different individuals because the resulting estimates would not be precise making estimation impossible. In this type of panel data, it is assumed that the intercepts are different for different individuals, but the slope coefficients are constant for all individuals. The individual heterogeneity is absorbed by the intercepts which are called fixed effects. The fixed effects (FE) estimator and the least squares dummy variable estimator are two methods to estimate a model that controls for individual specific and time-invariant characteristics. This study will apply the FE estimator because it can control for individual-specific and time-invariant characteristics that may impact or bias the predictor due to the omitted time-invariant characteristics (Hill et al., 2012). Another critical assumption of the FE estimator is the error term of each entity, and the intercept should not be correlated with the others because each entity is distinct.
time-13
invariant variables because it assumes the variation across entities is random and uncorrelated with the predictor or independent variables. Both RE and FE estimators are biased in a dynamic panel data model because the lagged dependent variable has a positive correlation with the fixed effects in the error term (Nickell, 1981). Nonetheless, Castro (2013) argues that the FE estimator has an advantage in a long time dimensional in dynamic panel data model because the result becomes consistent and the bias from the correlation between the lagged dependent variable and the independent variable becomes small.
In many studies, the determinants of the asset quality of banks are estimated by dynamic panel approaches because the past NPLs influence the current realization of the NPLs. Most of these studies employ either difference GMM as proposed by Arellano and Bond (1991) or system GMM as suggested by Arellano and Bover (1995) and Blundell and Bond (1998) (Abid et al., 2014; Bashir et al., 2017; Beck et al., 2013; Castro, 2013; Dimitrios et al., 2016; Espinoza & Prasad, 2010; Louzis et al., 2012; Us, 2016).
In general, the difference and system GMM estimators are applied to a linear functional relationship; the dependent variable rely on its past realizations; the past and possibly current realizations of the error are correlated with the independent variables; fixed individual effects; there is no heteroscedasticity and autocorrelation across individual but within them; and it is used when time dimensional is smaller than cross-sectional units (Roodman, 2009). However, some studies also apply these estimators despite the time dimensional being larger than cross-sectional units (Castro, 2013; Louzis et al., 2012).
In a dynamic panel data model, the correlation between the explanatory variable and the error term violates an assumption for the consistency OLS. To overcome the endogeneity problem, Roodman (2009) suggests two ways. First, the data is transformed to remove the fixed effects. This strategy is incorporated with the difference GMM. Despite the fixed effects are eliminated; the lagged dependent variable is still potentially endogenous and correlated with the idiosyncratic shocks. Second, , and other endogenous variables which are uncorrelated with the fixed effects are used as the instrument. It is incorporated with the system GMM.
The general model of the dynamic panel containing a lagged dependent variable and a single regressor X is as follows:
Equation (1)
= + , + ′ + +
14 Equation (1) can be re-written as:
Equation (2)
− , = ( , − , ) + ( ′ − ′ ) + −
∆ = ∆ , + ∆ ′ + ∆
Roodman (2009) argues that due to the transformation into first differences a correlation problem emerges between the lagged dependent variable in ∆ , term and in
∆ term. It imposes a bias in the estimation of the model. Louzis et al. (2012) suggest using
, term to solve the endogeneity problem. The , term is expected to correlate with
Δ , and not correlate with Δ for t =3,... ,T. The , term is used as an instrument in the estimation of equation (2). This suggests that lags of order two, and more, of the dependent variable satisfy the following moment conditions (Louzis et al., 2012):
Equation (3)
[ Δε ]= 0 for t = 3,…,T and s ≥ 2.
In the case of predetermined variables in that are not strictly exogenous, the error term is not correlated with all past and future values of the explanatory variable. It implies the following moment conditions (Louzis et al., 2012):
Equation (4)
[ Δ ] = 0 t = 3,…,T and for all s.
Hence, only current and lagged values of are valid instruments for a set of weakly exogenous or predetermined explanatory variables. The following moment conditions can be used (Louzis et al., 2012):
Equation (5)
[ Δ ] = 0 t = 3,…,T and for s ≥ 2.
