The impacts of risk and competition on bank profitability in China

Several rounds of banking reforms in China have aimed to increase the competitive condition and further enhance stability in the Chinese banking sector, while the joint effects of competition and risk-taking behaviour on the profitability in the banking sector have not been studied well enough so far in the literature. The current study contributes to the empirical literature by testing the impacts of risk and competition on profitability in the Chinese banking industry (state-owned, joint-stock and city commercial banks) over the period 20032011 under a one-step Generalized Method of Moments (GMM) system estimator. The results do not show any robust finding with regards to the impacts of competition and risk on bank profitability, while it is found that Chinese bank profitability is affected by taxation, overhead cost, labour productivity and inflation. The study provides policy implications to the Chinese banking industry and different ownership types of Chinese commercial banks. JEL classification: G21, C23,

A c c e p t e d M a n u s c r i p t 9 city commercial banks (CCBs) and a large number of other financial institutions, such as credit cooperatives, foreign banks, trust companies, finance companies of enterprise groups, etc. The proportion of assets of large-scale state-owned commercial banks (SOCBs) in the total banking sector assets keeps decreasing from 2003 to the lowest point in 2011, which is 47.3%, while on the other hand, the proportion of assets of joint-stock and city commercial banks in the total banking sector assets keeps increasing from 2003 to the highest points of 16. 22% and 8.81% in 2011, respectively. In summary, several rounds of banking reforms in China have aimed to increase the competitive condition and lower risk-taking behaviour, which is supposed to have influence on bank profitability. Our study examines the impacts of competition and risk on the profitability of Chinese banks.

Empirical literature on bank profitability
The empirical literature investigating bank profitability can be divided into two streams: one of which focuses on the analysis of bank profitability in multiple country studies, while the second group of literature places emphasis on the examination of bank profitability in single country studies. Table 1 gives a summary of the studies mentioned above.

<<Table 1-about here>>
There are a number of studies investigating profitability in the Chinese banking sector. The studies can be divided into three groups according to the methods used, which are principal analysis, fixed effect estimator and Generalized Method of Moments (GMM) estimator. Shih et al. (2007) investigate the performance of the big four, joint-stock and city commercial banks in China under the principal analysis. The results indicate that joint-stock commercial A c c e p t e d M a n u s c r i p t 10 banks (JSCBs) have better performance than state-owned and city commercial banks. They argue that the impact of size on bank performance is insignificant.
The second stream of studies uses the fixed effect estimator to investigate the determinants of bank profitability in China. Sufian (2009) uses this method to evaluate the determinants of profitability for four state-owned and twelve joint-stock commercial banks in China over the period [2000][2001][2002][2003][2004][2005][2006][2007]. The findings show that Chinese commercial banks with higher levels of credit risk, higher levels of capitalization and larger size in terms of total assets have higher profitability, while Chinese commercial banks with higher levels of liquidity and higher levels of overhead costs have lower profitability. The findings finally show that both economic growth and inflation precede an improvement in bank profitability in China. The same method has been used by Sufian and Habibullah (2009) to assess the determinants of Chinese bank profitability over the period [2000][2001][2002][2003][2004][2005]. Their findings suggest that higher levels of risk, higher levels of capitalization and higher levels of liquidity lead to higher profitability for state-owned commercial banks (SOCBs), whereas higher levels of risk and lower levels of cost lead to higher profitability for joint-stock commercial banks (JSCBs).
Finally, they find that larger bank size, in terms of total assets and higher levels of costs, lead to lower profitability for city commercial banks (CCBs), while city commercial banks (CCBs) with higher levels of capitalization and more diversified business have higher profitability.
Most of the research papers on Chinese banking profitability use the Generalized Method of Moments (GMM) estimator. Garcia-Herrero et al. (2009) use the two-step GMM system estimator to explain the low profitability of Chinese commercial banks over the period [1997][1998][1999][2000][2001][2002][2003][2004]. Their results show that higher profitability can be achieved by the banks with higher levels of capitalization, higher X-efficiency and larger shares of deposits. The findings also report that Chinese commercial banks have higher profitability in a less concentrated banking market. Finally, the results indicate that joint-stock commercial banks (JSCBs) have higher Page 11 of 77 A c c e p t e d M a n u s c r i p t 11 profitability than state-owned commercial banks (SOCBs) over the examined period. The two-step GMM system estimator is also used by Tan and Floros (2012a) to test the impact of inflation on bank profitability in China over the period [2003][2004][2005][2006][2007][2008][2009]. Their results show that Chinese banks with lower levels of diversification, lower levels of overhead cost and lower levels of taxation have higher profitability; while they find that higher developed banking market and higher developed stock market lead to profitability improvement in Chinese commercial banks. Finally, the results show that Chinese commercial banks have higher profitability in a higher inflationary environment. Rather than using the two-step GMM system estimator, Tan and Floros (2012b) use the one-step GMM system estimator to examine the impact of GDP growth on bank profitability in China over the period 2003-2009. Their findings suggest that Chinese commercial banks have lower profitability during the periods of economic boom (higher GDP growth rate).
Instead of using two-step GMM system estimators, Tan and Floros (2012c) use both the onestep difference estimator and one-step system estimator to investigate the impact of stock market volatility on bank performance in China over the period -2009 performance indicators are used, namely the return on equity (ROE), excess return on equity (EROE), Net Interest Margin (NIM) and Economic Value Added (EVA). Their results show that state-owned and joint-stock commercial banks with higher levels of taxation have lower ROE and EROE, while joint-stock commercial banks (JSCBs) with higher levels of capitalization have lower ROE and EROE. In addition, they report that state-owned and jointstock commercial banks with lower levels of overhead cost and higher levels of labour productivity have higher EVA and NIM. They also find that state-owned commercial banks (SOCBs) with higher levels of diversified business have lower EVA and NIM. Finally, the findings report that lower levels of risk, lower levels of taxation and a higher developed banking sector lead to higher EVA and NIM for joint-stock commercial banks (JSCBs).

