Author：Yilin Xing Student number: 11377011
July 2017
Supervisor: Dr. A.C.F.J (Aerdt) Houben Co- reader: Dr. W.E. (Ward) Romp Faculty of Business and Economics
The University of Amsterdam Master Program: Economics
This document is written by Student Yilin Xing who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it.
Non- performing loan (NPL) is an important indicator of systemic risk and financial fragility. Using fixed effects and dynamic panel data approaches estimated over 2009Q1- 2016Q4 on 15 commercial banks, the current study examines the determinants of non-performing loans in the Chinese banking sector. The main objective is to find empirical evidence on the macroeconomic and bank- specific variables that have an effect on NPLs. The results show that the macroeconomic variables GDP growth, inflation, money supply and house price significantly decrease NPLs, while real interest rate noticeably increases NPLs. Among bank-specific variables, size, capitalization and loan growth generally raise NPLs, while operating efficiency is negatively related to NPLs. Besides, the previous value of problem loans also has an impact on the current level of NPLs. The findings illustrate the countercyclical nature of NPLs and underscore the importance of improving banks' operating efficiency and risk management skills.
Key Words: Non- performing loans, bank- specific determinants, macroeconomic determinants, fixed effects model, dynamic panel data estimation, financial stability.
Contents
1. Introduction……… 2
2. Literature Review………..………... 2
3. Evolution of NPLs in Chinese banking sector………...……… 5
4. Methodology and Data Description………...…... 7
4.1 The choice of variables and data description………..…..…. 7
4.1.1
Non- performing loans ratio... 74.1.2
Macroeconomic determinants of NPLs………...……….. 94.1.3
Bank- specific determinants of NPLs………...………... 104.2 Empirical Methodology...………...………. 14
4.3 Test………...…………...………... 16
5. Empirical Analysis………...………... 17
5.1 Fixed effects estimation….………... 17
5.2 Dynamic panel data estimation…...……….. 18
6. Results nd Policy Implications………...………... 23
Tables
1) Summary of variables... 132) Fixed Effects Estimation Results... 19
3)
GMM Estimation Results...l... 25Figures
4) The NPLs ratio in five bank groups of China, 2007q2- 2016q4... 65) Weighted average NPLs ratio of 15 Chinese commercial banks 2009q2- 2016q4... 8
6) Macroeconomic variables in China banking sectors, 2005q4- 2016q4... 10
7)
Bank- specifics variables of 15 Chinese commercial banks 2009q2- 2016q4... 12Reference……….…. 27
1. Introduction
Non- performing loan (NPL) is an important indicator of systemic risk and financial fragility. Barr et al. (1994) find that banks usually have high amounts of non- performing loans prior to their failure. Reinhart and Rogoff (2010) point out that a large increase of NPLs can be seen as the beginning of a banking crisis. Messai (2013) claims that the deterioration in banks’ loan quality is the main cause of the financial crisis and the economic stagnation. A rising share of problem loans not only erode the profitability of individual banks, but may also shake the stability of the entire banking system due to spillover effects. Hence, limiting NPLs is a priority for authorities seeking to strengthen a banking system and to underpin financial stability.
A policy response by regulatory authorities to address NPLs problems requires a deep understanding of their underlying determinants. The current paper attempts to investigate the causes of increasing NPLs of Chinese commercial banks during the first quarter of 2009 to the last quarter of 2016. The main objective is to find empirical evidence to support the following hypothesis: the amount and ratio of NPLs in the Chinese banking system are affected by both macroeconomic and banking-industry specific determinants. Understanding the potential affect of these factors on the quality of loans provide a practical guidance for macroeconomic analysis and financial supervision as well as for bank risk management. According to the current economic situation in China, from a macro perspective, China faces an economic slowdown, with a relatively low GDP growth below 7%. Besides, the real estate and stock market are very volatile, which increases credit risk. Moreover, a tight monetary policy by central bank of China could also increase NPLs ratios. From a bank-specific perspective, improving efficiency and risk management skills are expected to lower NPLs. This is in line with the results of my study that the macroeconomic variables GDP growth, inflation, money supply and house price significantly decrease NPLs, while real interest rate noticeably increases NPLs. Among bank- specific variables, size, capitalization and loan growth generally raise NPLs, while operating efficiency is negatively related to loan quality. The rest of the paper is organized as follows. Section 2 discusses the literature on macroeconomic and bank- specific determinants of NPLs. Section 3 presents a short historical background of the Chinese banking system and describes the evolution of nonperforming loans in recent two decades. Section 4 provides the data description, methodology and empirical models. Section 5 analyses the empirical results. Section 6 summarizes the results and presents policy implications.
The definition of NPLs are different across countries. Following the definition given by the International Monetary Fund, a loan is non- performing when "payments of interest on principal are past due by 90 days or more, or at least 90 days of interest payments have been refinanced, capitalized or delayed by agreement, or payments are less than 90 days overdue, but there are other good reasons to doubt that payments will be made in full”. Most studies calculate the NPLs ratio by dividing the sum of problem loans by a bank' s total loans.
In recent decades, the literature discusses the determinants of NPLs from three main perspectives. One strand of the literature emphasizes the relationship between the macroeconomic circumstance and the loan quality. Macroeconomic variables are treated as exogenous. One of the earliest studies on this topic is by Lawrence (1995), who uses the life-cycle consumption model to prove that the default rate depends on customers' salary and the unemployment rate. The results show that lower incomes borrowers are more likely to default, as their incomes are less stable and their risk of future unemployment is higher. Besides, banks tend to charge higher interest rates to riskier borrowers, which further increases the default rate.
Bofondi and Ropele (2011) use a single-equation time series approach to study the quality of loans to individual borrowers and firms separately in the Italian banking sector from 1990Q1 to 2010Q2. They find that macroeconomic variables affect these two categories of borrowers differently. Both groups are effected by macroeconomic determinants such as GDP growth, house prices, unemployment and the short term interest rate, but these determinants affect the evolution of NPLs with different time lags. On average, it takes two times longer for them to effect the households' NPLs ratio than that of firms'.
Other research focusing on macroeconomic variables include Boss et al (2009), Cifter et al (2009) and Nkusu (2011). The former two articles study the banking system of Australia and Tunisia, respectively. The main conclusion is that default cycles are affected by macroeconomic variables in business cycles, such as GDP growth, real interest rate and unemployment rate. The latter study points out that there is an interaction between NPLs and macroeconomic vulnerabilities. The NPLs are determined by real economic activities, while, at the same time, a sharp increase in NPLs weakens macroeconomic performances and exacerbates macro financial vulnerabilities.
Summing up, the favorable macroeconomic environment such as sustained GDP growth, low interest and unemployment rate are associated with a better quality of loans, as borrowers receive sufficient streams of revenue and income which allow them to repay their debt easier. Another strand of the literature focuses on the impact of bank-specific determinants on NPLs. Berger and DeYoung (1997) apply Granger-causality techniques to test four hypotheses— bad luck, skimping, bad management and moral hazard— aiming at finding the intertemporal
relationships between loan quality, operating efficiency and capital adequacy of US commercial banks. The study finds empirical evidence in support of the bad luck hypothesis, the bad management hypothesis and the moral hazard hypothesis. More specifically, the bad luck hypothesis suggests that an increase in NPLs caused by exogenous events is likely to be followed by a decrease in operating efficiency, as extra expenses associated with monitoring or selling off problem loans will increase. The bad management hypothesis concludes that a decrease in operating efficiency leads to a future increase in NPLs, as low measured operating efficiency is a signal of poor management skills, and a higher level of NPLs will cause banks to spend more on monitoring and administering their loan portfolio. The moral hazard hypothesis indicates that banks with less capital have more incentives to take risks by rising the riskiness of loan portfolio. However, the research finds no evidence in favor of the skimming hypothesis, which posits that banks might appear to be cost efficient by skimping on the expenses used for loan underwriting and monitoring, but lead to the consequences of increasing problem loans in the future. Podpiera and Weill (2008) similarly find a negative relationship between operating efficiency and NPLs studying the Czech banking system from 1994 to 2005.