15
the fixed effects (Roodman, 2009). However, the weak instruments problem remains in the system GMM estimator (Ashley & Sun, 2016).
The basic model estimated in this paper is adopted from Equation (1) by transforming it into first difference as explained by Equation (2) as follow:
Equation (6)
∆ = ∆ + ∆ + ∆ + ∆
where the subscripts i denotes the cross-sectional units, whereas t implies time dimension of the panel sample. ∆ is the change of the non performing loan ratio. Due to correlation between the transformed error term ∆ and first lagged dependent variable, ∆ , the ordinary least square on the first differenced data in a dynamic model generates inconsistent parameter. Louzis et al. (2012) proposed using the second lagged dependent variable, , as an instrument for ∆ since [ Δε ]= 0. The ∆ term is the difference of a vector of macroeconomic factors, whilst the ∆ term is the difference of bank specific variables.
IV. Empirical Results
Prior to reporting the estimation results, this study provides the correlation matrix for measuring the relationship amongst variables. This study also conducts panel unit root tests to check the presence of the unit roots in all variables. Subsequently, this study will discuss the results of fixed effects (FE) estimator, random effects (RE) model, and Arellano-Bond (AB) GMM.
Table 3. Correlation matrix
NPL GGDP GREER INF ROA CIR LDR
NPL 1 GGDP 0.0369 1 GREER 0.0306 -0.3437* 1 INF -0.0750* -0.0211 -0.2052* 1 ROA -0.1156* 0.0215 -0.0547 0.0153 1 CIR 0.1598* -0.0193 0.0437 -0.0173 -0.9177* 1 LDR 0.1443* 0.0316 -0.0043 0.0237 0.1127* -0.1478* 1
* Indicates significance at the 5% level
16
and INF and ROA is negative, while the correlation coefficient between NPL and other variables are positive.
Table 4 provides the result of the panel unit root tests. This study applies the Fisher-type unit root tests to check for the presence of the unit roots because this unit root test can be applied in the panel data that has unbalanced data. The Fisher-type unit root tests that implemented in this study are conducted based on augmented Dickey-Fuller (Fisher-ADF) test. One lag is performed in all regression. The Fisher-ADF test shows that NPL is stationary at first difference I(1) at 5%, while other variables are stationary at the level I(0) because at least one panel is stationary.
Table 4. Panel unit root tests
Variables Fisher-ADF
Inv. Inv. N Inv. L M.Inv.
NPL 97.2408 1.0225 1.0213 1.1901 [0.12] [0.8467] [0.8458] [0.117] ∆NPL 326.2296*** -10.2768*** -12.7635*** 19.0711*** [0.0000] [0.0000] [0.0000] [0.0000] GGDP 2899.97*** -51.3033*** -125.149*** 220.0465*** [0.0000] [0.0000] [0.0000] [0.0000] GREER 619.7142*** -20.4541*** -26.6354*** 41.9884*** [0.0000] [0.0000] [0.0000] [0.0000] INF 137.6177*** -4.9592*** -4.5002*** 4.343*** [0.0001] [0.0000] [0.0000] [0.0000] CIR 112.412* -0.9438 -0.9399 2.3748** [0.0145] [0.1726] [0.1742] [0.0088] LDR 150.818*** -4.1075*** -4.5037*** 5.3738*** [0.0000] [0.0000] [0.0000] [0.0000] ROA 108.8885* -0.5422 -0.727 2.0996* [0.0252] [0.2938] [0.2340] [0.0179]
Notes: The Fisher-augmented Dickey-Fuller (Fisher-ADF) unit root tests are performed over an unbalanced panel for the period 2011q1-2016q2 with one lag for all regressions; the null hypothesis is that “all panels contain unit roots,” while alternative hypothesis is “at least one panel is stationary.” The p-values are in square brackets. The number of stars (*) denotes significance level: * p<0.05, ** p<0.01, *** p<0.001. Δ is the first difference operator.