Page 12 of 77
A c c e p t e d M a n u s c r i p t 12 Both the two-step GMM estimator and fixed effect estimator are used by Heffernan and Fu (2010) to test the determinants of performance for state-owned, joint-stock, city and rural commercial banks over the period [1999][2000][2001][2002][2003][2004][2005][2006]. Their findings show that banks with higher efficiency have better performance and bank listing contributes to the performance improvement. They further find that Chinese bank profitability is significantly affected by real GDP growth rate and unemployment, while the impacts of bank size and off-balancesheet activities on bank profitability are insignificant. Finally, compared to state-owned, joint-stock and city commercial bank, the rural commercial banks in China have better performance.

Empirical literature on bank competition and its measurement
The empirical literature uses a number of methods to estimate competition in the banking sector through analysing market power and efficiency. Bresnahan (1982) and Lau (1982) develop a method to estimate bank competition through measuring market power. This method has been recently used by Bikker (2003), Uchida and Tstsui (2005) and Qin and Shaffer (2014). The aggregate level of bank behaviour is analysed and average conjectural variation of banks is estimated through this method. This model is based on two structural equations, namely inverse demand equation and a supply equation, both of which are derived from the first order condition of profit maximization. This model estimates the mark-up of price over marginal cost as a measure of market power. Thus, this method is also called mark-up test. If the output price equals marginal cost, as reflected by a zero value of conjectural variation, there is perfect competition in the market, while the value equals to one, which indicates that there is a monopoly in the market.
The second approach that measured the bank competition is developed by Panzar and Rosse (1987). The model measures the extent to which a change in a vector of input prices is Page 13 of 77 A c c e p t e d M a n u s c r i p t 13 reflected in gross revenue. Thus, this method is also called the revenue test. The H-statistic is used in the method to estimate bank competition. The H-statistic is defined as the sum of the elasticities of the reduced-form revenues with respect to the input prices. A value equal to or smaller than zero indicates that the market is operated under monopoly, while A value of Hstatistic between zero and one suggests that the market is in a condition of monopolistic competition, and if the value of H-statistic equals 1, it shows that the market is perfectly competitive. This approach has been widely used in the empirical literature to measure banking sector competition (see Bikker and Haaf, 2002;Matthews et al., 2007;Goddard and Wilson, 2009;Moch, 2013;Barbosa et al., 2015, among others).
The third type of indicator to measure bank competition is through the analysis of market power. There are two indicators, which are Hirschman-Herfindahl index (HHI) and concentration ratio, both of which measure the degree of market concentration. These two indicators are used based on the structure-conduct-performance (SCP) hypothesis, which makes the assumption that the banks' behaviour is affected by market power, while the market structure plays a decisive role in bank performance. The idea of this hypothesis is based on the fact that in a more concentrated market where significant shares are occupied by a few banks, the competitive condition is lower, while higher concentration leads to greater market power, and the resultant increase in the collusive behaviour leads to higher profits.
These indicators were recently used by Al-Muharrami et al. (2006) andFu et al. (2014) to measure competition in the banking industry.
Market power can be also reflected from profitability due to the fact that higher profits achieved by the bank may indicate that there is a lower level of competition. The price-cost margin is an indicator used to measure profitability. The profit-cost margin can be estimated by the difference between output price and marginal cost, then divided by the output price.
The so-called Lerner index is widely used in the empirical literature to measure the M a n u s c r i p t 14 competition. The Lerner index ranges from zero to one. The Lerner index equals zero under the condition of perfect competition; the degree of competition decreases as the increase in the value of Lerner index. The market is operated under monopoly if the value of the Lerner index equals one. Empirical literature using the Lerner index to measure competition includes Cipollini and Fiordelisi (2012);Fungacova et al., (2014); among others.
The SCP hypothesis, as discussed above, suggests that market structure affects competitive behaviour which further influences bank performance. In other words, this hypothesis argues that bank profitability is derived from market structure. However, Demsetz (1973) develops the efficient-structure hypothesis, which argues that different profitability achieved by the bank is derived from efficiency. Thus, this hypothesis suggests that banks with higher efficiency have higher ability to increase their market shares and bank size, which further leads to excess profit. A new indicator was developed by Boone (2008) to measure bank competition. The so-called Boone indicator considers the impact of efficiency on performance with regards to profitability and market share. It also considers that competition improves the performance of efficient firms and weakens the performance of inefficient ones.
The Boone indicator can be positive as well as negative. A more negative Boone indicator indicates that there is a higher level of competition, while a larger positive value indicates the competition is lower. Considerable literature recently has used this measurement of competition in the banking sector, including Delis (2012), Tabak et al., (2012), among others.