The third strand of the literature combines macroeconomic and bank- specific determinants, pointing out that loan quality is related to both the macroeconomic circumstance and the characteristics of banks. Louizis et al. (2012) use dynamic panel data approach to examine the determining factors of NPLs among three different categories-- business loans, consumer loans and mortgages-- of the nine largest banks in Greece during 2003 and 2009. Empirical evidence is presented that the macroeconomic environment and bank-specific factors have a different impact on loan quality in different categories. In general, business loans are most sensitive to macroeconomic factors, such as GDP growth and unemployment rate, while mortgages are less affected by the macroeconomic development. Besides, the results also support the ”bad management” hypothesis that cost efficiency has a negative relationship with NPLs.
Similarly, Ghosh (2014) examines both regional economic determinants and bank- specific determinants of NPLs for commercial banks of 50 US states and the district of Columbia from 1984 to 2013. He uses fixed effects and dynamic panel data estimations to prove that NPLs is highly sensitive to bank- specific indicators such as capitalization, cost inefficiency, industry size and credit quality, but is also affected by personal income, state real GDP, unemployment, house prices and public debt.
In addition, Messai and Jouini (2013), Makri et al. (2014) find strong influence of both macroeconomic and bank- specific variables on problem loans of Countries in the Eurozone. The main conclusion is that economic growth, stock prices and bank profitability negatively influence NPLs while unemployment and real interest rate positively affect NPLs.
3. Evolution of NPLs in China’s banking sector
Similar to most developing countries, the financial system of China has a concentrated banking sector. Historically, the People's Bank of China (PBOC) is the only bank in the Chinese financial system. It is owned by the central government and served as both the central bank and a commercial bank, handling almost all financial transactions of China (Zhang and Zhao, 2012). The first market reform began in 1979, with the establishment of four state-owned banks-- the Bank of China, the Agricultural Bank of China, the China Construction Bank and the Industrial and Commercial Bank of China ( the" Big Four”)--which took over PBOC's commercial banking businesses and constituted the foundation of China's commercial banking system. The reforms progressed in the area of ownership structure, including the introduction of regional commercial banks, joint-stock banks and policy banks1( Liang et al., 2013).
The rapid expansion of the banking sector has stimulated the economy but has also caused problems. During last two decades, one of the most significant problems for China’s banking sector was the high level of NPLs held by state-owned banks, especially by the Big Four. Therefore, bringing the amount of non- performing loans to normal levels was one of the priorities of the Chinese financial system. In the late 1990s, the average NPLs ratio of commercial banks was around 40%, which created a huge burden for the economy. However, banks have made considerable progress in decreasing the amount of NPLs during 2000-2010. The Chinese government started the process of banking sector reforms including through recapitalization and establishing Asset Management Companies (AMC). This involved the transfer of NPLs, which resulted in a sharp decline of the NPLs ratio to 8.3% in 2006 and 0.9% in 2013. It is worth mentioning that the Chinese banking sector was doing relatively well in 2008 with a sharp decrease of the NPLs ratio, just when most developed countries were hit by the ongoing global financial crisis. (Figure 1.)
1 Established in 1994, the China Development Bank (CDB), the Export-Import Bank of China (Chexim) and the Agricultural Development Bank of China (ADBC) are three policy banks in China. These banks are responsible for state-invested projects and financing economic development as well as taking over the government-directed spending functions of the Big four. (via Wikipedia)
Figure 1. The NPLs ratio of the five largest banking groups of China, 2007q2- 2016q4 (Notes: LCB-- large commercial bank; JSCB-- joint- stock commercial bank; CCB-- city commercial bank; RCB-- rural commercial bank; FB-- foreign bank.)
Although market forces are the main determinant of NPLs, targeted policies have contributed to decreasing China's NPLs ratio. By the end of 2005, the China Banking Regulatory Commission published the "Core Indicators for Risk Regulation and Supervision in Commercial Banks”, which stipulate that NPLs ratio should be lower than 5% (Zhang et al, 2016). In order to improve banks' balance sheet, the Chinese government injected large amounts of capital into the banking system from 2003 to 2008, allowing state- owned banks to write off NPLs (Tan and Floros, 2013). Moreover, during the 2008 financial crisis, China's authorities launched a stimulus package of RMB 4 trillion, included policy lending2 and restructured loan3 to support the banks and firms that near bankruptcy. This resulted in a substantial credit expansion and a temporary decrease of the NPLs ratio.
Nevertheless, the decrease of NPLs did not imply that it would not again become a problem in the future. Firstly, the removal of NPLs with the help of government did not deal with the underlying problem creating new NPLs. Besides, because of the government intervention, banks had a lending bias in favor of state- owned enterprises (SOEs), especially those in trouble or those with high default risk (Lu et al, 2005). In fact, government support may create moral hazard incentives for banks if they expect government bailouts whenever facing financial losses. Banks will make more risky loans and become less efficient. A rebound of 2 Lending based on policy objectives or political connections rather than worthiness credit.
3 Restructured loan is a type of loan rescheduled to accommodate borrowers in financial difficulties and to avoid a default. It is often paid over a longer period with a lower installment amount.
NPLs ratio has been observed after 2013 and has increased steadily over 16 consecutive quarters, illustrating the new financial strains.
Secondly, according to some critics, the amount of NPLs in China's banking sector might be significantly underestimated. On the one hand, China has its own unique way to recognize and classify NPLs. Bank loans are classified as pass, special mention, substandard, doubtful and loss according to a five-category system and NPLs are loans classified in the latter three categories. However, as many banks are less prudent in their NPLs classification, a loan will not be classified as bad until the payment of the principal is delayed beyond the maturity date, or, more likely, until the borrower declares bankruptcy or goes into the process of liquidation. Instead, these problem loans would probably be regarded as "special mention” loans. By the end of 2016, although the average NPLs ratio has risen to around 1.7%, the ratio of special mention loans which are likely to turn into bad loans has been around 4%. On the other hand, a significant portion of problem loans was transferred to state-owned asset management companies (AMCs) with a goal to liquidate them. Therefore, the figures on outstanding NPLs in the Chinese commercial banking system do not include these transferred loans. For these reasons, the actual amounts of NPLs could be much larger than the reported official numbers (Allen et al, 2012). These point to the need for new actions to reduce the NPLs ratio to a sustainable level.
4. Methodology and Data Description
4.1 The choice of variables and data descriptionIn this paper, I consider a panel of 15 representative Chinese large commercial banks (LCBs) and joint- stock commercial banks (JSCBs) during 2009 Q1 to 2016 Q4 (The lists of banks are presented at the appendix). This is based on the consideration that the Chinese banking system is dominated by large commercial banks and joint- stock commercial banks. Policy banks and other saving institutions4 are excluded on account of their different business models.
4.1.1 Non- performing loans ratio
The NPLs ratio can be found in the quarterly reports of Chinese commercial banks, as the China Banking Regulatory Committee requires disclosure of financial performance information and operational details to the public. I examine the aggregate level of NPLs, rather than dividing them into separate categories, on account of data limitations.
It is worth mentioning that some ratings agencies such as Fitch argue that Chinese banks--notably the largest lenders--may be under-reporting their NPLs by masking loans as investments through accounting tricks. More specifically, the rapid development of Chinese shadow banking system since 2009 allows banks to shift traditional loans to off-balance sheet by cooperating with non-banking financial institutions. This type of asset is classified as "investment receivables” instead of "loans” and have limited disclosure and transparency. This practice, known as channel business in China, can hidden credit risks and threaten financial safety. Thus, the banks' NPLs ratio mainly reflects the risk of traditional loans while underestimates lenders' true risk position.