This study follows the analysis suggested by Beck et al. (2013). First, this study applies static panel data model. Fixed Effects (FE) estimator and Random Effect (RE) model are used to estimate static panel data as suggested by Shin, Min, and Mcdonald (2014). Table 5 provides the empirical results with FE and RE estimators. Hausman test is used to select the preferred model (FE or and RE model). If the random error component and any of the explanatory variables are correlated, then the RE model is inconsistent, while the FE model remains consistent (Hill et al., 2012). Second, this study uses the dynamic panel data model, namely Arellano-Bond (AB) GMM.
17
NPL can be explained by the regression model (Table 5, column (1) and (3)). The Hausman test shows that the RE estimator is the most appropriate estimator given that the Hausman test outcome is not statistically significant. This research does not discuss the result of static panel data method in detail because this article focuses on examining the relationship between macroeconomic, bank-specific variables, and NPLs based on dynamic panel data models. Table 5. Empirical results: Fixed Effects and Random Effects Model
FE RE FE RE FE RE (1) (2) (3) (4) (5) (6) GGDP(-1) -3.194* -3.156** -3.327** -3.189** -3.194* -3.156** (-2.66) (-2.64) (-2.71) (-2.66) (-2.66) (-2.64) GREER(-1) 0.378 0.415 0.321 0.41 0.377 0.406 (0.60) (0.66) (0.51) (0.66) (0.60) (0.65) INF(-1) -2.111 -2.053 -2.724 -2.172 -2.114 -2.062 (-1.20) (-1.17) (-1.58) (-1.25) (-1.20) (-1.17) CIR(-1) 0.000653 0.00129 (0.13) (0.74) LDR(-1) 0.00874*** 0.0016 (4.36) (1.45) ROA(-1) -0.00553 -0.0145 (-0.15) (-1.13) constant 0.0465 -0.0076 -0.632*** -0.0336 0.111 0.126*** (0.11) (-0.05) (-3.73) (-0.35) (1.53) (3.90) No. of Obs. 759 759 759 759 759 759 R2 0.0207 0.0303 0.0207 BIC 1411 . 1403.6 . 1411 . F 3.114 7.167 3.111 [0.0254] [0.0002] [0.0255] Hausman test 1.57 7.42 1.6 [0.8138] [0.1152] [0.8085]
Notes: Dependent variable is ∆NPL. Robust t-statistics are in parentheses. The number of stars (*) denotes significance level: * p<0.05, ** p<0.01, *** p<0.001. The model was estimated by fixed effects (FE) and random effects (RE). For each regression are presented the number of observations (No. of Obs.), the number of the coefficient of determination (R2), and the Schwarz's Bayesian information criterion (BIC). The F-test presents the statistics and respective p-values in square brackets for the test to the presence of fixed effects. The Hausman-test is used to select between a random and a fixed-effects estimator.
18
variables, whereas one lag independent variable allows for the delay of the economic shock effects, evading reverse causality issues and simultaneity problems.
Table 6. Empirical results: one step Arellano-Bond GMM
AB-GMM (1) (2) (3) ∆NPL(-1) -0.124 -0.117 -0.156 (-1.97) (-1.85) (-1.00) GGDP(-1) -3.249** -3.629** -3.155* (-2.70) (-3.01) (-2.64) GREER(-1) 0.166 0.344 0.234 (0.26) (0.56) (0.32) INF(-1) -1.9 -2.305 -1.757 (-0.91) (-1.13) (-0.86) CIR(-1) 0.0414* (2.07) LDR(-1) 0.011 (1.06) ROA(-1) -0.308** (-3.07) No. of Obs. 701 701 701 No. of Groups 41 41 41 No. of instruments 41 41 40 F 7.526 5.5 4.712 [0.000] [0.001] [0.002] AR1-test -4.68 -4.58 -2.36 [0.000] [0.000] [0.018] AR2-test -1.44 -1.31 -1.53 [0.151] [0.191] [0.127] Hansen test 39.4 39.55 37.88 [0.320] [0.315] [0.339]
Notes: Dependent variable is ∆NPL. Robust t-statistics are in parentheses. The number of stars (*) denotes significance level: * p<0.05, ** p<0.01, *** p<0.001. The model was estimated by one step AB GMM estimator. For each regression are presented the number of observations (No. of Obs.), a number of groups (No. of the group), and the number of the instrument (No. of the instrument). The F-test presents the statistics and respective p-values (in square brackets) for the test to the presence of fixed effects. AR1 and AR2 tests are the Arellano–Bond tests for first and second-order autocorrelation in first-differenced errors; The statistics and p-values (in square brackets) for the Hansen-test of overidentifying restrictions are also reported for the AB GMM estimations.