The impact of competition on bank profitability
There are some hypotheses in the empirical literature investigating the influence of competition on firm's profitability. Using concentration ratio and Herfindal index as the indicators of market structure, the Structure-Conduct-Performance (SCP) hypothesis argues that through offering lower deposit rates and charging higher loan rates, firms have the ability to extract monopolistic rents in a concentrated market. In other words, this hypothesis M a n u s c r i p t 15 suggests that lower competition resulting from higher concentration in the market leads to market power, which enables firms to earn monopolistic or abnormal profit. There are a number of pieces of research providing support to the SCP hypothesis, and the research is undertaken by Rose and Fraser (1976);Heggestad and Mingo (1977); Berger and Hannan (1989); Lloyd-Williams et al., (1994);and Samad (2005), among others. Moreover, Giber (1984) conducted a survey and provided a summary of 44 studies with regards to the relationship between market concentration and bank performance. The findings show that among the 44 studies, 32 of them reported that there is a significant and positive impact of concentration on bank performance.
Alternatively, it is argued that the significant impact of competition (concentration) on firm profitability does not arise from market power but from higher efficiency of firms with larger market share. This Efficient Structure Hypothesis (ESH), as indicated by Demsetz (1973), suggests that the firms with higher efficiency have higher ability to increase their market shares and firms' sizes, while this higher efficiency allowed the firms to concentrate and the resulting lower competition leads to higher profit (Lloyd-Williams et al., 1994). Lloyd-Williams et al. (1994) further argue that the profit can be maximized by the banks with higher efficiency mainly through two ways: 1) maintaining the current market size and pricing policies; 2) accommodating size expansion and price reduction strategies, while Berger (1995) suggest that the profit as well as the market share can be increased by a more efficient bank with superior management or production technology. The above two statements (Williams et al., 1994;Berger, 1995) can also be explained from the perspectives of two different efficiencies, namely X-efficiency and scale efficiency. The X-efficiency is related to superior management of production and technologies, the banks with higher X-efficiency have higher ability to lower the operation cost and increase the profit. On the other hand, the scale efficiency assumes that the banks have the same management level and technology; however, M a n u s c r i p t 16 they have different operation scales, and some banks produce in a more efficient scale than their counterparts, the higher efficient scales lead to a reduction of unit cost and an increase in unit profit. The efficiency-structure hypothesis is supported by studies of Brozen (1987) and Seelanatha (2010). Both of these two hypotheses hold the viewpoint that higher efficiency or larger market power increases concentration, which leads to a reduction in competition.
However, the contestable market theory (CMT), as developed by Baumol (1982), argues that if there was no barrier for new entrants to enter the market, a concentrated industry can behave competitively. In other words, in a higher concentrated banking market which is dominated by few large banks, there is still a higher level of competition. This positive relationship between concentration and competition can be further explained by the fact that in a contestable market, some firms are driven out of the market because of more competition, while a higher concentration is the result of more competition.
Through reviewing the empirical literature on profitability in the banking sector, most of the studies find that banks with higher risk have lower profitability, while the impact of competition on bank profitability is ambiguous. Furthermore, the empirical researches report that there are comprehensive factors influencing bank profitability, such as bank size, capitalization, liquidity, overhead cost, inflation and GDP growth rate.
Although there are studies investigating the impacts of risk and competition on bank profitability, they use traditional accounting ratio to measure the bank risk and concentration ratio as the measurement of competition. This study provides the robust analysis regarding the impacts of risk and competition by firstly attempting to use stability inefficiency as a risk indicator, which is cross checked by ratio of loan loss provision over total loans and Z-score.
More importantly, we use the Lerner index which firstly evaluates the competitive conditions Bank size: we use the natural logarithm of total assets to measure this variable. This measurement is widely used in the empirical literature (see Goddard et al., 2004;Athanasoglou et al., 2008;Dietrich and Wanzenried, 2011 among others). On the one hand, banks with larger size are able to reduce costs from economies of scale and scope (Akhavein et al., 1997;Bourke, 1989;Molyneux and Thornton, 1992;Bikker and Hu, 2002;Goddard et al., 2004;Iannotta et al., 2007;Mercieca et al., 2007;Elsas et al., 2010). On the other hand, Barros et al. (2007) argue that the asymmetric information problems associated with leading can be reduced by smaller and specialized banks, preceding a negative impact of size on bank profitability. Furthermore, Berger and Humphrey (1994) argue that small banks can obtain economies of scale by increasing their size to a certain point where further increase in size will result in diseconomies of scale. This is supported by Athanasoglou et al. (2008), who argue that profitability initially increases with size and then declines for bureaucratic and other reasons. So there is not a prior expectation on the impact of this variable on bank profitability.
Liquidity: we use the ratio of total loans over total assets to measure this variable. We follow A c c e p t e d M a n u s c r i p t 18 lower liquidity level. However, a large volume of loans implies that there is more interest revenue generated. Hence, a negative impact of liquidity on bank profitability is expected, which is consistent with the findings of Molyneux and Thornton (1992). However, it is in direct contrast with the findings of Bourke (1989) in relation to the European banking industry who argues that banks with higher liquidity levels have higher profitability. Higher volume of loans will lead to a decline in bank profitability if the bank does not have a good risk management system. So there is no a prior expectation for this variable.
Risk: we use the ratio of loan loss provision over total loans (LLPTL) to measure this variable. A higher ratio suggests that the bank has higher risk. Empirical studies indicate that an increase in risk exposure leads to a decrease in bank profitability (see Miller and Noulas, 1997). Therefore, we expect that there will be a negative impact of LLPTL on bank profitability in China. Due to the fact that in Chinese banking industry, all the banks are required to hold more than enough loan loss provision to enhance the risk management, this volume will be set at the beginning of the year, thus this variable should be treated as a predetermined variable. We complement this measurement and check the robustness of the result by employing another two alternative risk indicators, which are Z-score and stability inefficiency derived from frontier estimation 6 . The Z-score is used by a number of empirical studies as the risk/stability indicator in the banking sector (see Iannotta et al. 2007;Liu et al. 2013;Liu and Wilson, 2013) and it is calculated by using the sum of a bank's return on assets and equity to total assets ratio over the standard deviation of the bank's return on assets. A higher Z-score indicates that there is a higher stability and lower risk. However, Fang et al.
(2011) argue that the potential stability that each bank can achieve is not reflected by Z-score, and the deviation from the bank's current stability and the maximum stability given the 6 See Appendix for more detail about the estimation.
A c c e p t e d M a n u s c r i p t economic and regulatory conditions must be considered. Therefore, the so-called "stability inefficiency" is invented. So negative impact of risk on profitability is expected.
Capitalization: we use the ratio of shareholders' equity over total assets to proxy this variable (see Athanasoglou et al., 2008;Garcia-Herrero et al., 2009;Dietrich and Wanzenried, 2011).
This variable is important in explaining the performance of financial institutions. A positive impact of capitalization on bank profitability is expected for the following reasons: first, the funding cost can be reduced for the banks with higher capital levels due to the fact that a higher capital ratio indicates that the banks have higher creditworthiness. Second, the banks with higher capital levels are more likely to engage in prudent lending, which leads to an increase in bank profitability. Thirdly, capital plays an important role in absorbing the risk arising from higher risk assets, such as loans; the interest revenue generated from loans fosters bank profitability. Finally, the banks with higher capital levels need to borrow less, which reduces cost and further increases profitability. However, Berger (1995) argues that capitalization can have a negative impact on bank profitability because the higher the level of capitalization, the lower the relative risk position of the bank. According to the risk-return trade-off, lower risk leads to lower return. This is supported by Modigliani and Miller (1963) and more recently by Dietrich and Wanzenried (2011). So we do not have a prior expectation on the impact of this variable on bank profitability. While the opposite causation is produced according to the relaxation of one-period assumption, which argues that the increase in earnings is allowed to increase the capital ratio. Thus, capital is modelled as the endogenous variable 7 .
7 Clear process has been followed to find the endogenous variables. The same model has been tested twice, with the first time the capital treated as exogenous variable and second time as endogenous variable. The results of Sargan test for over-identifying restrictions indicate that the capital should be treated as endogenous variable. Furthermore, the same test has been conducted on other variables as well and the results show that the ratio of loan loss provisions over total loans should be treated as predetermined variable. M a n u s c r i p t 20 Cost management: we use the ratio of overhead cost over total assets to measure this variable. This measurement has been used extensively in the empirical literature (Dietrich and Wanzenried, 2011;Liu and Wilson, 2010;Garcia-Hererro et al., 2009;Kosmidou, 2008). Athanasoglou et al. (2008) argue that well-managed banks have the ability to reduce operating costs, which led to an increase in bank profitability in Greece. This opinion is supported by Bourke (1989);and Jiang et al. (2003), among others. However, there is a positive impact of operating expenses on profitability in the European banking industry (Molyneux and Thornton, 1992).This finding can be explained by the fact that large volume of operating expenses is derived from paying salaries and wages to staff, while higher levels of salaries and wages paid to staff significantly improve the staff productivity, while the resulting improvement in the bank profitability is much larger than the salaries and wages paid out by the bank. This explanation is also in line with the efficiency wage theory. The positive impact of operating expenses on profitability is also supported by Ben Naceur (2003) and Guru et al. (2003) in terms of the Tunisian and Malaysian banking industries, respectively. So there is no a prior expectation for this variable.
Diversification: we measure this variable by using the ratio of non-interest income over gross revenue. As argued by Tan and Floros (2012a), more income can be generated when banks are engaged in a number of different businesses. In addition, the banks with more diversified activities have the ability to reduce their costs from economies of scope. Thus, a positive impact of diversification on bank profitability is expected. This is in line with the findings of Jiang et al. (2003) in terms of the Hong Kong banking industry. However, Gischer and Jutter (2001) and Demirguc-kunt and Huizinga (1999) argue that there is a negative relationship between diversification and bank profitability due to the fact that, compared to the traditional interest income activity, there is stronger competition in the area of fee-income generating M a n u s c r i p t 21 business, which precedes a decrease in bank profitability. So there is no a prior expectation for this variable.
Labour productivity: we use the ratio of gross revenue over total number of employees to measure this variable. This variable has been widely used in the empirical literature to examine its impact on bank profitability (see Athanasoglou et al., 2008;Tan and Floros 2012a, 2012b, 2012c. Higher labour productivity not only reflects efficient bank management, but also increases the bank's efficiency and further fosters the bank's profitability. So we expect that this variable has a positive impact on bank profitability.
Taxation: the ratio of tax over operating profit before tax is used to measure this variable.
This measurement has been used by Tan and Floros (2012a, 2012b, 2012c in the Chinese banking industry. The findings of Tan and Floros (2012a, 2012b, 2012c show that there is a significant and negative impact of taxation on bank profitability in China. This can be explained by the fact that higher level of taxes incurred by the banks increase the bank cost and further leads to a reduction in bank profitability, thus, we expect that this variable has a negative impact on bank profitability.