Figure 4 depicts the weighted average NPLs ratio of 15 China commercial banks for 2009-2016. There is a decline in NPLs during 2009 to 2012, followed by a rebound from about 0.9% to 1.7% now. The downturn trend can be explained by government write- offs and transferring of NPLs as well as banks' dramatic increase in off- balance sheet financing. Nevertheless, according to Durden (2015), since 2012 banks have come under increasing pressure to manage their reported NPLs level. The loan loss reserves of Chinese commercial banks are insufficient to cover the increasing NPLs even as NPLs have been partly offset by disposals or write-offs. This illustrates the recent deterioration of balance sheets in the Chinese banking industry.
Following the calculation method of Klein (2013) and Ghosh (2014), the dependent variable is calculated by log(NPLs/(1-NPLs)). NPLs denotes the non- performing loans ratio. Ghosh (2014) explains that with logit transformation the dependent variable span the interval [-∞,+∞] instead of [0,1], and produces a symmetric distribution. All other variables discussed below are also expressed in logarithmic forms.
4.1.2 Macroeconomics determinants of NPLs.
NPLs are countercyclical to a nation's overall macroeconomic conditions. In the economic expansion phase, both households and firms have rising incomes and revenues to service their debts, which leads to a relatively small number of problem loans. By contrast, when income decline in the recession phase, it is much more difficult for clients to repay debts (Ahlem, Selma, Messai, 2013). Based on the literature mentioned above, this paper uses GDP growth, inflation, house price, real interest rate and M2 growth as macroeconomic indicators. Several sources are used to collect the quarterly data, including CEIC5, National Bureau of statistics of China andThe People’s Bank of China6. Inflation is calculated by the percentage change of Consumer Price Index provided by CEIC. Figure 2 provides a brief view of their trend in the most recent ten years.
The expected impact of macroeconomic variables on NPLs is described as follow:
GDP: GDP is expected to have a significant negative impact on NPLs, as positive real GDP
growth is usually associated with rising revenues and incomes for households and higher profitability for firms. This improves the capacity of borrowers to repay their debts (Bangia et al 2002). Hence, improvements in the real economy increase the loan quality of commercial banks. This indicator is measured on the basic of quarterly real GDP growth in China.
Inflation: The impact of inflation on non-performing loans is ambiguous. Shu (2002) and
Ghosh (2014) believe that higher inflation weakens borrowers' debts paying ability by reducing their real income when nominal wages are sticky. However, Rinaldi and Arellano (2006) find that inflation are positively related to NPLs, as higher inflation can decrease the real value of outstanding loans and make debt servicing easier. This study uses the percentage change of CPI in China as the proxy.
House price: House prices growth are expected to be negatively related with problem loans.
Rising house prices increase the collateral value of house, which helps borrowers face unexpected adverse shocks (Nkusu, 2011; Beck et al., 2013). Also, higher house prices increase wealth, which lower the default rate since borrowers have additional means to serve their debt. The quarterly growth of house prices in China is used to measure this determinant.
Real interest rate: The real interest rate is expected to be positively related with NPLs. An
increase of the real interest rate can leads to a rise in problem loans especially for loans with 5 CEIC Database is a up-to-date (be updated three times per week), highly detailed database, containing a large group of macroeconomic indicators, included M2 supply, house price and consumption price index of China. 6 Since 2002, the PBOC has revised the system of financial and monetary statistics in accordance with the IMF Manual on Financial and Monetary Statistics. The data are more comprehensively and accurately reflecting the monetary policy and financial operations of banking institutions.
floating rates, as it raises the real value of debt and burdens borrowers' repayment capacity. I use the one- to- three years lending rate in China adjusted by inflation to measure the real interest rate.
Money supply: Money supply is expected to have a negative relationship with NPLs, as it
means that firms and individuals will own more money to service their debts. The default rate will therefore decrease. I measure money supply by China' s quarterly growth of M2.
Figure 2. Macroeconomic variables in China banking sectors, 2005q4- 2016q4
4.1.3 Bank- specifics determinants of NPLs.
Bank- specific data are retrieved from the quarterly reports and balance sheets of each bank. These are available at the ifind7 website under "Finance indicators of Banks". The chosen variables include Return on equity (ROE), Capital adequacy requirement (CAR), loans-to-asset ratio, size, the non-interest income-to-total income ratio and the operating expense-to-operating income ratio. Figure 3 shows the developments of the average bank-specific data of 15 commercial banks.
The expected influence of bank- specific variables on NPLs is described as follow:
Bank profitability: The effect of bank profitability on NPLs can be positive or negative. On
the one hand, bank profitability is expected to have a negative relationship with NPLs, as profitable banks have less incentives to engage in risky activities, according to Ghosh (2014). Inversely, in line with the bad management hypothesis, poor performance with respect to 7 ifind Info financial database is one of the most powerful and comprehensive database for clients who need the complete, accurate and real-time information on Chinese bonds, stocks, futures, funds and banks’ balance sheets. The data are frequently quoted by media, in academic papers, and in research reports.
lending implies a negative relationship between current earning and loan quality (Louizis et al. 2012). However, Rajan (1994) suggests that banks' profits may also be positively related to problem loans, when current earnings are inflated at the cost of increasing NPLs in the future. This pro-cyclical credit policy hypothesis suggests that banks manipulate current earning through a liberal credit policy to boost short- term profitability at the cost of balance sheets deterioration. I measure profits by the return on equity (ROE) of banks.
Bank capitalization: The moral hazard hypothesis set by Berger and DeYoung (1997)
suggests that banks with limited capital will have higher NPLs, as thinly capitalized banks increase the riskiness of their loan portfolio, leading to more problem loans in the future. Shareholders will have less incentives to monitor credit quality and borrowers, as they have limited liabilities in case of a bank failure. This moral hazard hypothesis supports that banks' equity is negatively related to their NPLs ratio. Zhang and Cai (2014) examine the impact of bank behavior on NPLs in China and find empirical results to support this hypothesis, suggesting that risk incentives potentially lead to further deterioration of the loan quality. I measure banks' capitalization by dividing total equity capital by risk weighted assets, which is also known as the capital adequacy ratio (CAR).
Bank diversification: Bank diversification is expected to be negatively related to NPLs since
diversification lowers credit risk. As we know, the revenues of commercial banks stem from interest and non-interest income. Interest income relies on traditional loan making activities and the other type of income includes insurance underwriting, derivatives, asset management fee, etc. The latter is regarded as the diversified source of income (Louizis et al. 2012). Due to the financial regulation in Chinese banking sector, the ratio of non-interest income- to- total income of banks continues to rise. This increasing diversification of banks reduce the relative size of the loans and thus the relative size of default risk. However, Acharya et al. (2006) hold the opposite view that bank' s diversification is no guarantee of greater bank safety or superior performance, as it may result in banks' producing of risker loans or banks' lowering of monitoring effectiveness and thus increase the NPLs ratio. I measure diversification by dividing non-interest income by total income, in a way similar to Louizis et al. (2012), Ghosh (2014).
Size: Following the too- big- to- fail hypothesis, large banks may take excessive risks by
extending loans to lower quality borrowers and by increasing their leverage, as it is hard to impose market discipline on systemic banks who may be protected by the governmentwhen
they face financial challenges (Stern and Feldman, 2004). Hence, NPLs may have a positive
relationship with the size of banks. I measure the size of banks by dividing assets of one bank by the total assets of 15 banks, similar to Louizis et al. (2012).
