学术沙龙研讨会

Market structure,risk taking,and the ef ?ciency of Chinese commercial banks

Xiaohui Hou a,?,1,Qing Wang b,2,Qi Zhang a,3

a School of Economics and Finance,Xi'an Jiaotong University,710061,PR China b

Xi'an Branch,The People's Bank of China,710075,PR China

a r t i c l e i n f o a

b s t r a

c t

Article history:

Received 29July 2013

Received in revised form 10June 2014Accepted 18June 2014

Available online 25June 2014We investigate the impacts of market structure and bank risk taking on the ef ?ciency of Chinese commercial banks,employing a two-stage semi-parametric data envelopment analysis model.Our empirical results show that the intense market competition compels Chinese commercial banks to develop advanced technical experience and skills,thus improving their technical ef ?ciency.Besides,the technical ef ?ciency is positive associated with the risk taking.Since more risk taking implies a credit expansion of Chinese commercial banks based on the soft risk constraint,the improvement of technical ef ?ciency may accompany an accumulation of banking risks in the current ?nancial system of China.

JEL classi ?cation:G21G28C14

Keywords:

Market structure Bank risk taking Technical ef ?ciency

Chinese commercial banks

1.Introduction

Remarkable progress has characterized the reform and opening up of China's banking sector to foreign competition following the country's entry into the World Trade Organization (WTO)in 2001.However,reforms in the banking sector have lagged behind those in other economic sectors.In particular,Chinese commercial banks have faced numerous challenges in their attempt to realize stable development.

China's banking industry opened its doors to the world with the implementation of the Regulations of the People's Republic of China on Administration of Foreign-funded Banks on 11December 2006.This move resulted in the emergence of more challenges,further stimulating the intense competition in the

Emerging Markets Review 20(2014)75–88

?Corresponding author at:P.O.Box 1787,School of Economics and Finance,Xi'an Jiaotong University,Xian Ning West Road No.28,710049Xi'an,Shaanxi Province,PR China.Tel.:+86139********.

E-mail address:hxiaoh2006@https://www.360docs.net/doc/cd10878725.html, (X.Hou).1

Associate Professor of ?nancial economics,Xi'an Jiaotong University.2

Analyst of ?nancial stability,Xi'an Branch,The People's Bank of China.3

PhD candidate in Finance,Xi'an Jiaotong

University.

https://www.360docs.net/doc/cd10878725.html,/10.1016/j.ememar.2014.06.0011566-0141/?2014Elsevier B.V.All rights

reserved.

Contents lists available at ScienceDirect

Emerging Markets Review

j o u r n a l h o m e p a g e :w w w.e l s e v i e r.c o m /l o c a t e /e m r

76X.Hou et al./Emerging Markets Review20(2014)75–88

entire banking industry.In addition,since the end of2007,the global?nancial crisis has disrupted the stability of the world?nancial markets,but China's banking sector has apparently experienced limited impact.Nevertheless,the business model of Chinese commercial banks is being questioned now more than ever,especially in terms of their main pro?t sources(i.e.,interest margin and arbitrary charges for ordinary customer services).

Therefore,the ability of Chinese commercial banks to perform ef?ciently needs to be improved, especially under the in?uence of double shocks caused by increasing competition in the global banking industry and the global?nancial crisis in recent years.Naceur and Omran(2011)already investigate the impacts of bank concentration and regulation on commercial banks'ef?ciency across a broad selection of Middle East and North Africa countries.Pessarossi and Weill(2013)had begun to be concerned about the effect of capital risk on Chinese banks'ef?ciency,while Fungacova et al.(2013)?nd that no signi?cant relation between competition and ef?ciency for a sample of Chinese banks,which differs from the results generally observed for other countries.Although a number of studies have concerned the determinants of Chinese bank ef?ciency,the effects of market structure and risk taking,including capital risk,credit risk, and liquidity risk,on the bank ef?ciency still remain an important,yet largely unexplored,?nancial issue, which motivates our study in this paper.