Industry-specific determinants
Competition: we use Lerner index/Herfindahl-Hirschman index to measure this variable. The SCP hypothesis argues that firms in a financial system with less competition (higher Herfindahl-Hirschman index and higher Lerner index) tend to have larger scales of operation, which leads to a higher degree of profit. This is supported by Claessens and Laeven (2004).
However, on the other hand, the competition-efficiency hypothesis argues that in a higher competitive environment, banks managers have more incentive to improve efficiency; the cost reduction derived from efficiency improvement further precedes an increase in profitability. So we do not have any a prior expectation on the sign of this variable.

Page 22 of 77
A c c e p t e d M a n u s c r i p t

22
Banking sector development: we use the ratio of banking sector assets over GDP to measure this variable. This is a country-specific rather than a region-specific variable and it is widely used by empirical studies (see Tan and Floros, 2012a, 2012b, 2012c. Tan and Floros (2012a) suggest that there is a significant and positive impact of banking sector development on bank profitability in China. They argue that in a higher developed banking sector, the demand for banking services increases, which will attract more potential competitors to enter the market.
However, although the Chinese banking sector has been gradually opened up through several rounds of banking reforms, it is still difficult for new banks to enter the market. The reduction in supply of banking services relative to the increased demand increases the prices of banking services and further increases the profitability of existing banks. So we expect that this variable has a positive impact on bank profitability.
Stock market development: we use the ratio of market capitalization of listed companies over GDP to measure this variable. According to Demirguc-Kunt and Huizinga (1999) and Bashir (2000), banks in countries with well-developed stock markets normally have higher profitability. The reasons can be explained as follows: 1) higher developed stock market increases the number of firms to obtain funds from stock market rather than banks, this not only reduces the volume of loan service provided by banks, but decreases the risk of loan default, the risk reduction leads to an increase in bank profitability. Furthermore, the risk reduction also increases the borrowing capacity of banks, which fosters bank profitability.
Finally, the stock market provides more information on the public traded firms, which makes it easier for banks to evaluate and monitor the risk. Thus, it precedes a reduction of bank cost and an increase in bank profitability (Tan and Floros, 2012a). So we expect that this variable has a positive impact on bank profitability. In other words, it can be concluded that banks and the stock market complement each other and therefore they grow together when they develop.
This correlation between banking sector development and stock market development is M a n u s c r i p t 23 confirmed by the research undertaken by Demirguc-Kunt and Levine, 1996;Garcia and Liu, 1999;Li, 2007; among others).