19
the AR tests show negative and statistically significant first-order autocorrelation but no evidence of second-order autocorrelation.
The Hansen-test, which has an null hypothesis valid moment conditions, confirms the validity of the instruments used in this analysis (Roodman, 2009). In addition, in GMM estimator, the number of instruments is either the same or less than the number of groups (Beck et al., 2013). Given the data in this study has a relatively large time dimension; hence an unrestricted set of lags will introduce an enormous number of instruments. Consequently, it may be a possible loss of efficiency. Therefore in using AB GMM, this study applies a restricted set of lags in constructing the GMM instruments. The bank-specific variables are separately included into the estimation to keep the number of instruments lower than the number of groups in GMM specifications3.
The estimation result shows that the first-order autocorrelation test is negative and statistically significant, whereas the second-order test is also negative but statistically not significant. It indicates the consistency of the AB GMM estimator. Moreover, the validity of the instruments used in this analysis is also confirmed by the Hansen test because it strongly rejects the null hypothesis.
a. The relationship between macroeconomic factors and NPL
One lag of the dependent variables (∆NPL) is included in the set of regressors for the AB GMM estimators as shown in Table 6. The lagged NPL is negative but statistically not significant. The negative relationship implies that banks reduce the NPLs that have lasted for three months by writing off the loan facility or by liquidating the collateral.
The one lag of economic growth is included into all models. All estimation results find that growth in real GDP is negative and statistically significant. This finding indicates that one percentage point higher the growth of the real GDP in previous quarter reduces the contemporaneous NPLs by about 0.032 percentage points (Table 6, column (1)). The higher the growth of the real GDP enhances the ability of borrowers to service their debts so that the NPLs decreased. This finding support previous studies by Beck et al. (2013) that include Indonesia in their sample.
In examining the association of the growth of real effective exchange rate on NPLs, one lag of the growth of real effective exchange rate (GREER) is implemented in all models. The empirical results find that the previous quarter of the growth of real effective exchange rate is positive but statistically not significant. The positive coefficient of the growth of real effective exchange rate indicates that higher growth of real effective exchange rate leads to a higher of current realization NPLs. Castro (2013) argues that the appreciation of real value exchange rate weaken the competitiveness of export-oriented firms which eventually leads to a decline in the ability of these companies to serves their debts.
3
20
This study also includes one lag of the inflation rate (INF) to analyze the relationship between inflation rate and NPLs. All estimation results find that the previous quarter of inflation rate is negative but statistically not significant. The negative coefficient of inflation rate implies that the high inflation rate reduces the contemporaneous NPLs due to a decline in the real value of loans burden. Statistically, the inflation rate last quarter does not affect the contemporaneous NPLs because the high inflation rate also reduces the actual income of borrowers. The absence of a significant relationship between the inflation rate and NPLs is in line with a study on the effects of the macroeconomic developments in five European countries (Greece, Ireland, Portugal, Spain, and Italy) by Castro (2013).
b. The relationship between bank-specific variables and NPL
There are three variables included in this study to analyze the relationship between the bank-specific variables and NPL, i.e. operating expenses to operating income ratio (CIR), loan to deposit ratio (LDR) and return on assets (ROA). These variables are included separately in all models.