Allen et al.(2007)provide a comprehensive examination on all aspects of China's?nancial system. They principally conclude that China's?nancial system is dominated by a large,inef?cient banking sector. Despite the entry and growth of many domestic and foreign banks in recent years,China's banking sector is still mainly controlled by large state-owned banks.In addition,continuing efforts to control the risks of the major banks within normal levels,thereby avoiding a banking crisis,are the most important aspect of reforming the country's banking system.Hence,against this background,we investigate the impact of market structure and risk taking on the ef?ciency of Chinese commercial banks.

Market structure is one of the factors that may explain some of the ef?ciency differences that remain after controlling for the ef?ciency concept and measurement method(Berger and Mester,1997). Especially in China,banking sector reform as part of the reform of the entire?nancial system is carried out predominantly by the Chinese government,indicating that the change in market structure in China's banking industry is almost entirely exogenous.Therefore,there is a good opportunity to investigate the effect of the change in market structure on the ef?ciency of Chinese commercial banks.

Commercial banks can solve the potential moral hazards and adverse selection problems caused by the imperfect information between borrowers and lenders.These banks assess and manage risks,write contracts, and monitor contractual performance.The ability of banks to ef?ciently perform also depends in part on their amount of risk taking(Hughes and Mester,2010).Moreover,the amount of risk taking reveals their strategic niche,which is one of the important potential correlates of bank ef?ciency(Berger and Mester,1997).In recent years,the regulatory authorities have increasingly focused on the risk management of Chinese commercial banks,but the bank risk constraint is still soft derived from underlying credit guarantees by the government.Thus,another motivation of the current study is to examine the impact of risk taking on the bank ef?ciency.

Aside from these studies on the banks in developed countries,ef?ciency research on commercial banks in China has become increasingly popular in recent years.The two most frequently used approaches are stochastic frontier analysis(SFA)and data envelopment analysis(DEA)methods.The SFA approach requires a speci?c functional form,which may be unsuitable for the data during the industry transition period(Yao et al.,2008).Although the banking industry in China has undergone dramatic changes,many papers have used the DEA approach to estimate technical ef?ciency,and sometimes examine the impacts of some environmental variables on the ef?ciency estimates(Luo et al.,2011;Su?an and Habibullah,2011; Yao et al.,2008;Ye et al.,2012).

However,Simar and Wilson(2007,2011)argue that the conventional two-stage DEA approach to inference employed in these papers is invalid due to complicated,unknown serial correlations among the estimated ef?ciencies.To obtain a valid statistical inference,they propose a two-stage,semi-parametric model in which second-stage regressions are well-de?ned and meaningful,such that the form of the second-stage regression equation is determined by the structure of the model in the?rst stage where the initial DEA estimates are obtained.Empirical studies examining the determinants of the ef?ciency are abundant.Nevertheless,to our knowledge,papers that use a two-stage,semi-parametric model to investigate the impacts of market structure and risk taking on the technical ef?ciency of Chinese commercial banks have

not been published.Furthermore,outliers are often treated as particularly troublesome to the DEA model (Bogetoft and Otto,2011);hence,they should be removed from the data when the DEA model is employed,which is typically ignored in the present literature.

The current paper aims to ?ll the gap in the literature,estimate the technical ef ?ciency of Chinese commercial banks,and investigate the effects of market structure and risk taking on the ef ?ciency estimates,employing a two-stage semi-parametric DEA model after the outliers are removed from the sample observations.The empirical results show that the average technical ef ?ciency of Chinese commercial banks in the sample period is 0.886;the mean technical ef ?ciency has shown a declining trend.