Macroeconomic determinants:
Inflation: we use the annual inflation rate to measure this variable. Inflation is an important determinant of bank performance. The impact of inflation on bank profitability is firstly examined by Revell (1979) and further investigated by Perry (1992). Both of them argue that the effect depends on whether inflation is anticipated or unanticipated. If inflation rate is fully anticipated, banks can adjust the interest rates or manage the operating expenses accordingly to make the revenues increase faster than costs, which leads to higher profitability, while if inflation is not fully anticipated, the loan losses will be accumulated, which leads to a decrease in bank profitability, thus, there is no a prior expectation for the impact of inflation on bank profitability. GDP growth rate: some of the researchers argue that it has a positive impact on bank profitability due to the fact that the demand for lending increases during cyclical upswings (see Dermirguc-Kunt and Huizinga, 1999;Bikker and Hu, 2002;Athanasogolou et al., 2008).
However, Tan and Floros (2012b) find that GDP growth rate has a negative impact on bank profitability in China. They suggest that higher economic growth improves the business environment and lowers the bank entry barriers. The consequently increased competition dampens bank's profitability. So we do not have any a prior expectation for this variable.

Variable selection
The main goal of this paper is to test the impacts of risk and competition on bank profitability in China while controlling for comprehensive bank-specific, industry-specific and macroeconomic variables. There are four profitability indicators considered in the study:  Table 2). ROA, ROE and NIM are used to compare the results with the findings reported in the literature, while the inclusion of PBT as one of the profitability indicators follows the study of Demirguc-Kunt and Huizinga (1999).
ROA shows the profits earned per unit of assets and reflects the management ability to utilize banks' financial and real investment resources to generate profits (Hassan and Bashir, 2003).
ROA has emerged as the key ratio for the evaluation of bank profitability and has become the most common measure of bank profitability in the literature (Athanasoglou et al., 2008;Garcia-Herrero-et al., 2009;Golin, 2011).    Athanasoglou et al., 2008;Tan and Floros 2012a, 2012b, 2012c. It reflects how successful a bank's investment decisions are relative to its interest expenses. The difference between ROA and NIM is that the former emphasizes the profit earned per unit of assets while the latter focuses on the profit earned on the interest generating activities. Figure 1c shows that the NIM of city commercial banks (CCBs) was highest over the period 2004-2011, while the NIM of state-owned commercial banks (SOCBs) is slightly higher than joint-stock commercial banks (JSCBs) in general.
<<Figure 1c---about here>> Finally, the profit margin is used as one of the profitability indicators. The difference between ROA and profit margin lies in the fact that the latter considers the tax expenses in bank profitability. We include this indicator to see whether tax has an influence on Chinese bank profitability. When measured by the profit margin, we find from Figure 1d that the profitability of state-owned commercial banks (SOCBs) is higher than city commercial banks (CCBs) in general, while the joint-stock commercial banks (JSCBs) have the lowest profitability. This result indicates that joint-stock commercial banks (JSCBs) pay higher taxes. <<Figure 1d---about here>> In summary, we use comprehensive profitability indicators to examine the Chinese bank profitability. They reflect different aspects of banking operations. To be more specific, ROA represent bank's ability to generate profits from assets, while ROE is related to banks' financing and leverage decisions, NIM more specifically focuses on lending activities, while Page 26 of 77 A c c e p t e d M a n u s c r i p t 26 profit margin consider the tax effect on bank profitability. The different profitability indicators reflect the Chinese banking profitability from different perspectives, which gives the widest overview of Chinese bank profitability to government and it is helpful for the government to make accurate policies.

Empirical model
This study follows and expands the specification proposed by Athanasoglou et al. (2008), which can be expressed as follows: (1) Where i refers to year and t refers to an individual bank, it II represents the profitability indicator for the specific bank at a specific year, C is constant term,    Figure 3a shows that the volume of loan loss provision is the highest in city commercial banks after 2004. We explain this finding by mainly two reasons: 1) compared to the state-owned and joint-stock commercial banks, the city commercial banks (CCBs) have a short history and they lack the ability of risk management; 2) the city commercial banks (CCBs) normally make loans to local enterprises within the city, and the small enterprises have higher probability of default on loans. We further report that, the jointstock commercial banks (JSCBs) have higher risk than state-owned commercial banks  Table 3 shows the profitability conditions of Chinese banks over the examined period.
Furthermore, the profitability of three different groups of Chinese banks is also presented.
From the table we can see that over the whole examined period, the state-owned commercial banks (SOCBs) and city commercial banks (CCBs) have a higher profitability in terms of ROA and ROE than joint-stock commercial banks (JSCBs), while the NIM of city commercial banks (CCBs) is higher than joint-stock and state-owned commercial banks.
With regards to PBT, the state-owned commercial banks (SOCBs) have the highest profitability, while the profitability of joint-stock commercial banks (JSCBs) is the lowest.
The higher profitability in term of ROA and ROE in state-owned commercial banks (SOCBs) can be largely attributed to the fact that the overall income generated by the state-owned commercial banks (SOCBs) is higher, which indicates that state-owned commercial banks  M a n u s c r i p t 33 Furthermore, the Boone indicator also suffers from two disadvantages. First, as argued by Tabak et al. (2012) it makes the assumption that part of the efficiency gains achieved by the banks will be passed onto consumers. In addition, this indicator also suffers from idiosyncratic variation, i.e. uncertainty.
The Lerner index is used in this study mainly because of the following reasons: 1) it can be easily estimated by each bank at each year; and also it matches with its determinants, which are bank-level variables at each year; 2) we can estimate the competitive conditions (market power) for three different ownership types of Chinese banks. Moreover, compared between the K-bank concentration ratio and Herfindahl-Hirchman index, the latter is preferred and used in the current study due to the fact that it takes into account the relative size of the firms in an industry.