Operating efficiency: The impact of operating efficiency on NPLs can be ambiguous.
cost-efficient if they spend less effort monitoring borrowers' behavior and loan quality. Inversely, following the bad management hypothesis, decreases in operating efficiency will cause an increase in nonperforming loans, as this signals poor management skills in monitoring loans quality, borrowers behavior and credit scoring. In general, most of the empirical literature favors the bad management hypothesis over the skimping hypothesis. I measure operating efficiency by dividing operating expenses by operating income, in line with Louizis et al. (2012).
Loan growth: Banks' loan amount is expected to have a positive relationship with the future
problem loans. Ghosh (2014) claims that banks tend to accelerate their loan supply by lowering the minimum credit standard and extending loans to lower quality borrowers, thus leads to an increase in the default rate. Besides, as risky loans are less liquid, a surge in loan supply will also increase liquidity risk. For the study on Chinese banking industry, Zhang et al. (2012) find that NPLs have a positive relationship with the total amount of loan of banks. I use the loan- to- asset ratio to measure the loan growth of banks and its impact on NPLs is expected to be positive.
Table 1 summarize all the variables and their sources.
Figure 3. Bank- specifics variables of 15 Chinese commercial banks 2009q2- 2016q (Notes: NI is the non- interest income- to- total income ratio; INEF is the operating expense- to- operating income ratio.)
Table 1. Summary of variables.
Variables Description Source Expected sign Hypothesis
Banking- specific determinants
Operating efficiency Operating expense- to- operating income ifind +/- Bad management hypothesis
Skimming hypothesis
Capitalization Capital adequacy ratio ifind +/- Moral hazard hypothesis
Diversification Non- interest income- to- total income ifind - Diversification hypothesis
Loan growth Loan- to- assets ratio ifind + Too- big- to- fail hypothesis
Industry size Assets of one bank- to- total assets of 15 banks ifind + Too- big- to- fail hypothesis
Banks profitability Return on equity ifind -/+ Pro- cyclical credit policy hypothesis
Macroeconomic determinants
GDP Quarterly GDP growth National Bureau of
statistics of China
-Inflation Percentage change in CPI CEIC
+/-Real interest rates 1-3 years lending rates People's bank of China +
House price Quarterly house price growth CEIC
-4.2
Empirical Methodology
4.2.1 Empirical models
Following the studies of Louzis et al (2012) and Ghosh (2014), I apply both fixed effects estimation and dynamic panel data approach to examine the determinants of NPLs in the Chinese banking system.
Fixed effects estimation model: The model to be estimated is given by:
( )
j( )
kit j it k t i it
j k
Y
X
X
(1)where Yitrepresents the logit transformation of the NPLs ratio for banks iin periodt, ( j)
it X
represents the vector of bank- specific variables and ( k)
t
X denotes the macroeconomics determinants. irepresents each bank and t each quarter. uiis the unobserved bank- specific fixed effect,
itis the error term. The fixed effects model can control the effect of unobservedtime- invariant factors.
Dynamic panel data estimation:
Fixed effects estimation may lead to inconsistent results, as the error term is correlated with the lagged dependent variable. Besides, the endogeneity problem between independent variables and NPLs should also be taken into consideration. For example, increasing NPLs reveal a deterioration of banks' loan quality, which may decrease banks' profits or do harm to banks' operating efficiency. Also, a dramatic increase in NPLs can weakens a country' s macroeconomic performances. To deal with this problem a dynamic panel data estimation8 will be adopted.
The baseline model is given by:
1
( )
,| | 1, 1,..., ; 1,...,
it it it i it
Y
Y
L X
i
N t
T
. (2)8 Developed in 1990s, the dynamic panel data approach are designed for situation with independent variables that are not strictly exogenous and are correlated with past and current errors; with fixed effects; and with heteroscedasticity and autocorrelation within individuals (Roodman, 2007).
where i refer to the individual dimension and t is the time dimension of the panel, Yit
represents the logit transformation of NPLs, ( )L denotes the 1 k lag polynomial vector,
it
X denotes the k 1vector of explanatory variables except for Yit1,
i is the unobservedvariables that vary per bank and itis the error term.
The Generalized Method of Moments (GMM) technique developed by Arellano and Bond (1991) will be used to estimate Eq. (2). According to Arellano and Bond, the GMM estimation is based on the first difference transformation of Eq. (2) , which can eliminate the time invariance bank- specific effects:
1 ( )
it it it it
Y
Y
L X
. (3)
whererefers to the first difference operator.
The lags of the dependent variable should satisfy the condition: E y[ it s
it] 0 for3,...,
t Tand
s
2
. Following Louzis et al (2012),Yit1in Eq. (3) is correlated with the error term
it and may lead to a bias in the model estimation. However, yit2 is correlated with Yit1while not correlated with
it fort 3,...,T . Hence, yit2 can be seen as an instrument variable in Eq. (3) .Besides, explanatory variables should satisfy the condition:E X[ it s
it] 0 fort 3,...,Tand
s
2
. This is because bias may stem from the endogeneity between independent variables and NPLs as well as their correlation with the error term. To avoid these problems, only the current and lagged value of Xitcan be used as instruments.Eq. (4) is a baseline model examining the effect of macroeconomics variables on NPLs:
2 , 1 2 3 1 1 1 1 4 5 1 1 2 T T T it i t j j t j j t j j t j j j j j T T j t j j t j it j j NPL NPL RGDP RIR M housepr Infl
(4) With | | 1
, i= 1,..., 15 banks and t= 1,..., 36 quarters. j denotes the lags of explanatory variables.examine their explanatory power. Only one type of bank-specific variable will be added at a time to reduce the amount of additional instruments, which is known as "restricted GMM procedure”(Judson and Owen, 1999). The baseline model in Eq. (4) is extended by:
2 , 1 2 3 1 1 1 1 4 5 6 , 1 1 1 2 T T T it i t j j t j j t j j t j j j j j T T T j t j j t j j i t j it j j j NPL NPL RGDP RIR M housepr Infl X
(5) Where Xitcontains all bank-specific variables presented in Table 1.Definitions of all variables are shown in the appendix. I apply four lags for bank-specific indicators to obtain the dynamics of explanatory variables over the previous quarters, similar to Louzis et al. (2012). What is worth mentioning is that I only use the contemporaneous value of “size” variables, as most of findings in the existing literature indicate that banks size is a more permanent feature of the banking sector and its lags have no explanatory power. Following Berger and DeYoung (1997), due to the time delay in changing management decisions, current bank- specific variables is assumed to have no effect on current level of NPLs. However, according to Louzis et al. (2012), the current level of variables also contributes to the changes of the loan quality. Thus, aside from the individual lag regression, I also run a long run effect regression :
2 , 1 2 3 4 5 6 1
2
it i t j t t t t t it it jNPL
NPL
RGDP
RIR
M
housepr
Infl
X
4.3. Tests
Panel unit root test
I perform both common panel unit root tests (Levin et al, 2002) and individual panel unit root tests (Im et al, 2003) on the variables. The null hypothesis is that all the variables are non-stationary (the results are presented at the appendix). Variables that reject the null hypothesis are first- differenced to induce stationary.
Sargan test ( Over- identification test)
The Sargan test is used to test for the over-identifying restriction in the dynamic panel data model. The results should not reject the null hypothesis that all the instruments are exogenous and valid. The results are shown in Table 4.
In GMM methods, AR(1) and AR(2) are used for testing the first and second order autocorrelation of the residuals. In order to get consistent GMM results, the results should rejected the null hypothesis of no first order serial correlation of the residuals while should not reject the null hypothesis of no second order serial correlation of the residuals. The results are presented in Table 4.
5. Empirical Analysis
In this section, I will discuss the results of both fixed effects estimation and dynamic panel data estimation.
5.1 Fixed Effects Estimation
Table 3 presents the estimation results using the fixed effects model.
