1 Introduction
The business organizations measure their success based on the profit earned during the operating period. However, the success of banking sectors is assessed not only on the basis of profitability but also based on the quality of assets held by them. The main challenge for the banking sectors is to sustain in the market by maintaining profitability growth through quality assets. The asset quality is, however, degraded due to the prevalence of non-performing loans (NPLs). The rapid increase in non-performing loans introduces the chance of failure of banking sectors as well as creates unemployment in the economy
[1]. Hence, minimizing non-performing loans is necessary for a sound banking system, which in turn is essential for an overall financial and economic stability.
The banking sector is an indispensable part of the financial system because it promotes development plans by channelizing funds for the productive purpose. The past studies on the Asian financial crisis and the recent 2008 global financial crisis have shown that most banking failures or crises are caused by NPLs. Some Asian economies that have the market power in America and Europe were severely affected by the global financial crisis with unexpected force than predicted. After those crises, NPLs have increasingly gained international attention in the developing and developed markets. Particularly, since the 1990s, many of the researchers and the policy makers have become interested in studying the NPLs due to increased financial instability in the global economy. When financial vulnerabilities were examined, the NPLs were found to be the prime cause of bank's failure and financial crises both in the developing and the developed countries
[2]. Hence, we choose the NPL ratio (NPL to total loans) to measure the assets quality and the credit risk of the banking system in this study. Similarly, the macroeconomic indicators are chosen as the potential determinants of NPL because they directly reflect the economic stability, which in turn, is linked with the sound financial system of a country.
The NPL is an economic problem, which discourages the economic activity by deteriorating the assets quality and destabilizing the banking system. The banking sustainability and the economic growth are impossible in the absence of regular and effective management of the NPLs. However, it is difficult to efficiently manage NPLs without learning what determines it. Following the global economic crisis in 2007, several studies have attempted to assess the relationship between various economic indicators and NPL. However, such studies are largely lacking in the Asian banking industries. Moreover, to the best of our knowledge, no previous studies have assessed the macroeconomic determinants of NPLs based on the income levels of countries. This study presents the first empirical study on the macroeconomic determinants of NPL in 19 Asian countries ranked with their income levels. The non-performing loans in this study are categorized based on the definitions by Alton and Hazen
[3] and Fofack
[4]. According to Alton and Hazen
[3], the unpaid obligation of principal and interest within 90 days or beyond 90 days by the borrowers is categorized as non-performing loans (NPLs). Similarly, Fofack
[4] defined NPLs as those loans, which are not generating income for at least ninety days.
The rest of paper is organized as follows. Section 2 provides a brief overview of the macroeconomic determinants of NPL. Section 3 describes the data sources, introduces the study variables and the hypotheses, and outlines the econometric framework. Section 4 presents the estimation results and discusses the findings. The final section summarizes the study.
2 Literature Review
This section provides a brief overview on the determinants of NPL. These determinants are broadly categorized as the firm level variables, variables of legal issues and macroeconomic variables. Different studies have used different variables based on their statistical research design. However, most cross-country studies have included macroeconomic variables to find the cause of NPLs growth. Here, we discuss the findings of contemporary literature and use them as a basis of selecting explanatory variables for this study.
Louzis, Vouldis
[5] and Salas and Saurina
[6] applied dynamic panel data to investigate the factor affecting NPLs in the banking system. They concluded that NPLs were caused primarily by macro-economic factors like gross domestic product (GDP), unemployment rate, interest rates and management quality. Similarly, Klein
[7] found that macroeconomic factors like GDP growth rate, unemployment rate and inflation strongly affects NPL.
With the help of quarterly data over 2004–2012, Jakubík and Reininger
[8] analysed the determinants of NPLs in 9 central, eastern, south-eastern European countries (Bulgaria, Croatia, Czech Republic, Hungary, Poland, Romania, Russia, Slovakia and Ukraine). GMM estimation technique was applied to show the relationship between macroeconomic factors and NPLs. They found that the nation's exchange rate, the private credit to GDP ratio and one period lagged NPLs increase NPLs. However, they found that the real GDP growth and the national stock price index have a negative association with non-performing loans.
Using unbalanced panel dataset of 141 developing countries (low and middle-income countries) over 2000–2011, Ebeke, Loko
[9] investigated the effect of remittance on credit quality. They applied ordinary least square estimation technique taking NPLs as a dependent variable and found a negative correlation with remittance.
By taking the sample of 75 countries, Roland, Petr
[10] studied the impact of macroeconomic factors on NPLs over an eleven-year period (2000–2010) using the GMM estimation technique. The results showed that the lending interest rate, the exchange rate, the share prices and the GDP growth rate significantly affect the non-performing loans.
karica
[11] used the 6 years (2007–2012) quarterly data of Central and East European countries to explore the macroeconomic determinants of NPLs. The empirical results concluded that the unemployment and inflation rates have positive roles on the NPLs ratio. The findings suggest that increase in the real GDP growth is essential to reduce the NPLs growth in the financial system of European countries.