The ratio of operating expenses to operating income (CIR) variables is statistically significant in the AB GMM estimator. The positive coefficient of CIR variable indicates that one percentage point increased of CIR leads to a rise of contemporaneous NPLs by 0.041 percentage points (Table 6, column (1)), ceteris paribus. This finding support ‘bad management hypothesis' as suggested by Berger and DeYoung (1997). A poor quality of bank‘s management in controlling their operating expenses also reflects the insufficient skills of the bank’s managers in credit scoring, to value the collateral, and to monitor the borrowers after loan disbursement that eventually leads to a rise of NPLs in the future.
The FE estimator finds the loan to deposit ratio (LDR) variable is positive and statistically significant, yet the RE estimator and AB estimator do not find that loan to deposit ratio significantly affect the NPLs. It indicates that high ratio of loan to deposit in the past three months also increases the current realization of NPLs. This finding supports the previous study by Dimitrios et al. (2016) that took a sample of banks in the euro-area banking system. In addition, AB GMM estimator finds the return on assets (ROA) is negative and statistically significant. As suggested by Dimitrios et al. (2016), the negative coefficient of ROA variable reflects a more efficient of the bank’s management in converting assets into profits leading to a lower NPLs. Moreover, banks that have high profit can allocate its profits to write-off the NPLs.
c. Robustness tests
21 Table 7. Robustness checks: System GMM
System GMM (1) (2) (3) (4) ∆NPL(-1) -0.289* -0.17 -0.274* -0.111 (-2.21) (-1.17) (-2.08) (-1.90) GGDP(-1) -3.211** -3.695** -3.322** -3.236* (-2.83) (-3.22) (-2.91) (-2.33) GREER(-1) 0.507 0.455 0.0634 0.248 (0.86) (0.73) (0.09) (0.19) INF(-1) -2.209 -2.664 -2.168 -1.421 (-1.13) (-1.41) (-1.24) (-0.92) CIR(-1) 0.0308 (1.82) LDR(-1) 0.0189* (2.16) ROA(-1) -0.279* (-2.08) SOB -0.0404 (-1.25) TBK -0.0797* (-2.17) constant -2.462 -1.476* 0.618* 0.155*** (-1.74) (-2.03) -2.55 -4.47 No. of Obs. 747 747 747 747 No. of Groups 41 41 41 No. of instruments . 40 40 26 F 5.876 4.204 5.379 7.453 [0.000] [0.004] [0.001] [0.000] AR1-test -2.87 -2.87 -2.82 -4.4 [0.004] [0.004] [0.005] [0.000] AR2-test -1.86 -1.21 -1.89 -1.66 [0.063] [0.226] [0.059] [0.096] Hansen test 36.64 32.68 35.65 22.53 [0.347] [0.532] [0.391] [0.259]
Notes: Dependent variable is ∆NPL. Robust t-statistics are in parentheses. The number of stars (*) denotes significance level: * p<0.05, ** p<0.01, *** p<0.001. The model was estimated by one step system GMM estimator. For each regression are presented the number of observations (No. of Obs.), a number of groups (No. of the group), and the number of the instrument (No. of the instrument). The F-test presents the statistics and respective p-values (in square brackets) for the test to the presence of fixed effects. AR1 and AR2 tests are the Arellano–Bond tests for first and second-order autocorrelation in first-differenced errors; The statistics and p-values (in square brackets) for the Hansen-test of overidentifying restrictions are also reported for the system GMM estimations.
22
instrument. A collapsing method is applied to limit the number of instruments to keep the number of instrument lower than the number of groups in GMM specifications4.
The results of system GMM are presented in Table 7. The first lag of NPLs (NPL) is negative and statistically significant. The estimation result by the system GMM confirms the finding by FE, RE, and AB GMM models. The NPLs that have lasted for three months will be either written off or repaid off by liquidating the collateral. The first lag of the real GDP growth (GGDP) and return on assets (ROA) are negative and statistically significant, whereas the first lag of loan to deposit ratio (LDR) is positive and statistically significant. The result of system GMM confirms that real GDP growth, return on assets (ROA), and loan to deposit (LDR) remain statistically significant.