5.21a Lerner index
The Lerner index is defined as the difference between a bank's price and the marginal cost, divided by the price. The index value ranges from a maximum of 1 to a minimum of zero, with higher numbers indicating greater market power and hence less competition. The Lerner index represents the extent to which a particular bank has market power to set its price above the marginal cost.
The price is computed by estimating the average price of bank production (proxied by total assets) as the ratio of total revenue over total assets following Fernandez de Guevara et al., Once the marginal cost is estimated and the price of output computed, we calculate the Lerner index for each bank and obtain a direct measure of bank competition. We use the same three input prices to calculate the marginal cost, which are the price of funds, price of capital and price of labour. The are the coefficients estimated from equation (2). The definition of the variables used to estimate the Lerner index is summarized in Table 5.

5.21b Herfindahl-Hirchman index (HHI)
The HHI sums the squared market shares of firms in the relevant market.
where MS is the market share of the firm and n is the number of firms in the market. Al-Muharrami et al. (2006) indicate that this indicator assigns a greater weight to larger banks than smaller banks. In other words, it attaches importance to the larger banks. Furthermore, each bank is incorporated individually to avoid the arbitrary cut-offs and insensitivity to the share distribution.

Measurement of stability inefficiency
Fang et al. (2011) argue that the potential stability of banks cannot be necessarily reflected by the Z-score. The deviation from the bank's current stability and the maximum must be considered. We provide a measure of the bank's stability inefficiency by estimating a stochastic frontier (Aigner et al., 1977;Meeusen and Van den Broeck, 1977) with the Z-score as the dependent variable of the translog specification.
where W represents the input price; we consider two input prices which are the price of funds (interest expenses to total deposits) and the price of capital (non-interest expenses to total assets). Y represents four outputs, which are total loans, total deposits, other earning assets and non-interest income. The sub-index i and t represent bank i operates at time t, while j and k represent different output. indicates that the bank risk is higher, while lower stability inefficiency means that the bank is more stable.

Method to investigate the determinants of bank profitability in China
Empirical literature uses a variety of methods to investigate the determinants of bank profitability. Fixed effects is used by Sufian (2009)  GMM estimator is used in the current study due to the fact that a number of problems in estimating the determinants of bank profitability including endogeneity, unobserved heterogeneity, autocorrelation and profit persistence cannot be solved by fixed effects. To be more specific, comparing between difference and system GMM estimators, we prefer the latter because the system GMM estimator addresses the issue of unit root property and produces more precise results (Bond, 2002), while compared to two-step GMM estimator, the one-step estimator is chosen due to the fact that it produces a smaller bias and a smaller standard deviation of the estimation (Judson and Owen, 1999). Besides using the one period lag of profitability indicators, through the Sargan over-identifying test, we confirm that the capital will be treated as endogenous variable, while one of the risk indicators (the ratio of loan loss provisions over total loans) will be treated as predetermined variable, other variables do not suffer any endogenous issue. In order to make sure there is no second order Page 37 of 77 A c c e p t e d M a n u s c r i p t 37 autocorrelation in the estimation, the predetermined variable is instrumented using levels lagged by one year period, while the endogenous variable is instrumented using levels lagged by two years periods.  (Arellano and Bond, 1991), but this case is rejected by the test of AR(2) errors.