To study the impact of macroeconomic conditions on NPLs in Indian banking sectors, Rajan and Dhal
[12] employed the panel regression method. They found that the GDP growth rate, maturity, cost and terms of credit, bank's size and credit orientation significantly affected the NPLs. Bucur and Dragomirescu
[13] explored the interrelationship between macroeconomic conditions (real GDP growth rate, inflation rate, interest rate, money supply, foreign exchange rate fluctuation and unemployment rate) and credit risk in the Romanian banking system during 2008–2013 with the help of multidimensional statistical analysis. The regression analysis confirmed that the exchange rate fluctuation negatively, while the unemployment rate positively affected the NPL growth. However, the GDP growth rate was not significantly correlated with the credit risk.
Buncic and Melecky
[14] estimated panel regressions using annual data for 54 countries ranging from 1994–2004 to evaluate the determinants of NPLs. The GMM estimation technique was used to estimate the selected explanatory variables (lagged NPLs ratio, real GDP growth, inflation rate, real interest rate, changes in the nominal US dollar exchange rate for each country). They found a significant effect of all explanatory variables on NPLs except changes in nominal US dollar exchange rate for each country.
De Bock and Demyanets
[15] employed the dynamic panel regressions and structural panel vector autoregressive regression (VAR) to identify the factors affecting NPLs in 25 emerging market economies during 1996–2010. The authors concluded that the real GDP contraction, the currency depreciation against US dollar, the weaker term of trade, and the outflow of debt-creating capital (portfolio debt and bank loans) are the main factors that affect non-performing loans.
By employing the correlation and regression tests for empirical analysis, Saba, Kouser
[16] found a significant effect of real GDP per capita, inflation rate, and total loans on NPLs in US banking sector during 1985–2010. Nkusu
[17] analysed the link between non-performing loans and macroeconomic factors using panel regression and panel vector autoregressive (PVAR) model. A thorough study of the 26 advanced economies during the period from 1998 to 2009 confirmed that the adverse macroeconomic determinants are the major causes of higher NPLs.
Touny and Shehab
[18] identified the macroeconomic determinants of nonperforming loans of 9 Arab countries (Egypt, Morocco, Tunisia, Jordan, Lebanon, Saudi Arabia, Kuwait, Oman, and United Arab Emirates) during 2000–2012 using the dynamic panel data approach (GMM estimation). The findings suggest that the inflation rate, the government spending, and the GDP growth rate have a negative impact on NPLs, whereas an increase in aggregate debt burden has a significant positive relationship with the NPL level.
Castro
[19] analysed the link between macroeconomic indicators and credit risk in Greece, Ireland, Portugal, Spain, and Italy (GIPSI) over the period 1997q1–2011q3 employing pooled fixed effect, random effect and dynamic panel data method (GMM estimation). This study indicated that the decrease in GDP growth rate, share prices, housing price indices and rise in unemployment rate, interest rate, real exchange rate and credit growth significantly increase the non-performing loans in GIPSI countries.
Tanasković and Jandrić
[20] focused on macroeconomic and institutional factors to find the determinants of non-performing loans in CEEC and SEE countries during 2006–2013. They showed that a high GDP and a developed financial market significantly reduce the NPL level. Similarly, the foreign currency loan ratio and the level of exchange rate significantly increase NPL rate. The inflation is, however, found to be statistically insignificant.
Using GMM estimator, Islam and Nishiyama
[21] empirically studied the bank specific, industry specific, and macroeconomic variables to find its effect on non-performing loans in South Asian countries (Bangladesh, India, Nepal, and Pakistan) over 1997–2012. They found moral hazard problems, adverse selection of borrower, cost inefficiency, income diversification, bank size, inflation rate and GDP growth rate as the potential determinants of default risk in the banking industry during their study period.
3 Methodology
3.1 Variables and Data Sources
All sorts of investments and credit risk of the financial system are mainly affected by different macroeconomic factors. Frequent Changes in economic policies (changes in monetary and tax policies, economic legislation changes, and import and export trade policy), political instability, and slow economic growth are the main causes of credit risk. However, these factors are difficult to measure quantitatively and therefore, in this study we have included only those macroeconomic factors that could be quantitatively measured (e.g., inflation rate, remittance, official exchange rate, GDP per capita, etc.) to determine the potential determinants of credit risk in Asian countries.