Model 1 only examines the explanatory power of five macroeconomic variables, namely GDP growth, money supply, inflation, real interest rate and house price. The regression results are shown in the first column. All macroeconomics variables have the right sign and all macroeconomic variables except money supply are statistically significant.
The increasing amount of NPLs is negatively related to the GDP growth, which is compatible with the theory that a growing economy raises borrowers’ incomes and improves their capacity to repay debts. As mentioned above, the Chinese economy has witnessed a steady slowdown in recent years. In order to achieve the goal setting by Chinese policymakers that the country must at least post 6.5% annual economic growth through 2020, the government directs state-owned banks to stimulate the economy by issuing unprecedented levels of debt. As lenders are encouraged to provide loans to struggling businesses with high default risks, the credit quality is eroded. A 1% decrease in GDP growth rate increases NPLs by 8.6%. Inflation and house price have a significant negative impact on NPLs. A 1% increase in inflation decreases NPLs by 2.7%. This is because inflation reduces the debt’s real value and makes debt servicing easier. Likewise, a 1% increase in house price lowers NPLs by 0.4%. This is consistent with my expectation that the rising house price in China increase the collateral value of house and give borrowers additional means to face unexpected shocks, therefore decreasing the mortgage default rate.
The coefficient for the real interest rate is significant and positive as expected. A 1% increase in real interest rate higher NPLs by 4.9%. This is, again, consistent with most existing studies that a rising real lending rate will increase the real value of debts and lead to higher default
rates. The coefficient of money supply is statistically insignificant.
Overall, these results confirm the countercyclical nature of NPLs that both individual borrowers and firms will increase their debt serving ability in a favorable macroeconomic environment. Conversely, during the downward phase of business cycles, a deterioration in the macroeconomic circumstances is associated with rising NPLs. Besides, an easing monetary policy environment with lower interest rate and rising inflation will lead to a decrease in NPLs.
Turing to bank- specific variables, Column [2] to [7] present the results when bank- specific variables are included. Only the indicator of operating efficiency measured by operating expense divided by operating income is found to have a significantly positive relationship with problem loans, while the rest of variables are all statistically insignificant. This result provides strong evidence in support of the "bad management” hypothesis in line with the finding of Berger and DeYoung (1997). As the increase of operating expense- to- operating income ratio indicates lower operating efficiency, a 1% decrease of banks' operating efficiency will lead to an increase of NPLs by 4.5%. This shows that poor management skills and high operating cost in the Chinese banking sector have a negative impact on loan quality. There is no empirical evidence in favor of the "moral hazard” hypothesis, the "too-big-to fail” hypothesis and the "diversification” hypothesis for the Chinese banking system following the fixed effects estimation.
The coefficients of macroeconomic variables remain stable after introducing bank- specific variables, with results close to the estimation of the baseline model.
5.2 Dynamic Panel Data Estimation
Panel A of Table 4 presents the long- run coefficients estimation and Panel B presents the individual lag one- step GMM estimation. The results of Sargan test and AR test are shown at the bottom of the Table, indicating that all instruments are exogenous and valid and the GMM results are consistent.
Gleaning first at the coefficient of macroeconomic variables shown in model [1] of panel A, after introducing the lagged dependent variables, the results confirm some findings from the fixed effects model. Both the GDP growth rate and inflation are statistically significant at 1% level and negatively influence NPLs as expected. A 1% rise of GDP and inflation lower NPLs by 7.2% and by 3.9%, respectively. Besides, a 1% increase of real interest rate leads to an increase of 3.6% in the NPLs ratio. The coefficient of house price are now only statistically significant at 1% level in several models. These finding, again, illustrates the countercyclical nature of non- performing loans. Moreover, the previous quarter's NPLs positively impacting
Table 3 Fixed Effects Estimation Results.
The coefficients in bold denote statistically significant values.
Both bank and time dummies are included while are not shown in the Table.
Variables [1] [2] [3] [4] [5] [6] [7] GDP -0.0862*** -0.0770*** -0.0835*** -0.0827*** -0.0845*** -0.0885*** -0.0873*** (0.0164) (0.0168) (0.0169) (0.0159) (0.0164) (0.0159) (0.0162) Infl -0.0268*** -0.0240*** -0.0251*** -0.0264*** -0.0290*** -0.0271*** -0.0281*** (0.00813) (0.00761) (0.00815) (0.00788) (0.00844) (0.00799) (0.00783) RIR 0.0494*** 0.0478*** 0.0505*** 0.0482*** 0.0467*** 0.0517*** 0.0497*** (0.0117) (0.0121) (0.0112) (0.0124) (0.0122) (0.0114) (0.0118) housepr -0.0044** -0.00357* -0.00456** -0.00399** -0.00358 -0.00417* -0.00403* (0.0019) (0.0018) (0.00193) (0.00181) (0.00205) (0.00195) (0.00204) M2 0.0303 0.0164 0.039 0.0229 0.0138 0.0251 0.0208 (0.0338) (0.0329) (0.0343) (0.0334) (0.0383) (0.0342) (0.0346) INEF 0.0451** (0.018) CAR 0.0764 (0.0475) ROE 0.00505 (0.00595) NI -0.04 (0.034) LTA 0.0317 (0.0237) size -0.0379 (0.0232) Constant -0.255*** -0.230*** -0.167* -0.246*** -0.265*** -0.262*** -0.322*** -0.062 -0.0587 -0.0878 -0.0569 -0.0624 -0.0605 -0.0667 R-squared 0.426 0.432 0.432 0.427 0.428 0.429 0.43 F-stat 34.77 40.64 42.00 35.04 26.44 34.34 49.77
Notes: Robust standard errors are reported in parentheses. *** Denotes significance at 1% respectively (p<0.01). ** Denotes significance at 5% respectively (p<0.05). * Denotes significance at 10% respectively (p<0.1).
the present NPLs value by 89%, which reveals the persistence of NPLs. This indicates the prolonged effect of NPLs on future loan quality. That is to say, an increase of current problem loans has a persistent impact on the future NPLs. Similarly, an effort to reduce the current NPLs ratio will also lead to persistently lower levels in the future.
Most of the estimation results shown in columns [2]- [7] are also consistent with the ex- ante findings. The coefficients of macroeconomic variables exhibit the same significant and sign as in the baseline model. The rising previous NPLs continue to have a strong and positive influence on the increase of current NPLs. Banks’ operating inefficiency are statistically significant and positive related to NPLs. This, again, is consistent with the results found by the fixed effects estimation and supports the " bad management" hypothesis.
Interestingly, banks size is now observed to have a noticeable positive impact on NPLs. A 1% rise in bank size increases NPLs by 1.2%, supporting the "too- big- to- fail” hypothesis that large sized state- owned banks in China tend to take excessive risk by extending loans to borrowers with lower quality, as they believe the government will provide protection in the case of bank failures. In addition, some literature assume that the size of bank have a significantly positively effect on their level of disclosure. This is due to the fact that larger banks have higher agent cost and that greater transparency in financial reports can reduce this cost (Chipalkatti, 2002; Marston et al., 2004). Thus, large size banks are less likely to hide their real NPLs value on the balance sheet.
Nevertheless, banks’ capitalization, banks’ probability, loan-to-asset ratio and diversification are still statistically insignificant and do not show explanatory power on loan quality. Following Berger and DeYoung (1997), banks who meet the capital adequacy requirement are less likely to have excessive risk taking behavior, thus the capital amounts of banks are less correlated with loan quality deterioration. Since 2011 the China Banking Regulatory Commission requires all commercial banks to meet the minimum 8% capital adequacy requirement according to Basel I, which may limit bank managers' moral hazard incentives. Besides, following Louzis et al. (2012), a possible explanation for the rejection of the "diversification” hypothesis could be ascribed to the "dark sides" of diversification. That is to say, banks do not have comparative advantage when entering a business they are not familiar with or have no experience in. Thus, bank' s diversification is no guarantee of lower default risk.