The significant coefficients of the lagged dependent variables (ROA, NIM and PBT) confirm
the dynamic character of model specification. δ takes values of 0.08, 0.4 and 0.34 when profitability is measured by ROA, NIM and PBT, respectively, which shows that the profitability of Chinese banks does not persist, it implies that departure from a perfectly competitive market structure in the Chinese banking sector is not very large. In other words, the Chinese banking sector has a relatively competitive structure.
Turning to other explanatory variables, the coefficient of bank size is significant and negative, suggesting that there is a negative impact of bank size on ROA, NIM and PBT. This finding can be explained by the fact that the smaller banks, such as city commercial banks (CCBs), are easier to be managed and bank managers can concentrate on a smaller number of business engaged, which leads to a higher profitability. Liquidity is found to be significantly and negatively related to ROA, NIM and PBT; this result indicates that higher degree of loan exposure (lower liquidity) leads to an increase in bank profitability. It reflects the fact that Turing to the macroeconomic variables, inflation is found to be significantly and positively related to bank profitability (ROA, NIM and PBT). This result indicates that during the examined period, the inflation is fully anticipated by Chinese banks, so the bank regulatory authorities and bank managers adjust the interest rates accordingly. Furthermore, it also reflects that Chinese bank managers manage the expenses very well during this period, which leads to an increase in bank profitability. Pasiouras and Kosmidou (2007) have the same finding for the EU banking industry as well as Sufian (2009)  A c c e p t e d M a n u s c r i p t 41 competition, as measured by Herfindahl index, has a significant and positive impact on PBT of Chinese commercial banks, which is in contrast with our finding in Table 6. <<Table 7---about here>> Table 8 reports the empirical results regarding the impacts of risk and competition on bank profitability using Z-score as risk indicator and Lerner index as competition indicator.
Compared to the results reported from Table 6 and Table 7 higher economic growth leads to higher NIM and PBT of Chinese commercial banks. Finally, the results confirm that joint-stock commercial banks and city commercial banks have lower profitability compared to state-owned commercial banks.
<<Table 8---about here>> Table 9 reports the empirical results regarding the impacts of risk and competition on bank profitability using LLPTL as risk indicator and Lerner index as competition indicator. The following findings are in accordance with the results reported from Table 6, Table 7 and   credit situation, and this information substantially reduces the cost of monitoring risk and further leads to an increase in NIM.
<<Table 10---about here>> 2) overhead cost and labour productivity have significant and positive impact on NIM.
<<Table 11---about here>> The impacts of competition and risk on profitability of city commercial banks (CCBs) are reported in Table 12. The table shows that stability inefficiency has a significant and negative impact on ROE, which indicates that higher insolvency risk reduces the profitability of city commercial banks (CCBs). Furthermore, we notice that bank size is significantly and negatively related to profitability of Chinese city commercial banks (CCBs). This can be explained by the fact that smaller city commercial banks (CCBs) normally operate within the city where they were found; however, larger city commercial banks (CCBs) have branches in other cities as well. The local city government provides support and give beneficial policies (i.e. tax reductions) to their own city commercial banks (CCBs), while city commercial banks (CCBs) from other cities do not have this cost advantage, and the resulting increase in cost leads to a decline in bank profitability. Liquidity is found to be significantly and negatively related to NIM of city commercial banks (CCBs), indicating a large volume of loans leads to improvement in NIM of city commercial banks (CCBs). This finding can be explained by the fact that city commercial banks (CCBs) have higher ability of loan monitoring and credit checking. This table also has similar findings compared to the previous tables which are: 1) taxation has significant and negative impacts on ROA and ROE; 2) overhead cost and labour productivity have significant and positive impact on NIM. We noticed that diversification has a significant and negative impact on NIM of city commercial banks (CCBs). This can be explained by the fact that loan business is still the focus for city commercial banks (CCBs), M a n u s c r i p t 45 and more diversified activities reduces the volumes of funds available for making loans, which further precedes a decline in banks' NIM.
<<Table 12---about here>> 7 Discussion on the empirical results for Chinese banking industry Table 13 provides a summary for the empirical results using different risk and competition indicators across various profitability measurements. Starting from single profitability measurement, we find that one period lagged ROA is significantly and positively related to the current ROA (except for the case when Herfindahl index is used as the competitor indicator). The results indicate that the profitability of Chinese banks tends to be persistent to some extent and it confirms our correct choice of dynamic model in the empirical analysis.
We further report that the taxation is significantly and negatively related to ROA of Chinese banks. This can be explained by the fact that higher taxes paid by banks increase the noninterest expenses, which further precedes a decline in bank profitability. Labour productivity, which is another significant variable in terms of ROA, is found to be positively related to bank profitability. This finding indicates that Chinese banks should provide more training and professional opportunities to the staffs in order to increase their productivity and further improve profitability. In addition, Chinese banks should also recruit more staffs with higher productivity. In terms of the industry-specific variables, it is found that banking sector development is significantly and positively related to ROA of Chinese banks.
When using ROE as the profitability indicator, the results confirm the findings that higher taxes paid by banks leads to lower profitability, while banks with higher labour productivity have higher profitability. In addition, the results show that banks with higher overhead cost leads to higher bank profitability. This finding can be explained by the fact that the Chinese banks have the ability to transfer operational cost to the depositors; the resulting reduction in Page 46 of 77 A c c e p t e d M a n u s c r i p t 46 cost and increase in income promote the improvement in bank profitability. This finding also indicates that higher salaries paid to staff, especially experienced staff, significantly improve the productivity, while the resulting increase in the profitability is larger than the salary cost.
With regards to NIM, one period lag of NIM is significant for all the cases which indicates that the NIM of Chinese commercial banks at the current year is significantly affected by the situation at the previous year. Furthermore, the values of one period lagged NIM are less than 0.4, which indicates that Chinese bank profitability does not persist to a large extent. In other words, it is relatively competitive in the Chinese banking industry. Compared to ROE, we also find that banks with high overhead cost and labour productivity have higher profitability, while the results report that large banks have lower profitability. This can be explained by the fact that larger volumes of funds are used by larger banks to engage in diversified activities, while the reduction in the volumes of funds for traditional services further precedes a decline in bank profitability (NIM). In addition, the liquidity variable is significant and positive for all the cases indicating lower liquidity (higher volumes of loans) leads to higher profitability.
This finding shows that in general, Chinese banks have improved their ability and skills in credit checking, monitoring and managing loans, which leads to an improvement in bank profitability. The diversification is found to be significant and negative for all the cases, indicating the larger variety of business engaged by banks reduces the NIM. This can be explained by the fact more diversified business engaged by the banks reduces the volumes of funds available for traditional loan services, which further precedes a decline in bank profitability. The positive and significant impact of inflation on Chinese bank profitability reflects the fact that Chinese banks have the ability to forecast inflation, and the interests rates are adjusted accordingly, the revenues increase faster than costs, which leads to higher profitability.
M a n u s c r i p t 47 Finally, the table confirms a number of findings when using PBT as the profitability indicator, which can be summarized as follows: 1) the profitability of Chinese commercial banks tends to be persistent to some extent as reflected by the significant one period lagged dependent variable; 2) larger bank have lower profitability; 3) larger volume of loans (lower liquidity) leads to higher profitability; 4) Chinese commercial banks with higher taxes have lower profitability; 5) Chinese commercial banks with larger volumes of overhead cost and labour productivity have higher profitability; 6) higher inflationary environment leads to improvement in Chinese bank profitability. In addition, the results indicate that higher developed banking sector improves Chinese bank profitability, which is in line with our expectation, while higher GDP growth also has a significant and positive impact on Chinese bank profitability. This can be explained by the fact that during the period of economic boom, the volumes of leading increase, and the resulting increase in interest income further leads to an improvement in bank profitability. Table 13 shows that, compared to state-owned commercial banks, both joint-stock commercial banks and city commercial banks have lower profitability (ROA, NIM and PBT) as reflected by the significant and negative coefficients.