The study consists of a balanced panel of 19 Asian countries covering the period from 1998 to 2015. These countries were further divided into high-income economies (7 countries), middle-income economies (8 countries), and low-income economies (4 countries) using gross national income per capita (GNIPC) based on the criteria set by World Bank (see Figure 1). The data consists of South Asian (Bangladesh, India, and Pakistan), West Asian (Armenia, Israel, Jordan, Kuwait, Oman, and Saudi Arabia), South East Asian (Indonesia, Malaysia, Philippines, Singapore, and Thailand), East Asian (China, Japan, and South Korea), and Central Asian (Kazakhstan and Kyrgyzstan) countries as the samples in this study. We could not include other Asian countries due to limited availability of data over the study period. The independent variables were selected based on existing studies (see above), which reflected the general state of economy, economic stability, monetary policy, capital formation, and debt burden of individuals and business firms.
Figure 1 The 19 Asian countries sampled in this study. These countries have been categorized into high-income, middle-income and low-income countries based on their average gross national income per capita during the study period (1998–2015) |
Full size|PPT slide
The data for this study is obtained from the World Bank and the International Monetary Fund (IMF). The variables include non-performing loans to gross loans (%), GDP per capita (current US＄), inflation rate (%), unemployment rate (%), foreign direct investments (net inflows, % of GDP), official exchange rate local cost unit per USD, personal remittance received (% of GDP), and domestic credit to private sectors by bank (% of GDP). In this paper, credit risk is measured as the ratio between bank's NPLs and the total gross loans. In order to provide better interpretation of the coefficient results, logarithm form of all the study variables are used. Figure 2 shows the trend of non-performing loans in the 19 Asian countries and Figure 3 shows the average values of all variables during the study period. From these figures it is clear that high non-performing rate is mostly led by low-income countries where Bangladesh has the highest average (19.54%) followed by Pakistan (14.54%). On the other hand, Singapore (3.15%) and South Korea (2.44%), which are categorized as high-income economies (Figure 1), have the lowest non-performing rates of all the Asian countries. This indicates that the high-income countries usually have low NPL rates, while the low-income countries, in general, have high NPL rates.
Figure 2 Trend of non-performing loans in 19 Asian countries during the study period (1998–2015) |
Full size|PPT slide
Figure 3 Average values of non-performing loans and other study variables of 19 countries. For ease of comparison, the values of these variables have been separately shown for high-income, middle-income and low-income economies |
Full size|PPT slide
3.2 Data Description and Hypotheses
Based on some existing studies and our preliminary analysis (see Figure 4), we propose following hypotheses for our study.
Figure 4 Scatterplots showing the relationship between non-performing loans and macroeconomic variables of the 19 Asian countries. Blue lines in each plot represent trend-lines and the grey areas around these lines represent 95% confidence intervals. The pairwise correlation coefficient and -value are given for each relation |
Full size|PPT slide
Hypothesis 1 GDP per capita (GDPPC) has negative effect on NPL.
The general level of the economy is represented by GDP per capita (GDPPC). It describes the level of economic activities of an individual and firms. Since GDP per capita increases the income level, it may contribute to reduce the default loans. Hence, we assume that the GDP per capita has a negative impact on the growth of NPLs.
Hypothesis 2 Increase in unemployment rate (UR) increases NPL.
The level of the unemployment rate (UR) describes the level of income and purchasing capacity of an individual. The fall in unemployment rate indicates more employed people and an increased purchasing power of the households. It also reflects the increased economic transactions and the higher tendency of borrowers to repay their debts. Hence, we assume that the unemployment rate has a positive impact on the growth of NPLs.
Hypothesis 3 Foreign direct investment (FDI) reduces NPL level.
The foreign direct investment (FDI) inflow, represented as the percentage of GDP, reflects the foreign participation in the capital formation for a higher productivity. The increased productivity boosts up a firm's ability to repay loans by generating higher revenues. Borén
[22] suggested that the foreign investment contributes to create a low interest rate environment in the economy, which makes servicing debt on time and hence reduces the cost of the loan. Therefore, we assume that the foreign direct investment improves the assets quality of banks and reduces NPLs.
Hypothesis 4 Inflation rate (IR) has positive effect on NPL.
The inflation rate (IR) is regarded as a significant macroeconomic determinant of NPLs because it represents economic flexibility. The volatility of inflation rate is a sign of economic instability. The rise in inflation decreases the purchasing power of money and the real value of an income, which weakens the debt servicing and hence increases the NPL level. More specifically, a higher inflation leads to increased lending rate and hinders the payment of debt on time. Therefore, as shown by Abid, Ouertani
[23] and Klein
[7], we assume a positive link between inflation and NPL in this study.
Hypothesis 5 Domestic credit to private sectors (DCPS) reduces NPL level.