Panel B presents the individual lag one- step GMM estimation. I apply two lags for macroeconomic variables and four lags for bank- specific variables.
For the macroeconomic variables, the coefficients of the lag value of GDP and inflation are consistently statistically significant and negatively related to NPLs. A 1% increase of GDP in the previous one and two quarters lowers NPLs by 7.4%- 9.3% and by 3.8%- 4.8%,
respectively. Likewise, a 1% increase of inflation in the previous one quarter decreases NPLs by 3.5%- 4.5%. The first order lag of real interest rate also has a noticeable effect on NPLs. It is worth mentioning that the previous one quarter of M2 supply are shown to be statistically significant and have a negative relationship with NPLs, indicating that money supply have a delayed effect on firms' and individuals' ability to serve their debts. In general, these results signal the strong dependence of borrowers' debt serving ability on the past economy growth and monetary policies.
Turing to bank- specific variables, the coefficient of both the first and second order lag of the dependent variable NPLs is positive and statistically significant. In model [2], bank size has a strong and positive impact on NPLs, continuing support the "too-big-to-fail” hypothesis. In model [3], the previous four- periods lagged value of the indicator of banks' operating efficiency is statistically significant and affects the current level of NPLs by 4.6%, suggesting the delayed pass through from lower cost efficiency to higher NPLs spans four quarters. Model [4] again rejects the negative relationship between capital adequacy ratio and problem loans. Instead, the fourth order lag of CAR is positively related to NPLs, which is at odds with the "moral hazard” hypothesis mentioned above. This can be explained by several reasons. First, banks with more capital tend to apply liberal lending policies and result in rising NPLs. Second, bankers may build wrong confidence in bank management and become less sensitivity to portfolio risk due to the increasing capitalization. (Diana Teixeira, 2013). Also, banks with higher capital are less likely to hide their real NPLs value on the balance sheet, as they have more incentive to enhance their disclosure transparency to both the public and stakeholders.
Model [5] indicates that the ROE is statistically insignificant. In model [6], the first order lags of the loan- to- asset ratio have a noticeable positive effect on the present NPLs value. A 1% rise in loans- to- asset ratio raises NPLs by around 12%. This is in line with the theory mentioned above that banks with loan growth lowers its underwriting standards and attract lower quality borrowers, therefore increase the default risk.
Combining the results in Panel A and Panel B, we can see that some variables such as money supply and loan- to- asset ratio do not affect NPLs immediately but after a lag of one quarter, while other variables such as house prices and the size of bank efficiency affect the NPLs ratio simultaneously. GDP growth, inflation, real interest rate and banks’ operating efficiency have both simultaneous and delayed impacts on the current level of problem loans.
Table 4 GMM estimation results:
Panel A: Long run coefficients estimation
Variables [1] [2] [3] [4] [5] [6] [7] ΔNPLt-1 0.892*** 0.861*** 0.890*** 0.891*** 0.875*** 0.883*** 0.882*** (0.035) (0.0345) (0.0346) (0.0352) (0.036) (0.0346) (0.0346) ΔNPLt-2 0.0528 0.0605* 0.0566* 0.049 0.0572* 0.0543 0.0447 (0.0333) (0.0324) (0.0332) (0.0338) (0.0331) (0.0331) (0.0334) ΔGDP -0.0716*** -0.0633*** -0.0726*** -0.0668*** -0.0687*** -0.0690*** -0.0657*** (0.0108) (0.0107) (0.0108) (0.0111) (0.0108) (0.0108) (0.0109) ΔInfl -0.0353*** -0.0322*** -0.0353*** -0.0353*** -0.0357*** -0.0363*** -0.0371*** (0.00403) (0.00406) (0.00406) (0.00403) (0.00399) (0.00399) (0.00402) ΔRIR 0.0363*** 0.0354*** 0.0368*** 0.0339*** 0.0344*** 0.0340*** 0.0313*** (0.00826) (0.00798) (0.00819) (0.0082) (0.0082) (0.00826) (0.00832) Δhousepr -0.00211* -0.00176 -0.00216* -0.0015 -0.00214* -0.0018 (0.00146 (0.00118) (0.00114) (0.00117) (0.0012) (0.00119) (0.00118) (0.00118) ΔM2 -0.014 -0.0246 -0.0138 -0.0237 -0.0136 -0.0188 -0.0237 (0.0214) (0.0208) (0.0215) (0.0216) (0.0216) (0.0214) (0.0213) ΔINEF 0.0524*** (0.0161) ΔCAR -0.00862 (0.0339) ΔROE 0.00486 (0.00424) ΔNI 0.0279 (0.0294) ΔLTA -0.0051 (0.0187) Δsize 0.0121** (0.00518) Constant -0.403*** -0.424*** -0.409*** -0.406*** -0.431*** -0.422*** -0.427*** -0.0419 -0.0394 -0.0557 -0.0414 -0.0436 -0.0406 -0.0405 Sargan 337.0454 405.1335 400.6441 404.2096 406.5651 412.044 412.6237 test [0.1485] [0.1445] [0.1830] [0.1519] [0.1335] [0.100] [0.0937] AR(1) [0.0007]*** [0.0013]*** [0.0007]*** [0.0007]*** [0.0008]*** [0.0008]*** [0.0007]*** AR(2) [0.4123] [0.4066] [0.4346] [0.4455] [0.4268] [0.4268] [0.3736]
Notes: Standard errors are reported in parentheses.
The P- values for the Sargan test and the AR test are reported in brackets. *** Denotes significance at 1% respectively (p<0.01).