Our main empirical focus is the impact of banking market structure and bank risk taking.The results suggest that the intense market competition reduces the market power of the banks,which compels them to develop advanced technical experience and skills,thus improving their technical ef ?ciency.In addition,the technical ef ?ciency is positive associated with the risk taking.Since more risk taking implies a credit expansion of Chinese commercial banks based on the soft risk constraint derived from underlying guarantees by the government,the improvement of technical ef ?ciency may accompany an accumulation of banking risks in the current ?nancial system of China.

This paper is organized as follows.Section 2presents the literature review.Section 3outlines the empirical methodology and shows the data on Chinese commercial banks.Section 4discusses the empirical results.Section 5concludes.2.Literature review

Over the past two decades,a number of studies have measured the ef ?ciency of commercial banks and analyzed the determinants of bank ef ?ciency (Banker and Natarajan,2008;Berger and Humphrey,1997;Berger et al.,2009;Drake and Hall,2003;Hughes and Mester,2010;Kumbhakar and Wang,2007;Pastor,1999;Wheelock and Wilson,2009).However,empirical evidence employing the non-parametric DEA approach on the ef ?ciency of Chinese commercial banks is relatively scarce.

Chen et al.(2005)examine the technical ef ?ciency of 43Chinese banks in 1993to 2000employing the DEA method.The results show that large state-owned banks and small banks are more ef ?cient than medium-sized banks.The ?nancial deregulation in 1995was found to have improved technical ef ?ciency.

Yao et al.(2008)evaluate the technical ef ?ciency of Chinese commercial banks using the DEA method to analyze the ef ?ciency levels of these banks over the 1998–2005period,which covers both the pre-WTO era and the ?rst few years of the post-WTO era in China.The DEA results indicate that the Big Four state-owned banks are not necessarily less ef ?cient than their joint-equity counterparts.Two state-owned banks,CCB and BOC,continuously outperform their state-owned peers and most joint-equity banks.

Ariff and Can (2008)employ the DEA technique to investigate the ef ?ciency of 28Chinese commercial banks during 1995–2004.In the second-stage regression,they examine the effects of ownership,size,risk pro ?le,pro ?tability,and key environmental changes on the bank ef ?ciency using the Tobit regression model.The results imply that the joint-stock banks exhibit higher cost and pro ?t ef ?ciency relative to the state-owned counterparts,and the medium-sized banks are more ef ?cient than their small and large peers.

Su ?an (2010)explores the impact of risks on the technical and scale ef ?ciency estimates of Chinese commercial banks using the DEA model.The empirical ?ndings suggest that scale inef ?ciency has a greater in ?uence than pure technical inef ?ciency in determining the total technical ef ?ciency of the Chinese banking sector.

Su ?an and Habibullah (2011)provide empirical evidence on the impact of economic globalization on bank ef ?ciency in https://www.360docs.net/doc/cd10878725.html,ing the DEA method,they compute the technical ef ?ciency of the Chinese banking sector during 2000–2007.In the second stage,their regression results suggest that greater economic integration,cultural proximity,and political globalization have a signi ?cant positive in ?uence on bank ef ?ciency levels.

Luo et al.(2011)evaluate the effectiveness of stock listing on the ef ?ciency of Chinese commercial banks using the DEA approach.Employing data on the 14listed banks in 1999to 2008,they adopt the DEA-CCR-CRS model and the DEA-BCC-VRS model for estimation.Their empirical results con ?rm that banking reform in China over the past 10years has achieved remarkable progress,and ownership restructuring via transforming the banks into shareholding companies is an effective way to enhance their performance.

X.Hou et al./Emerging Markets Review 20(2014)75–88

78X.Hou et al./Emerging Markets Review20(2014)75–88

Ye et al.(2012)use the DEA method with the panel data of the14largest nationwide banks in China to test?ve hypotheses that have been proposed in the literature on the relationship between market structure, pro?tability,and ef?ciency.The empirical results suggest that neither the structure-conduct performance hypothesis nor the ef?cient-structure hypothesis holds in China.