The domestic credit to private sectors (DCPS) by commercial banks is used to examine the aggregate impact of private sectors on the assets quality of banking sectors. It reflects the allocation of financial resources like loans and securities to household and business in the economy. Based on an empirical study of Glen and Mondragón-Vélez
[24], the private companies increase economic activities by efficiently mobilizing the credits provided by commercial banks. The increased financing to private companies offer both challenges and opportunities for the economic growth. The efficient mobilization of debt by private sectors generally improves the living and credit standard of borrowers through high economic growth. Thus, we assume that the development of private sectors plays a significant role in accelerating the economic growth as well as improving the debt payment capacity of borrowers.
Hypothesis 6 Official exchange rate (OER) has positive effect on NPL.
The exchange rate differs among countries and it depends on their overall economic conditions. In order to measure the value of local currency, we used the exchange rate between local cost per unit and U.S. dollars for all Asian countries assuming that the foreign currency loans are mostly the U.S. dollars denominated loans. To capture the impact of exchange rate dynamics on the asset quality for different group of Asian countries, we assume exchange rates are directly associated with the default loans. This is based on the findings of Akinlo and Emmanuel
[25] for Nigeria, Jakubík and Reininger
[8] for central eastern and south-eastern European countries and Gremi
[26] for Albania.
Hypothesis 7 Remittance (RM) has positive effect on NPL.
Many studies have shown that remittance could increase bank's loanable fund causing higher credit growth. A high credit growth in the market can be associated with a higher credit risk in the financial system. Hence, hypothetically it can be assumed that there is a positive association between remittance and NPL. In contrast, Aggarwal, Demirg-Kunt
[27] and Ebeke, Loko
[9] have shown a negative association between remittance and NPL in their studies, which indicated that the increase in remittance helps to create a sound banking system.
3.3 Econometric Framework
The dynamic panel data models include the lagged dependent variable as regressors. The lagged dependent variable is correlated with an idiosyncratic error in static panel model estimators, such as fixed effects and random effects estimators, thus giving inconsistent results. Moreover, the endogeneity concern is not solved well in static panel estimators. Therefore, the Generalized Method of Moments (GMM) propounded by Arellano and Bond
[28] is employed by most studies to remove the dynamic panel bias and to obtain consistent results. Both difference GMM and system GMM are designed to remove dynamic panel bias
[28-30]. The system GMM designed by Blundell and Bond
[31] and Arellano and Bover
[30] is an extended estimator of Arellano and Bond
[28].
Since the system GMM uses more instruments than the difference GMM, it may not be appropriate to use system GMM with our dataset, which includes only 19 countries. Therefore, in order to prevent over identifying instrumental variables and to validate the available instruments, the difference GMM estimation developed by Arellano and Bond
[28] was applied in this study. The difference GMM also avoids the endogeneity problem introduced by instruments and produces consistent results. It captures the country-specific effects, unobserved variables and persistence of dependent variable and controls the heterogeneity biases and omitted errors. The dynamic panel data (DPD) models estimated by Generalized Method of Moments (GMM) have become an important technique in the empirical research involving country level panel data. An important baseline of Arellano and Bond
[28] is the first order autoregressive (AR (1)) model with unobserved individual effects.
3.4 Econometric Model Specification
Here, we employed the dynamic panel model used by Touny and Shehab
[18], Espinoza and Prasad
[32], Louzis, Vouldis
[5] and Ghosh
[33], to explore the determinants of NPLs in Asian banking system. The baseline model is captured by the following general equation:
where NPL denotes the logit transformation of NPL ratio for country at time ; NPL is the NPL ratio of country at time ; represents the matrix of macroeconomic factors of country at time . is the constant term; and are the corresponding coefficient vectors. The coefficient of lagged dependent variable is expected to be positive and less than unity. is the unobserved individual (country specific) effect and is the idiosyncratic error term. has the standard error component structure. We assume that the innovations have the following characteristics:
In Equation (1), the fixed effects are correlated with the regressors. In order to remove such time in-variant country specific effect, a transformation like first differencing is the option to obtain valid moment conditions. First differencing removes the individual effect and omitted-variable bias from the equation. The following expression represents the first difference:
After the first differencing,
becomes correlated with the error term (
in Equation (3). The instrumental variable estimator of Anderson and Hsiao
[34] and Generalized Method of Moments (GMM) of Arellano and Bond
[28] are the widely followed measures to remove such problem
[35]. Arellano and Bond
[28] is a tricky econometric estimator, which takes lagged value of independent variables as instrumental variable within the equation assuming that the series of independent variables are predetermined and there is no contemporaneous correlation.
In the above specification, the lagged dependent variable captures the relationship between the past and the present values. Furthermore, the above specification may be subjected to fixed individual effects, autocorrelation within individuals and unobservable differences between countries. According to Salas and Saurina
[6], the non-performing loan ratio is closely related to the previous period rate. Since NPLs are not immediately written down from a bank's balance sheet, they are assumed to be auto-regressive and bear positive coefficient.