** Denotes significance at 5% respectively (p<0.05). * Denotes significance at 10% respectively (p<0.1).
Panel B: Individual lag coefficients estimation Variables [1] [2] [3] [4] ΔNPLt-1 0.845*** 0.843*** 0.886*** 0.879*** (0.0413) (0.0407) (0.0424) (0.042) ΔNPLt-2 -0.0629* -0.0835** -0.0717* -0.0676* (0.0381) (0.0382) (0.0389) (0.0391) ΔRIRt-1 0.0136 0.00954 0.0239* 0.0153 (0.0123) (0.0124) (0.013) (0.0132) ΔRIRt-2 -0.0139 -0.0174 -0.0171 -0.0131 (0.0109) (0.0109) (0.0111) (0.0113) ΔInflt-1 -0.0413*** -0.0450*** -0.0359*** -0.0353*** (0.0102) (-0.0103) (0.0103) (0.0105) ΔInflt-2 -0.00839 -0.00985* -0.00772 -0.00618 (0.00569) (0.00571) (0.006) (0.00583) ΔGDPt-1 -0.0802*** -0.0798*** -0.0837*** -0.0737*** (0.0201) (0.0202) (0.0205) (0.0217) ΔGDPt-2 -0.0414*** -0.0449*** -0.0423*** -0.0384** (0.0147) (0.0148v (0.0156) (0.0157) Δhouseprt-1 -0.00142 -0.00157 0.000161 -0.0023 (0.00253) (0.00254) (0.0028) (0.00281) Δhouseprt-2 0.00317* 0.00387** 0.00124 0.00236 (0.00184) (0.00185) (0.00212) (0.00199) ΔM2t-1 -0.0715** -0.0754** -0.0694** -0.0583* (0.0321) (0.0323) (0.035) (0.0332) ΔM2t-2 -0.0227 -0.0233 -0.0119 -0.024 (0.0299) (0.03) (0.0317) (0.0316)
Δsize 0.0264*** ΔINEFt-1 -0.0176 ΔCARt-1 -0.0375
(0.00561) (0.022) (0.0465) ΔINEFt-2 -0.00852 ΔCARt-2 0.0664 (0.0237) (0.0547) ΔLINEFt-3 -0.0331 ΔCARt-3 -0.0656 (0.0217) (0.0513) ΔLINEFt-4 0.0456** ΔCARt-4 0.0743* (0.0188) (0.0426) Constant -0.597*** -0.628*** -0.499*** -0.457*** (0.0758) (0.0733) (0.0724) (0.0896) Sargan test 287.9917 329.0893 325.8852 331.3967 [0.0894] [0.2550] [0.2779] [0.2515]
Variables [5] [6] [7] ΔNPLt-1 0.883*** 0.846*** 0.857*** (0.0408) (0.0421) (0.0413) ΔNPLt-2 -0.0758* -0.0637* -0.0698* (0.0388) (0.0385) (0.0386) ΔRIRt-1 0.0272* 0.0247* 0.0233* (0.0145) (0.0131) (0.0135) ΔRIRt-2 -0.0174 -0.0165 0.021* (0.0115) (0.0111) (0.0115) ΔInflt-1 -0.0350*** -0.0372*** -0.0348*** (0.0104) (0.0102) (0.0105) ΔInflt-2 -0.00585 -0.00853 -0.00906 (0.00588) (0.00583) (0.00586) ΔGDPt-1 -0.0863*** -0.0934*** -0.0771*** (0.0223) (0.0217) (0.0208) ΔGDPt-2 -0.0471*** -0.0475*** -0.0484*** (0.0162) (0.0152) (0.0152) Δhouseprt-1 -0.000918 0.000108 -0.000806 (0.00308) (0.00264) (0.00257) Δhouseprt-2 0.0022 0.000492 0.00233 (0.0026) (0.00209) (0.00187) ΔM2t-1 -0.0816** -0.0598* -0.0763** (0.0367) (0.0344) (0.0332) ΔM2t-2 -0.00495 -0.00456 -0.0343 (0.0349) (0.0315) (0.0312)
ΔROEt-1 -0.00296 ΔNIt-1 -0.0598 ΔLTAt-1 0.120***
(0.00853) (0.0439) (0.0438)
ΔROEt-2 0.00757 ΔNIt-2 0.0635 ΔLTAt-2 -0.0806
(0.00885) (0.0454) (0.0569)
ΔROEt-3 -0.00769 ΔNIt-3 0.0648 ΔLTAt-3 -0.0705
(0.00845) (0.0397) (0.0523)
ΔROEt-4 0.00819 ΔNIt-4 -0.0069 ΔLTAt-4 0.0399
(0.00824) (0.0419) (0.0418)
Constant -0.491*** -0.582*** -0.545***
(0.0742) (0.0748) (0.076)
Sargan test 315.7926 320.1681 309.922
6.Results and Policy Implications
A thorough understanding of the determinants of increasing NPLs can help in the design of polices promote financial stability of the banking sector. In this study, I adopt both fixed effects and dynamic panel data method to estimate the drivers of NPLs in the Chinese banking system. The findings indicate that macroeconomic variables, specifically the GDP growth, inflation, money supply, house prices and real interest rate have a strong effect on NPLs. Besides, bank- specific variables such as banks size, capitalization, loan growth and operating efficiency have additional explanatory power when introduced in the baseline model. This supports the " too- big- to- fail” hypothesis and the "bad management” hypothesis. Furthermore, empirical evidence also indicates that the past value of NPLs has an effect on the current value of NPLs. The results are thus consistent with the hypothesis that both macroeconomic and bank- specific determinants have an effect on loan quality, and that NPLs are subject to persistence.
The results have implications in term of policy and regulation. During the last two decades government policies have played a vital role in decreasing NPLs in the Chinese banking sector. A key policy introduced by the Chinese government to deal with bad loans is the establishment of four state-owned asset management companies (AMCs). A second policy is that the government directly issues bonds to state banks and purchases NPLs from them to convert problem loans into state shares. In addition, the government also uses private financial companies to purchase banks’ non- performing loans at a discounted rate.
However, such government support may increase banks‘ too- big- to- fail problem and bad management problem, as the large state-owned banks may become less prudent in lending activities and may expect government bailouts when they face financial challenges. Besides, banks will have less incentive to enhance their operating efficiency and management skills, therefore raising the underlying level of NPLs. Moreover, this approach increases the government burden by transforming NPLs into explicit budgetary debt.
Thus, to obtain a sustainable decrease of NPLs in China, it is important for regulators to break the vicious cycle between the state-owned banks, (SOBs) the state-owned entrepreneur (SOEs) and the state budget. Instead of imposing policy loans on commercial banks by the government, proposals to resolve the NPLs problem should be conditional on enhancing banks' capacity to deal with bad loans. Efficient cost management and risk management should be prerequisites to improve the banks' balance sheet. Banks should be required to meet the capital adequacy requirement and establish effective monitoring of their lending behavior. Also, banks need to promote the transparency of their balance sheets. In addition, bank ownership structure reform should be deeper to allowed banks operating in a modern corporate system environment. Last but not least, bankruptcy frameworks should be improved
and bankruptcy law should be enforceable to discipline inefficient SOBs.
For the macroeconomic perspective, as the empirical evidences show that the recession phase of business cycles has an adverse impact on NPLs, it is also very important to improve the overall economic health of China. Besides, understanding the relationship between macroeconomic determinants and NPLs is important to build a more specific macro- stress testing model and financial stability assessment function, as the credit risk scenarios used in macro- stress testing exercises depend on macroeconomic circumstances of a country. This could therefore contributes to strengthen the banking system and enhance resilience to financial stress.
This research has some limitations due to the continuously changing regulatory environment in the Chinese banking sector and the lack of specific data of certain loan types. In future studies, other macroeconomic variables could be included such as the unemployment rate, personal income growth and credit growth in the baseline model. Other variables specific to banks such as the liquidity ratio and loan loss provision may also have explanatory powers on NPLs. Researcher can also extend the study by dividing NPLs into different types of loans according to Louzis et al. (2012) and see which category is most sensitive to changes of variables. In addition, the market share of regional commercial banks will increase because of the deeper banking reform. This will require more data in future empirical studies. Moreover, other econometric methods such as VAR models can be used to investigate the feedback effects of non- performing loans on determinant factors.
Reference
Abid, L., Ouertani, M., Zouari-Ghorbel, S., 2013. Macroeconomic and Bank-Specific Determinants of Household's Non-Performing Loans in Tunisia: a Dynamic Panel Data. Procedia Economics and Finance 13, 58 – 68
Allen, F., Qian, J., Zhang, C., Zhao, M., 2012. China's Financial System: Opportunities and Challenges. NBER Working Paper No. 17828
Arellano, M., Bond, S., 1991. Some tests of specification for panel data. Monte Carlo evidence and an application to employment equations. Rev. Econ. Stud. 58,277–297.
Arellano, M., Bover, O., 1995. Another Look at the Instrumental Variable Estimation of Error-Component Models. Journal of Econometrics 68, 29–51.
Bangia, A., Diebold, F., Kronimus, A., Schagen, C., Schuermann, T., 2002. Ratings migration and the business cycle, with application to credit portfolio stress testing. Center for Financial Institutions Working Papers, 26 ( 2–3) : 445-474
Barr, R. S., Seiford, L. M., Siems, T. F., 1994. Forecasting bank failure: a non- parametric frontier estimation approach. Recherches Economiques de Louvain/Louvain Economic Review 60 (4), 417–429
Beck, R., Jakubik, P., Piloui, A., 2013. Non-performing Loans What Matters in Addition to the Economic Cycle? European Central Bank Working Paper Series 1515.
Berger, A., DeYoung, R., 1997. Problem loans and cost efficiency in commercial banks. J. Bank. Finance. 21, 849–870.