As an another robust test, this study use a state-owned bank (SOB) dummy to investigate the impact of state-owned banks and also listed bank (TBK) dummy to consider the association between listed banks and NPLs as suggested by Bashir et al. (2017). The state-owned banks (SOB) imply that the shares of the Indonesian government in those banks are more than 50%. The subsidiary of a state-owned bank is not considered as the state-owned bank. Listed banks (TBK) mean that banks that are listed on stock market so that public can own a share of these banks through the stock market.
The empirical results show that both SOB and TBK dummy are negative, but only the TBK dummy variable is statistically significant. The negative coefficient indicates the state-owned banks and listed banks have lower NPLs than other banks. It reflects that the state-owned banks and listed banks have a better control and monitor regarding business activities such as loan expansion. Our finding differs from that of Bashir et al. (2017) who found that the Chinese state-owned banks significantly have higher NPLs than other banks. However, the negative coefficient of TBK dummy variable support to that of Bashir et al. (2017) who also found that the Chinese listed banks have a negative relationship on NPLs.
The result of system GMM estimator is consistent because there is a first-order serial correlation, but there is no second-order serial correlation in the residual. The Hansen test confirms the validity of the instruments used in this analysis because it strongly rejects the null hypothesis.
V. Conclusions
This study employs static and dynamic panel data model in examining the relationship between macroeconomic, bank-specific variables, and loan quality of Indonesian commercial banks. NPLs are used in this study as the measurement of loan quality. This study considers forty-one commercial banks which consist of four state-owned commercial banks and thirty-seven foreign exchange commercial banks in the period of first quarter 2011 to the second quarter of 2016.
4
23
The empirical results conclude that the real GDP growth is the only macroeconomic factors that significantly affect NPLs. All bank-specific variables are statistically significant influencing the NPLs. This finding indicates that a lower economic growth reduces the repayment capacity of the borrowers. As a result, the borrowers are not able to service all loan obligations which lead to higher NPLs. On other hands, the bank management quality in controlling expenses, expanding credit, and converting assets into profit significantly affect NPLs. Low-cost efficiency and high loan to deposit ratio in the past three months lead to an increase of the contemporaneous NPLs. The more efficient a bank’s manager to convert its assets into profit leads to a lower NPLs because the profits can be used to write-off the NPLs. In term of policy implication, the slowdown of Indonesia economic growth has an enormous impact on a rise of NPLs. Hence banks have to carefully evaluate the feasibility of every project investment to minimize the credit risk in recession economic phase. The credit expansion is suggested to the high productive sector and closely monitors the borrowers’ business activities after the loan disbursement to maintain the quality of earning assets. The intermediary financial regulator can set a regulation to restrain the loan expansion during recession economic period a view to preventing loss due to the credit risk. In addition, banks are suggested to enhance the efficiency in controlling operating costs and converting assets into profits to improve the loan quality.
This study provides some suggestions for future research. First, the prospective research is suggested to include microeconomic shocks such as shock on firm and household level that affect the decision or the inability of borrowers to serve their debts. For instance, the future research takes a sample of the firms in particular industry to examine the impact of the shock at the firm level toward NPLs. Second, it is suggested including the non foreign exchange commercial banks and rural development banks in the sample.