To avoid multicollinearity among the study variables, we build 7 models in total, including two models for high-income group, two models for middle-income group and three models for low-income group, to determine the macroeconomic factors affecting non-performing loans in Asian countries. The models are free from multicollinearity problems and provide consistent results.
Prior to analyses, the unit root test was conducted in order to avoid spurious regression coefficients and misleading results. Studies have shown that Fisher unit root test performs better compared to other panel unit root tests
[36]. Therefore, the preference was given to the Fisher-ADF unit root test to assess the level of integration with the null hypothesis of non-stationarity against the stationary alternative. The simplest form of Autoregressive (AR) model for testing unit root is
where , . means identically and independently distributed with a zero mean and a constant variance. The null hypothesis : is tested against an alternative hypothesis : . If , Equation (4) becomes a random walk model without drift, which is known as a non-stationary stochastic process.
In order to test the absence of autocorrelation assumptions, Arellano-Bond first order serial correlation was used. According to Arellano and Bond
[28], a consistent GMM estimator should reject the Arellano-Bond first-order serial correlation and should not reject the second order serial correlation.
4 Estimation Results
In this section, the empirical results are presented based on one step GMM estimator of Arellano and Bond
[28]. Estimations were conducted in multiple models for different groups based on variance inflation factors to avoid multi-collinearity issue.
Tables 1 to
3 reports the estimation results of high-income, middle-income and low-income Asian countries, respectively.
Table 1 GMM estimation results for high-income countries |
| Variables | Model 1 | Model 2 |
| Lagged nonperforming loans | 0.6748 *** | 0.5428 *** |
| | (5.8625) | (5.1525) |
| Foreign direct investment inflow | 0.0650 | 0.0047 |
| | (0.7074) | (0.0707) |
| Domestic credit to private sectors | 0.1623 | 0.06208 |
| | (0.8974) | (0.3659) |
| Inflations | -0.0961 * | -0.0770 * |
| | (-1.6839) | (-1.8028) |
| Unemployment rate | 0.0622 | |
| | (0.3156) | |
| Remittance | 0.7795 * | 1.0671 *** |
| | (1.6594) | (2.9140) |
| Official exchange rate (LCU per US Dollar) | | -0.7736 |
| | | (-1.5125) |
| Gross domestic product per capita | | -0.4866 *** |
| | | (-3.4635) |
| AB test for autocorrelation (AR2) | 0.1960 | 0.1796 |
| | (-1.2929) | (-1.3421) |
| No observations | 71 | 74 |
| Note: -statistics are reported in parenthesis. , , . |
Table 2 GMM estimation results for middle-income countries |
| Variables | Model 3 | Model 4 |
| Lagged nonperforming loans | 0.8795 *** | 0.5972 *** |
| | (9.5012) | (2.9491) |
| Foreign direct investment inflow | | -0.0573 |
| | | (-1.1035) |
| Domestic credit to private sectors | 0.3133 *** | |
| | (3.6907) | |
| Inflations | -0.0728 *** | -0.0729 *** |
| | (-6.0168) | (-2.7102) |
| Unemployment rate | 0.0048 | |
| | (0.0597) | |
| Remittance | -0.0952 | |
| | (-0.7490) | |
| Official exchange rate (LCU per US Dollar) | | 0.0457 |
| | | (0.1072) |
| Gross domestic product per capita | | -0.1098 |
| | | (-0.5365) |
| AB test for autocorrelation (AR2) | 0.0798 | 0.0504 |
| | (-1.7517) | (-1.9567) |
| No observations | 120 | 122 |
| Note: -statistics are reported in parenthesis. * , ** , *** . |
Table 3 GMM estimation results for low-income countries |
| Variables | Model 5 | Model 6 | Model 7 |
| Lagged nonperforming loans | 0.3622 * | 0.3914 * | 0.8036 *** |
| | (1.6872) | (1.8473) | (8.9641) |
| Foreign direct investment inflow | | -0.0412 | -0.0964 |
| | | (-0.4232) | (-1.2447) |
| Domestic credit to private sectors | -0.2392 | | |
| | (-1.2557) | | |
| Inflations | 0.6385 | 0.0766 | 0.0939** |
| | (1.1140) | (1.3992) | (2.0285) |
| Unemployment rate | 0.4130 * | | 0.4429 * |
| | (1.9493) | | (1.6836) |
| Remittance | -0.1727 * | | |
| | (-1.7224) | | |
| Official exchange rate (LCU per US Dollar) | 0.3534 | 0.7650 *** | 0.2216 ** |
| | (1.5579) | (2.6144) | (2.2276) |
| Gross domestic product per capita | | -0.5888*** | |
| | | (-2.6699) | |
| AB test for autocorrelation (AR2) | 0.6539 | 0.3497 | 0.7132 |
| | (0.4484) | (-0.9352) | (-0.3676) |
| No observations | 64 | 64 | 64 |
| Note: -statistics are reported in parenthesis. * , ** , *** . |
The official exchange rate was positively related to the growth of NPL ratio and significant in model 6 and 7 of low-income countries (Table 3) but insignificant in middle-income countries (Table 2). This indicates that the depreciation of domestic currency tend to increase NPLs in low-income countries. The result also suggests that the middle-income countries are not using domestic currency as credit placements. In the model of high-income countries (see Table 1), the coefficient of the official exchange rate was opposite to those of middle and low-income countries (see Tables 2 & 3). The estimation result showed an insignificant negative correlation between official exchange rate and non-performing loans.