Blundell, R., Bond, S., 1998. Initial conditions and moments restrictions in dynamic panel data models. J. Econometrics. 87, 115–143.
Bofondi, M. and Ropele, T; 2011. Macroeconomic Determinants of Bad Loans: Evidence from Italian Banks, Occasional Papers, 89.
Bond, S., 2002. Dynamic Panel Data Models: A Guide to Micro Data Methods and Practive, Portuguese Economics Review 1 (2).
Bond, S., Windmeijer F, 2002. Finite Sample Inference for GMM Estimations in linear Panle Data Model: A Comparision of Alternative Test mimeo, Institute for fiscal studies, London. Boss, M., Fenz, G., Pann, J., Puhr, C., Schneider, M., Ubl, E., 2009. Modeling Credit Risk Through the Austrian Business Cycle: an Update of the OeNB Model. OeNB Financial Stability Report 17, 85–101.
Chipalkatti, N., 2002. Market Microstructure Effects of the Transparency of India banks, National Stock Exchange, India Working Paper.
Cifter, A., Yilmazer, S., Cifter, E., 2009. Analysis of sectoral credit default cycle dependency with wavelet networks: evidence from Turkey. Econ. Modell. 26,1382–1388.
Espinoza, R., Prasad, A., 2010. Nonperforming loans in the GCC Banking System and their Macroeconomic Effects. International Monetary Fund Working Paper 10/224.
Ghosh, A.,2014. Banking-industry specific and regional economic determinants of non-performing loans: Evidence from US states. Journal of Financial Stability 20, 93–104. Im, K., Pesaran H., Shin Y., 1997. Testing for Unit Root in Heterogeneous Panels. Duscussion Paper, University of Cambridge.
Jimenez G., Saurina, J., 2006. Credit Cycle, Credit risk and Financial Regulation. International Journal of Central Banking. Vol.2, pp 65- 98.
Judson, R. A. And Owen, L.A., 1999. Estimating Dynamic Panel Data Models: A Guide for Macroeconomists. Economics Letters 65, 9- 15.
Klein, N., 2013. Non-performing Loans in CESEE: Determinants and Impact on Macroeconomic Performance. International Monetary Fund Working Paper 13/72.
Lawrence, E., 1995. Default and the life cycle model. J. Money Credit Bank. 27,939–954. Levin, A., Lin, C., Chu, C.J., 2002. Unit root tests in panel data: asymptotic and finite sample properties. J. Econometrics. 108, 1–24.
Liang, Q., Xu, P., Jiraporn, P., 2013. Board characteristics and Chinese bank performance. Journal of Banking & Finance 37 (8), 2953–2968.
Louizis, D., Vouldis, A., Metaxas, V., 2012. Macroeconomic and bank-specific determinants on non-performing loans in Greece: a comparative study of mortgage, business and consumer loan portfolios. J. Bank. Finance. 36, 1012–1027.
Lu, D., Thangavelu, S.M., Hu, Q., 2005. Biased lending and non-performing loans in China's banking sector. The Journal of Development Studies 41 (6), 1071–1091.
Makri, V., Tsagkanos, A., Bellas, A., 2014. Determinants of non- performing loans: the case of Eurozone. Panoeconomics 2. 193- 206.
Marston C., and Polei, A., 2004. Corporate reporting on the Internet by German companies. International Journal of Accounting Information Systems, 5, no. 3, 285–311
Messai, A., Jouini, F., 2013. Micro and macro determinants of non-performing loans. Int. J. Econ. Financ. Issues 3, 852–860.
Nkusu, M., 2011. Nonperforming Loans and Macro- financial Vulnerabilities in Advanced Economies. International Monetary Fund Working Paper 11/161.
Podpiera, J., Weill, L., 2008. Bad luck or bad management? Emerging banking market experience. J. Finance. Stab. 4, 135–148.
Rajan R., 1994. Why Bank Credit Policies Fluctuate: A Theory and Some Evidence. The Quarterly Journal of Economics 109, 339- 441.
Reinhart, C., Rogoff, K., 2010. From Financial Crash to Debt Crisis. National Bureauof Economic Research Working Paper 15795.
Renaldi L., A, Sanchis- Arellano, 2006. Household debt sustainability: What explains household non- performing loans? An empirical analysis. ECB working paper, no. 570. Roodman, D., 2006. How to do xtabond2: an introduction to difference and system GMM in Stata.
Salas, V., Saurina, J., 2002. Credit risk in two institutional regimes: Spanish commercial and saving banks. Journal of Financial Services Research 22, 203- 224.
Shu C., 2002. The Impact of Marcoeconomic Environment on the Asset Quality of Hong Kong's banking sector. Hongkong Monetary Authority Research Memorandums.
Tan, Y., Floros, C., 2013. Risk, capital and efficiency in Chinese banking. Journal of International Financial Markets, Institutions and Money 26, 378–393.
Zhang, D., Cai, J., Dickinson, D., Kutan, A., 2015. Non-performing loans, moral hazard and regulation of the Chinese commercial banking system. Journal of Banking & Finance 63, 48–60.
Appendix
1. List of commercial banks.
No. Banks
1 Bank of China
2 Ping An Bank Co., Ltd.
3 China Construction Bank
4 Agriculture Bank of China
5 Bank of Communications
6 Industrial and Commercial Bank of China
7 Industrial Bank Co., Ltd
8 China Minsheng Banking Co., Ltd
9 Bank of Beijing
10 China Merchants Bank Co., Ltd
11 HUA XIA BANK
12 Bank of Nanjing
13 Bank of Ningbo
14 Shanghai Pudong Development Bank
2. Definition of variables and statistics summary.
Variables Definition Obs. Mean Std. Dev. Min Max
NPL Non- Performing Loans 480 0.0108411 0.0047249 0.0034 0.0432 CAR Capital Adequacy Ratio 480 0.1207804 0.0155897 0.0746 0.2412 ROE Profits/ Total Equity 480 0.1160644 0.0500688 0.0251 0.2458 LTA Total Loans/ Total Assets 480 0.4919451 0.0746354 0.2788833 0.7055043
size Assets of a bank/ Total
Assets of 15 banks 480 0.0666667 0.0670405 0.002421 0.2366242 NI Non- interest Income/
Total Income 480 0.2227931 0.19814 -0.7846 3.19
INEF Operating Expenses/
Operating Income 480 0.5038729 0.0779409 0.2538473 0.7595665 RIR Real Interest Rate 480 0.0336554 0.0153136 0.0035758 0.0703768
housepr House Price 480 0.0897906 0.0731536 -0.0771 0.2305
Infl Inflation 480 0.0234875 0.0171669 -0.0153 0.0627
M2 Money Supply 480 0.1596594 0.0499995 0.1088 0.2898
RGDP Real GDP 480 0.0574582 0.020267 0.0294533 0.1182012
3. Panel unit root test
Variables Levin, Lin and Chu t P- Value Im. Pesaran and Shin W-statz P- Value
NPL 3.4657 0.9997 3.3056 0.9995 ΔNPL -3.4933 0.0002 -8.3270 0.0000 RIR -10.8908 0.0000 8.8183 1.0000 Infl -8.8018 0.0000 -1.6573 0.0487 housepr -10.7507 0.0000 -9.7492 0.0000 M2 -7.5855 0.0000 1.8773 0.9698 RGDP -2.6032 0.0046 -0.2639 0.3959 Size -2.1007 0.0178 0.2835 0.6116 CAR -4.5428 0.0000 -3.6020 0.0002 NI -1.6689 0.0476 -3.6623 0.0001 ROE -19.6913 0.0000 -12.6701 0.0000 INEF 0.7847 0.7837 -2.4189 0.0078 ΔINEF -15.1180 0.0000 -14.3139 0.0000 LTA -3.7921 0.0001 -13.3079 0.0000