Studies investigating the ef?ciency of China's banking sector mainly concentrate on the impact of ownership form,bank size,?nancial sector reforms,and macroeconomic environment.Empirical evidence on the effects of market structure and risk taking on the technical ef?ciency of Chinese commercial banks, especially after the banking industry opened its doors to the world,remains elusive.Furthermore,outliers are often treated as particularly troublesome to the DEA model(Bogetoft and Otto,2011);thus,they should be removed from the data when DEA models are employed,which is usually ignored in the present literature.

In addition,the conventional two-stage DEA approach employed in the present literature to examine the impacts of environmental variables on the technical ef?ciency estimates of Chinese commercial banks is invalid due to complicated,unknown serial correlations among the estimated ef?ciencies.Given these knowledge gaps,this paper seeks to estimate the technical ef?ciency of Chinese commercial banks,and investigate the effects of market structure and risk taking on the ef?ciency estimates employing a two-stage,semi-parametric DEA model after the outliers are removed from the sample observations.

3.Methodology and data

Data envelopment analysis(DEA)is described as the mathematical programming approach to the construction of frontiers and the measurement of ef?ciency relative to the constructed frontiers.The?rst DEA model is known as the CCR model,named after Charnes et al.(1978).Technical ef?ciency re?ects the ability of a producer to achieve the possible maximum level of output given an input level.Vast studies on the DEA model use quantity data to estimate the technical ef?ciency of the producers.

3.1.Detecting outliers

Outliers are producers that differ to a large extent from the rest of the producers,and therefore may end up being badly captured by the model,or having too large an impact on the model.They are often treated as particularly troublesome to the DEA model because an outlier helps span the frontier and may have a signi?cant impact on the evaluation of several other producers(Bogetoft and Otto, 2011).

Bogetoft and Otto(2011)describe the data cloud method to identify outliers for the DEA model.Let X=(x1,…,x K)and Y=(y1,…,y K)be K×m and K×n matrices with inputs and outputs for K producers. The combined matrix[X Y]contains all the observations.These observations,the different rows in the combined matrix,can be seen as a cloud of points,where each point represents a producer.The volume of the cloud is proportional to the determinant of the combined matrix[X Y]′[X Y]:

eT:e1TVolume of data cloud≈D X;Y

This determinant can be interpreted as the generalized sum of the quadratic residuals from the linear model of Y conditioned on X.If we remove a producer from the data,then the volume of the data cloud may decrease.If the removed producer is in the middle of the cloud,then the volume will be unchanged. Otherwise,the volume will be much smaller,giving an indication that the producer is an outlier.To identify one or more outliers,we can therefore look at the way the volume of the cloud changes when we remove one or more observations.Let D(i)be the determinant after removing producer i,and consider the ratio of the new volume of the data cloud to the old volume:

R ieT?D ieT=D:e2T

If producer i is not an outlier,then R(i)will be close to1.Otherwise,it will be much smaller than1.To identify outliers,we must simply look for small values of R(i).In practical applications,we can use a

graphical model in which we plot the ordered pairs [r ,log(R (r )/R (r )min )],where r is the number of deleted producers.In the graph,we look for isolated points;the r with isolated points gives an indication of r outliers.Finally,Barnett and Lewis (1995)suggest that the reasonable upper bound of the proportion for outliers is (n )1/2/n .

3.2.Two-stage,semi-parametric DEA model

Numerous papers have regressed the DEA estimates of ef ?ciency on environmental variables in two-stage procedures to account for exogenous factors that might affect the performance of producers.Simar and Wilson (2007)argue that a serious problem in all of the two-stage DEA studies arises from the fact that DEA ef ?ciency estimates are serially correlated.Consequently,standard approaches to inference in the second stage are invalid.They propose single and double bootstrap procedures;both procedures permit valid inference,and the double bootstrap procedure improves statistical ef ?ciency in the second-stage regression.