In high-income (Table 1) and low-income countries (Table 3), the coefficient of GDP per capita was negative and statistically significant. This indicates that the economic growth increases the payment capacity of individuals and business firms as well as improves the assets quality and financial system of an economy. However, the GDP per capita of middle-income countries did not have a significant effect on NPLs.
The inflation rate was statistically negatively significant in all models of high-income and middle-income countries stating higher inflation made debt servicing easier by reducing the real value of the unpaid loan that caused low NPL level. These results are beyond our hypothesis and suggest that the economic stability policy is similar in these economies too.
On the other hand, the inflation rate was positively significant in model 7 of low-income countries (Table 3) at 95% significant level, which means the rise in inflation is an indication of low economic growth in low-income economy. The result suggests that the rise in inflation level increases the lending rate that causes high cost of debt and consequently high NPL level.
Similarly, the coefficients of unemployment rate were positive in all models but not statistically significant as expected. The -value of the unemployment rate was statistically reliable only in low-income countries (Table 3). The unemployment decreases the sources of individual income to consume commercial goods as well as the ability to pay loans and consequently increases NPLs.
The domestic credit to private sectors is the ratio between domestic credit to private sectors by commercial banks and GDP. The empirical results suggest that, in high-income and middle-income countries, private sectors could not mobilize the debt provided by commercial banks to improve the quality of loans during the study period. The probability of increasing default loans was positively associated with the DCPS but significantly reliable only in middle-income group. There was a negative insignificant effect in a low-income group.
The foreign direct investment showed a negative insignificant effect in middle-income and low-income groups (
Tables 2 and
3). Surprisingly, in a high-income group, a statistically unacceptable positive effect of foreign direct investment inflows on NPL growth was found. This is in contrast to the findings of Festic and Beko
[37] and Borén
[22] who found a negative impact of foreign direct investment on the NPL ratio.
As hypothesized, the GMM estimation showed a positive effect of remittance on NPLs in the high-income group with statistically acceptable value based on 99% level of significance. This indicates that the remittance could expand credit level by increasing the bank's loanable fund in the market. The high volume of credit level could be associated with a low credit standard and a higher credit risk in the financial system.
On the other hand, the remittance had a negative insignificant relationship with NPLs in the middle-income countries (Table 2) and statistically significant relationship in the low-income countries (Table 3), which confirms that remittance enhances the living standard and payment capacity of the borrowers in developing countries due to stable transfers and serves as a collateral (income stabilizing effect) through its effect on the overall economic growth. This helps to increase the credit quality of borrowers and decrease NPLs. Based on the empirical results, it is evident that the impact of remittance on NPL differs from economy due to its unstable nature.
Due to the presence of lagged dependent variable as an explanatory variable, serial correlation occurs in the estimated models. The Arellano Bond test for autocorrelation was conducted to test serial correlation in the error term of the estimated models where the null hypothesis is, there is no autocorrelation. In all models, the null hypothesis of Arellano Bond was accepted, which means that the presented models are free from autocorrelation.
All the study variables were expressed in logarithmic form, while the dependent variable was transformed into logistic form. In order to avoid stochastic or deterministic trends in the estimations, we applied the Augmented Dickey-Fuller tests to find whether the series in panel regressions equations are stationary or not. Table 4 presents the results of panel unit roots test where the null hypothesis is non-stationary. The first difference of the series was taken to ensure stationarity for those variables that exhibit unit roots in their levels.