Conclusion
The aim of this paper is to test the impacts of competition and risk on bank profitability in China over the period 2003-2011. We check the robustness of the results by using different risk and competition indicators. To be more specific, we use two different measurements of bank competition, which are Lerner index and Herfindahl-Hirschman index. There are three alternative measurements in terms of bank risk: stability inefficiency, Z-score and ratio of loan loss provision over total loans. Besides the analysis on the effects of risk and competition on bank profitability, we also control for comprehensive bank-specific, industry-M a n u s c r i p t 48 specific and macroeconomic determinants of bank profitability. With regards to the econometric estimation, the one-step GMM system estimator is applied.
The results suggest that Chinese bank profitability persists to a small extent, while we do not find any robust impact of risk and competition in the Chinese banking industry, the unclear impact of risk on bank profitability can be attributed to the fact that the Chinese government still have strong influence or provide strong support to Chinese banks (especially state-owned commercial banks) through four assets management companies and capital injections. This study does not support the traditional SCP hypothesis; this may be explained by the fact that in the Chinese banking industry, the efficient-structure hypothesis may be prevailing, thus, future research can be conducted, including efficiency as one of the determinants to test this hypothesis. The findings further show that taxation has a significant and negative impact on Chinese banking profitability with regards to ROA, NIM and PBT, while Chinese banks with higher labour productivity have higher profitability. In addition, the results report that Chinese banks with higher overhead cost have higher profitability (ROE, NIM and PBT) and more diversified Chinese banks have lower profitability with regards to NIM. In terms of the macroeconomic environment, the findings suggest that Chinese banks have higher profitability (ROA, NIM and PBT) in a higher inflationary environment. Finally, we find that compared to state-owned commercial banks (SOCBs), joint-stock commercial banks (JSCBs) and city commercial banks (CCBs) have lower ROA, NIM and PBT.
The results with regards to the separate estimations on different ownership types of Chinese commercial banks show that government still has strong influence and give strong support to state-owned commercial banks (SOCBs), which to some extent reduces their profitability, while with regards to the joint-stock commercial banks (JSCBs), it is argued that non-interest income generating business earns higher income compared to the traditional loan services.
M a n u s c r i p t 49 Finally, it is found that large city commercial banks (CCBs) have lower profitability and city commercial banks (CCBs) with higher business diversification have lower NIM.
The findings have several policy implications to Chinese government, regulatory authority and bank managers in order to improve bank profitability: 1) Chinese government should reduce the tax rate for commercial banks; 2) recruit more productive and experienced staffs and provide more training opportunities to existing staffs. With regards to different ownership types of Chinese commercial banks, the following policies should be considered: 1) reduce the degree of influence and support by the government to the state-owned commercial banks (SOCBs); 2) encourage the joint-stock commercial banks (JSCBs) to engage in more   1993-2000McShane and Sharpe (1985 and Angbanzo (1997)  A c c e p t e d M a n u s c r i p t     Page 70 of 77 A c c e p t e d M a n u s c r i p t 70 •*,** and *** denote significance at 10%, 5% and 1% levels, respectively. •ª the endogenous variable is instrumented using levels lagged by two periods •¹ the test for over-identifying restrictions in GMM dynamic model estimation •² Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (H0: no autocorrelation) •³ Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (H0: no autocorrelation) Page 71 of 77 A c c e p t e d M a n u s c r i p t 71 164 321 *,** and *** denote significance at 10%, 5% and 1% levels, respectively. •ª the endogenous variable is instrumented using levels lagged by two periods •¹ the test for over-identifying restrictions in GMM dynamic model estimation •² Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (H0: no autocorrelation) •³ Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (H0: no autocorrelation) Page 72 of 77 A c c e p t e d M a n u s c r i p t 72 174 *,** and *** denote significance at 10%, 5% and 1% levels, respectively. •ª the endogenous variable is instrumented using levels lagged by two periods •¹ the test for over-identifying restrictions in GMM dynamic model estimation •² Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (H0: no autocorrelation) •³ Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (H0: no autocorrelation) Page 73 of 77 A c c e p t e d M a n u s c r i p t 73 173 *,** and *** denote significance at 10%, 5% and 1% levels, respectively. •ª the predetermined variable is instrumented using levels lagged by one period •º the endogenous variable is instrumented using levels lagged by two periods •¹ the test for over-identifying restrictions in GMM dynamic model estimation •² Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (H0: no autocorrelation) •³ Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (H0: no autocorrelation) Page 74 of 77 A c c e p t e d M a n u s c r i p t 74 29 36 *,** and *** denote significance at 10%, 5% and 1% levels, respectively. •ªthe endogenous variable is instrumented using levels lagged by two periods •¹ the test for over-identifying restrictions in GMM dynamic model estimation •² Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (H0: no autocorrelation) •³ Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (H0: no autocorrelation) Page 75 of 77 A c c e p t e d M a n u s c r i p t 75 35 53 *,** and *** denote significance at 10%, 5% and 1% levels, respectively. •ª the endogenous variable is instrumented using levels lagged by two periods •¹ the test for over-identifying restrictions in GMM dynamic model estimation •² Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (H0: no autocorrelation) •³ Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (H0: no autocorrelation) Page 76 of 77 A c c e p t e d M a n u s c r i p t 76 57 85 *,** and *** denote significance at 10%, 5% and 1% levels, respectively. •ª the endogenous variable is instrumented using levels lagged by two periods •¹ the test for over-identifying restrictions in GMM dynamic model estimation •² Arellano-Bond test that average autocovariance in residuals of order 1 is 0 (H0: no autocorrelation) •³ Arellano-Bond test that average autocovariance in residuals of order 2 is 0 (H0: no autocorrelation) Page 77 of 77 A c c e p t e d M a n u s c r i p t 77 •lag represent the lag of dependent variable; BSV stands for bank-specific variables which include Size (bank size); Risk (bank risk); Liquid (bank liquidity); Tax (bank taxation); Cap (capitalization); Cost (overhead cost); DIV (diversification); LP(labour productivity); ISV is the industry-specific variables which include Comp (competition); BSD (banking sector development); SMD (stock market development); Mac is the macroeconomic variables considered in the study which include INF (inflation) and GDP (annual GDP growth rate).
• "+" represents that the variable is significantly and positively affect the bank profitability; "-" indicates that the variable is significantly and negatively affect the bank profitability; "insignificant" means that the variable is not significantly related to bank profitability; "inconsistent" underlines that different results are obtained for different indicators being used.