Furthermore,in the literature,only two statistical models have been proposed in which second-stage regressions are well-de ?ned and meaningful.Truncated regression provides a consistent estimation in the second stage in the model proposed by Simar and Wilson (2007),whereas in the model considered by Banker and Natarajan (2008),OLS provides a consistent estimation.Simar and Wilson (2011)examine,compare,and contrast the very different assumptions underlying these two models,and clarify that second-stage OLS estimation is consistent only under very peculiar and unusual assumptions on the data-generating process that limit its applicability.In either case,bootstrap methods provide the only feasible means for inference in the second stage.The implementation steps of double bootstrap procedure of a two-stage,semi-parametric DEA model are described in Appendix B .

The Monte Carlo experiment reported in the work of Simar and Wilson (2007)suggests that the double bootstrap described above performs very well in terms of both coverage for estimated con ?dence intervals and root mean square error.In addition,FEAR 1.13in R is the main software package employed to perform the ef ?ciency analysis in our paper (Wilson,2008).3.3.Data description

Although the size of foreign banks in China's ?nancial system is minuscule (Allen et al.,2013),the implementation of the Regulations of the People's Republic of China on Administration of Foreign-funded Banks in 2006has further increased the pressure on China's banking sector;on the other hand,the restructuring of Chinese large commercial banks had been completed at the end of 2010.Accordingly,the technical ef ?ciency of Chinese commercial banks changed remarkably through this period,which provides an appropriate opportunity to study the impacts of market structure and risk taking changes on the bank ef ?ciency.4We collect data on 44major Chinese commercial banks,including those on the Big Four state-owned banks.Our sample covers over 90%of the banking assets in China.The data set is primarily drawn from the BankScope database,the Financial Yearbook of China,2007–2011,and the statements of various banks in annual ?nancial reports,2007–2011.In this sample period,we also investigate the in ?uence of global ?nancial crisis since the end of 2007on the technical ef ?ciency of Chinese commercial banks.

The relatively short sample period implies that no signi ?cant technical progress or regress has occurred during this period,which indicates an unvarying best-practice technology implying that ef ?ciency estimates for each producer are generated all against the same standard;and trends in technical ef ?ciency estimates of individual producers are relevant.All data of value variables are in ?ation-adjusted to eliminate the impact of price ?uctuations.

In determining the input and output variables,we adopt the intermediation approach,which assumes that banks function as an intermediary between savers and borrowers,to treat bank deposits as an input.

There is no any signi ?cant change that might affect the main results of our study since little structural shock in ?uences China's banking sector dramatically after 2011.

X.Hou et al./Emerging Markets Review 20(2014)75–88

80X.Hou et al./Emerging Markets Review20(2014)75–88

The output variables are total net loan,5TNLOAN,and other earning assets,OEA.The de?ned input variables are total deposits,TDEPOSITS;?xed assets,FASSETS;and number of employees,EMPLOYEES(Berger and Humphrey,1997;Berger et al.,1993;Luo et al.,2011;Su?an,2010;Su?an and Habibullah,2011).

Following the above research as well as the work of Berger et al.(2010),we de?ne the market structure variable of China's banking sector as the Her?ndahl–Hirschman Index in deposits,DHHI.Risk taking variables are de?ned as the ratio of equity to total assets,CAPRISK,which is an indicator of the capital risk of commercial banks;the ratio of loan loss provisions to total loans,an indicator of the credit risk, CREDRISK;and the ratio between total loans and deposits,an indicator of the liquidity risk of banks, LTODEP.Additionally,according to Berger and Mester(1997),we introduce LSIZE,OWNERSHIP, DUMMY2008,TREND,GDPGROWTH,and ROA as control variables to control for bank characteristics, organizational form,and macro-economy characteristics.LSIZE is the natural logarithm of the total assets. The state ownership dummy variable,OWNERSHIP,is0if the bank is stated-owned and1otherwise. DUMMY2008is another dummy variable,which is0before2008and1otherwise.The variable TREND captures the in?uence of unobserved time-varying environmental factors on the technical ef?ciency of Chinese commercial banks.GDPGROWTH and ROA are GDP growth rate and return on assets,respectively. Details of the description of the variables used in this paper are presented in Table A1of Appendix A.Value terms are expressed in million CNY;the number of employees is expressed in person;and the ratio variables are expressed as a percentage.