Table 4 Fisher-Augmented Dickey-Fuller unit root test results |
| Variables | High-income countries | | Middle-income countries | | Low-income countries |
| -value | statistics | -value | statistics | -value | statistics |
| NPL | 0.8658 | 8.42935 | | 0.0669 | 25.1689 | | 0.0975 | 13.4429 |
| FDI | 0.0065 | 30.5183 | 0.000 | 48.0159 | 0.2232 | 10.6359 |
| DCPS | 0.4718 | 13.7067 | 0.024 | 28.9907 | 0.8036 | 4.558 |
| IR | 0.0043 | 31.7956 | 0.000 | 62.5083 | 0.029 | 17.1043 |
| UR | 0.0028 | 33.0644 | 0.8531 | 10.2529 | 0.0176 | 18.5351 |
| OER | 0.1871 | 16.0906 | 0.8567 | 10.1862 | 0.5121 | 7.22936 |
| GDPPC | 0.0622 | 22.8838 | 0.3268 | 17.951 | 0.9823 | 1.95751 |
| RM | 0.0431 | 21.5325 | 0.3687 | 17.2637 | 0.000 | 96.1317 |
5 Discussion of Empirical Results
The estimated models of high-income and low-income countries show the negative significant relationship between NPLs level and GDP per capita (Models 2 & 6). With the increase in GDP, economic activities also increase. As a result, more revenues are gained by individuals and business institutions. This, in turn, creates a debt payment capacity environment, which significantly reduces bad loans. This is consistent with the findings of Endut, Syuhada
[38], Nkusu
[17], Espinoza and Prasad
[32], Rajan and Dhal
[12], and Shu
[39]. The effect of GDP per capita is statistically negatively insignificant in middle-income group (Model 4) which is in accordance with the result of Bucur and Dragomirescu
[13] in the Romanian banking system.
The GMM estimation result shows that the GDP per capita influences bad loan negatively suggesting that the GDP per capita growth indicates an improvement in the business performance and economic growth where a payment capacity is positively increased. The models recommend increased GDP per capita for low NPL level in Asian banking industry.
The economic stability is measured by an annual inflation rate. The empirical results of high-income (Models 1 & 2) and middle-income countries (Models 3 & 4) reveal a negative effect of inflation on NPLs, which is consistent with the findings of Shu
[39], Endut, Syuhada
[38], Vogiazas and Nikolaidou
[40] and Khemraj and Pasha
[2] who argued that higher inflation reduces money supply in the market and proportionally reduces credit expansion and the level of NPLs.
The higher inflation rate makes debt facilitation easier and increases credit growth. Due to increase in loan growth, the chances of low credit standard also increases, which creates bad loans in the financial system after a certain period. In the present study, this situation is found in low-income Asian countries (Model 7), which is consistent with the result of Rinaldi and Sanchis-Arellano
[41] and Klein
[7]. The result indicates that high inflation is an indicator of economic instability and cause of high NPLs. However, Models 5 and 6 show the insignificant effect of inflation on default loans, which is similar to the findings of Anjom and Karim
[42] and Tanasković and Jandrić
[20]. Both economic theory and empirical evidence strongly support that volatility in inflation is associated with banking instability, which leads to economic instability eventually. The above empirical results indicate that the relationship between NPLs and inflation could be either positive or negative depending upon the income level of countries.
As expected, the lagged value of NPLs showed a positive association with the current value of NPLs over the study period. The coefficient of lagged non-performing loans ranged from 0.36 to 0.87, suggesting that NPL is likely to have a prolonged effect on the banking system and economy as a whole. In contrast, Louzis, Vouldis
[5] found a negative relationship between lagged dependent variable and current dependent variables.
We found a positive relationship between the unemployment rate and non-performing loans in all models but statistically significant association only in low-income group (
Table 3, Models 5 & 7), which indicated that unemployment is negatively associated with the income level. A higher unemployment rate lowers the income level, which reduces the payment capacity of borrowers and increases the default loans. This result is consistent with those of Klein
[7], Vogiazas and Nikolaidou
[40], Nkusu
[17] and Makri, Tsagkanos
[43] who found that the unemployment rate negatively affects the reimbursing capacity of individuals, which increases debt burden and hence increases the default rate. The findings of Bofondi and Ropele
[44] in Italian banks, Gambera
[45] in American banks and Louzis, Vouldis
[5] in Greek banking system suggested that unemployment causes a loss in the production of firms and ultimately decreases their revenue and the payment capacity.
The impact of foreign direct investment on NPLs was negative but statistically insignificant in the middle-income and low-income groups (Tables 2 and 3). We supposed that the capital import increases the capital formation and the production capacity of a country, which in turn, reduces NPL level by increasing the export and employment rate. However, the empirical evidence did not support our hypothesis.
The foreign direct investment showed an insignificant positive effect on NPL in high-income countries similar to the results of Ahmad and Bashir
[46] and San
[47]. A high degree of foreign direct investment inflows significantly increases liquidity and creates additional loan supply
[48]. An increasing credit growth brings low credit standard, causing default loans to increase in banking sectors. The foreign participation on capital formation is found to have insignificant effect on NPLs in all groups, which indicates that the foreign direct investment does not seem to influence NPL.