24
Appendix
Table A.1. Bank sample State Owned & Listed Banks
Listed Banks Non Listed Banks
PT BANK MANDIRI (PERSERO), Tbk
PT BANK ARTHA GRAHA INTERNASIONAL, Tbk
PT BANK ANTARDAERAH PT BANK NEGARA
INDONESIA (PERSERO), Tbk
PT BANK BUKOPIN, Tbk PT BANK GANESHA
PT BANK RAKYAT INDONESIA (PERSERO), Tbk
PT BANK BUMI ARTA, Tbk PT BANK HSBC INDONESIA
PT BANK TABUNGAN NEGARA (PERSERO), Tbk
PT BANK CAPITAL INDONESIA, Tbk
PT BANK ICBC INDONESIA PT BANK CENTRAL ASIA, Tbk PT BANK INDEX SELINDO PT BANK CHINA
CONSTRUCTION BANK INDONESIA, Tbk
PT BANK KEB HANA INDONESIA
PT BANK CIMB NIAGA, TBK PT BANK MASPION INDONESIA
PT BANK DANAMON INDONESIA TBK
PT BANK MESTIKA DHARMA
PT BANK JTRUST INDONESIA, TBK PT BANK MULTIARTA SENTOSA PT BANK MAYAPADA INTERNATIONAL, Tbk PT BANK NATIONALNOBU PT BANK MAYBANK INDONESIA, Tbk PT BANK RABOBANK INTERNATIONAL INDONESIA
PT BANK MEGA, Tbk PT BANK SBI INDONESIA PT BANK MNC INTERNASIONAL, Tbk PT BANK SHINHAN INDONESIA PT BANK NUSANTARA PARAHYANGAN,Tbk PT BANK SINARMAS PT BANK OCBC NISP, TBK PT BANK UOB INDONESIA PT BANK PERMATA, Tbk PT BANK QNB INDONESIA, Tbk PT BANK TABUNGAN PENSIUNAN NASIONAL, Tbk PT BANK VICTORIA INTERNATIONAL, Tbk PT BANK WOORI SAUDARA INDONESIA 1906, Tbk PT BRI AGRONIAGA, Tbk
PT PAN INDONESIA BANK,
Tbk
25
References
Aaron, M., Armstrong, J., & Zelmer, M. (2007). An overview of risk management at Canadian banks. Bank of Canada Financial System Review, 39–47.
Abid, L., Ouertani, M. N., & Zouari-Ghorbel, S. (2014). Macroeconomic and Bank-specific Determinants of Household’s Non-performing Loans in Tunisia: A Dynamic Panel Data.
Procedia Economics and Finance, 13(December 2013), 58–68. https://doi.org/10.1016/S2212-5671(14)00430-4
Ali, A., & Daly, K. (2010). Macroeconomic determinants of credit risk: Recent evidence from a cross country study. International Review of Financial Analysis, 19(3), 165–171. https://doi.org/10.1016/j.irfa.2010.03.001
Allen, F., & Gale, D. (2000). Financial Contagion. Journal of Political Economy, 108(1), 1– 33. https://doi.org/10.1086/262109
Allen, F., & Santomero, A. M. (2001). What do financial intermediaries do? Journal of
Banking and Finance TA -, 25(2), 271–294.
Arellano, M., & Bond, S. (1991). Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations. The Review of Economic Studies,
58(2), 277. https://doi.org/10.2307/2297968
Arellano, M., & Bover, O. (1995). Another Look at the Instrumental Variable Estimation of Error-Component Models. Journal of Econometrics, 68(1), 29–52. Retrieved from http://www.cemfi.es/~arellano/arellano-bover-1995.pdf
Ashley, R., & Sun, X. (2016). Subset-Continuous-Updating GMM Estimators for Dynamic
Panel Data Models. Econometrics, 4(4), 47.
https://doi.org/10.3390/econometrics4040047
Bank Indonesia. (2011). Indonesian Banking Statistics. Bank Indonesia (Vol. 10).
Bashir, U., Yu, Y., Hussain, M., Wang, X., & Ali, A. (2017). Do banking system transparency and competition affect nonperforming loans in the Chinese banking sector?