Based on the dynamic panel data estimation, the official exchange rate showed an insignificant effect on NPLs in high-income and middle-income groups (
Tables 1 and
2) which are in accordance with the result of
karica
[11] and indicates that the share of foreign currency loans in total loans is very low in these countries. Moreover, the local currency is stable against the U.S. dollar during the study period in high- and middle-income countries. However, in low-income countries (
Table 3), a positive significant relationship between NPLs and official exchange rate is found, which is consistent with the findings of Fofack
[4], Khemraj and Pasha
[2], Roland, Petr
[10] and Tanasković and Jandrić
[20] who argued that the depreciation of local currency may result in the borrowers exposition to higher debt servicing cost against the loan currency and result in the higher NPL ratios. Our finding together with previous reports suggests that there are more lending foreign currencies to unhedged borrowers in low-income countries.
A positive significant effect of domestic credit to private sectors on NPLs is found in middle-income countries (
Table 2), which indicated that the growth in lending to private sectors lead to low credit standard and high chances of granting risky loans thereby increasing the probability of default loans. Similarly, due to credit expansion in private sectors, banks may not be able to effectively monitor the performance of their borrowers frequently, resulting in higher default loans. The estimated results are consistent with the findings of Nkusu
[17], Jakubík and Reininger
[8], Akinlo and Emmanuel
[25]. They argued that high credit to private sectors causes liquidity problems and simultaneously increases the probability of default loans. However, in high-income countries (
Table 1), the positive insignificant association is found between the domestic credit to private sectors and NPLs, consistent with the finding of De Bock and Demyanets
[15]. In contrast, Model 5 (
Table 3) of low-income countries revealed the insignificant negative effect of credit to private sectors on the NPLs. It means that the portion of the domestic credit to private sectors is not enough for economic growth.
Our empirical study provides evidence that remittance is the main driver for reducing NPLs, particularly in low-income countries (
Table 3, Model 5) as it increases the household's incomes and improves the financial intermediation. This outcome showed a crucial role of remittance to the stable financial system in developing countries like India, Bangladesh, Pakistan and Kyrgyz Republic. Consistent with our finding, San
[47] and Clichici and Colesnicova
[49] also showed an inverse relationship between remittance growth rate and NPLs growth in Albania and Republic of Moldova, arguing that remittance increases the payment ability of individual and hence reduces NPL ratio by improving household income. However, there was an insignificant effect of remittance on NPL in middle-income countries.
The findings of high-income group (Table 1) revealed that remittance increases loanable fund due to the unstable transfer of remittance, which then reduces the demand for loans. This reduces the credit standard of loans and causes high NPLs. The summarized results of remittance suggest that the countries where remittance is significantly negatively associated with the NPL level are macro economically unstable because remittance plays a greater role in causing financial stabilization in such countries. Similarly, a positive association between remittance and NPL level indicates high money supply and lack of proper investment policy for unstable income. The money supply can be controlled by the government by identifying investment areas to stabilize financial system under such condition. The -value of Arellano Bond AR (2) tests was more than 5% in all the models, which indicated the absence of serial correlation and hence confirmed the accuracy of the estimated models.
6 Conclusions
Based on the empirical results, the macroeconomic indicators, such as GDP per capita, remittance and inflation rate are the most effective determinants of NPLs in all Asian countries (high-income, middle-income and low-income economies). The study recommends that every economy should stabilize the inflation rate and foreign exchange rate to improve the assets quality of financial system. Moreover, the fiscal and monetary policy should be formulated in such a way that the entire economic growth is supported.
A deeper understanding of the determinants of NPLs is required by the central bank to build a credit risk model of the whole banking system in the changing macroeconomic environment. Our study can be a useful tool to manage credit risk in Asian banking system. This study has practical benefit in the macroeconomic analysis for improving the assets quality in Asian financial system. We found that the macroeconomic determinants of credit risk in Asian countries differ based on their income level. Therefore, economic level of a country should be carefully considered while formulating credit policy to minimize credit risks.
The government should work together with the fiscal and monetary policies to increase the GDP per capita and stabilize the lending process to private sectors, inflation, domestic currency and banking system. The findings of this study have important implications for policy-makers and credit risk experts in Asia to reduce their NPL levels and stabilize banking sectors.
This paper mainly focuses on the macroeconomic determinants of NPLs with few Asian countries, which is a limitation of this study. Bank specific factors, business environment and other institutional factors may also influence non-performing loans. Therefore, future studies involving other factors and covering the entire Asian countries over a longer period of time could provide more robust results.
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
PDF(483 KB)
Figure 1 The 19 Asian countries sampled in this study. These countries have been categorized into high-income, middle-income and low-income countries based on their average gross national income per capita during the study period (1998–2015)
Table 1 GMM estimation results for high-income countries