Applied Economics Letters, 0(0), 1–7. https://doi.org/10.1080/13504851.2017.1305082
Beck, R., Jakubik, P., & Piloiu, A. (2013). Non-performing loans: What matters in addition to the economic cycle? European Central Bank Working Paper Series, (1515), 34. https://doi.org/QB-AR-13-012-EN-N
Berger, A. N., & DeYoung, R. (1997). Problem loans and cost efficiency in commercial banks. Journal of Banking and Finance, 21(6), 849–870. https://doi.org/10.1016/S0378-4266(97)00003-4
Blundell, R., & Bond, S. (1998). Initial conditions and moment restrictions in dynamic panel data models. Journal of Econometrics, 87(1), 115–143. https://doi.org/10.1016/S0304-4076(98)00009-8
26 https://doi.org/10.1016/j.econmod.2013.01.027
Clair, R. T. (1992). Loan growth and loan quality: some preliminary evidence from Texas banks. Economic Review-Federal Reserve Bank of Dallas, 9.
De Haan, J., Oosterloo, S., & Schoenmaker, D. (2009). European Financial Markets and
Institutions (First Edit). Cambridge, UK: Cambridge University Press.
Dimitrios, A., Helen, L., & Mike, T. (2016). Determinants of non-performing loans: Evidence from Euro-area countries. Finance Research Letters, 18, 116–119. https://doi.org/10.1016/j.frl.2016.04.008
Espinoza, R., & Prasad, A. (2010). Nonperforming Loans in the GCC Banking System and
their Macroeconomic Effects. IMF Working Papers, 25.
https://doi.org/10.5089/9781455208890.001
Festić, M., Kavkler, A., & Repina, S. (2011). The macroeconomic sources of systemic risk in the banking sectors of five new EU member states. Journal of Banking and Finance,
35(2), 310–322. https://doi.org/10.1016/j.jbankfin.2010.08.007
Filip, B. F. (2015). The Quality of Bank Loans within the Framework of Globalization.
Procedia Economics and Finance, 20(15), 208–217.
https://doi.org/10.1016/S2212-5671(15)00067-2
Gerlach, S., Peng, W., & Shu, C. (2005). Macroeconomic conditions and banking performance in Hong Kong SAR: a panel data study. Bank for International Settlements
Press & Communications, (22), 481. Retrieved from http://www.bis.org/publ/bppdf/bispap22x.pdf
Hill, R. C., Griffiths, W. E., & Lim, G. C. (2012). Principles of econometrics (4th ed.). John Wiley & Sons Inc.
IMF, I. M. F. (2006). What Use Can Be Made of the Specific FSIs? In Financial soundness
indicators: compilation guide (pp. 155–160). Retrieved from http://www.imf.org
Louzis, D. P., Vouldis, A. T., & Metaxas, V. L. (2012). Macroeconomic and bank-specific determinants of non-performing loans in Greece: A comparative study of mortgage, business and consumer loan portfolios. Journal of Banking and Finance, 36(4), 1012– 1027. https://doi.org/10.1016/j.jbankfin.2011.10.012
Messai, A. S., & Jouini, F. (2013). Micro and Macro Determinants of Non-performing Loans.
International Journal of Economics and Financial Issues, 3(4), 852–860.
https://doi.org/10.1300/J079v27n04_02
Mileva, E. (2007). Using Arellano-Bond dynamic panel GMM estimators in Stata.
Economics Department, Fordham University, 1–10.
Nickell, S. (1981). Biases in dynamic panel models with fixed effects. Econometrica. https://doi.org/10.2307/1911408
27 115. https://doi.org/10.1016/j.fbj.2016.05.004
Otoritas Jasa Keuangan. (2016). Indonesian Banking Statistics (Vol.14 No.). Otoritas Jasa Keuangan.
Roodman, D. (2009). The Stata Journal. Stata Journal, 9(1), 86–136. https://doi.org/The Stata Journal
Shen, C. H., & Lin, C. Y. (2012). Why government banks underperform: A political interference view. Journal of Financial Intermediation, 21(2), 181–202. https://doi.org/10.1016/j.jfi.2011.06.003
Shin, S., Min, H., & Mcdonald, J. A. (2014). The Market-Discipline Effects of Subordinated Debt : Enhanced US Commercial Banking-Sector Efficiency and Stability, (September), 78–95.