The detailed descriptive statistics of the input and output variables of Chinese commercial banks are shown in Table1.The bank data are suf?ciently representative of the entire population.The difference between extreme values re?ects the comparison results across various banks in different years.The difference between extreme values in individual years is not as large as it appears.

We test the multi-collinearity between the explanatory and main control variables to guard against the impact of highly multi-collinear regressors on the parameter estimates.The condition number obtained from the test is1.92,and the mean VIF is1.21,which implies that high multi-collinearity has not been found.

4.Empirical results

4.1.Finding the outliers

According to Bogetoft and Otto(2011),Table2presents the r removed observations corresponding to a minimum value of R(r).The rows in Table2show which deletions give the minimum value of R(r);this minimum value is shown.The?rst row,r=1,shows that deleting observation33from the data set results in a value of R(1)at0.36,and this is the minimum R when only one observation is deleted from the data set.The same principle applies to the other rows.

A log-ratio graph can present a clear view of the minimum R s and how they depend on the

(r))).To number of simultaneously deleted observations by plotting the ordered pairs(r,log(R(r)/R min

?nd the outliers,we search the points in this graph where a gap exists between the points above0 and the point at0and eventually?nd that there are three outliers.6From Table2,the outliers are observations16,17,and33.

4.2.Empirical analysis based on a two-stage,semi-parametric DEA model

After removing the identi?ed outliers from the input and output data set,we compute the bias-corrected technical ef?ciency estimates,implementing the double bootstrap procedure of a two-stage semi-parametric

5The total net loan equals the total loans minus the loan loss provisions,which is the output variable of commercial banks concerned in the quality of loans.

6See implementation details of the log-ratio graph in Bogetoft and Otto(2011).For saving space,the plotted log-ratio graph is not displayed here,which is available upon request.

DEA model from steps [1]to [4]under the assumption of various returns to scale.The results for the average relative bias-corrected technical ef ?ciency estimates in the sample period are shown in Table 3.7

The total sample mean relative ef ?ciency of Chinese commercial banks is approximately 0.886,suggesting a signi ?cant potential for further ef ?ciency improvement.As shown in Fig.1,the annual mean technical ef ?ciency of Chinese commercial banks has shown a declining trend during the period under study.This downward trend is brie ?y reversed in 2008.However,the average technical ef ?ciency of Chinese commercial banks subsequently continues to decline.The decline in the annual mean technical ef ?ciency may be attributed to the global ?nancial crisis,which resulted in less demands by ?rms and households for loans and other ?nancial services.

In addition,we use parametric and non-parametric tests to determine whether the means of the estimates of the bias-corrected technical ef ?ciency and the conventional technical ef ?ciency are systematically different.8The null hypothesis of a two-sample mean-comparison t -test with unknown variance in a normal population is that no difference exists between the means of two groups.The P value for this test is 0,denoting that the null hypothesis is rejected.The P values of the Wilcoxon signed-rank test and the sign test are 0and 0.001,respectively.These results,derived using the non-parametric approach,also indicate that a systematic difference exists between the bias-corrected technical ef ?ciency scores and the conventional technical ef ?ciency scores at the 1%signi ?cance level.

After obtaining the estimates of bias-corrected technical ef ?ciency,we construct the truncated regression empirical Eq.(3)to implement steps [5]and [6]of the double bootstrap procedure of a two-stage,semi-parametric DEA model.

b ^δi ?β0tβ1DHHI tβ2CAPRISK tβ3CREDRISK tβ4LTODEP tx 0i βtεi

e3T

Table 1

Descriptive statistics of the input and output variables.Variable Mean S.D.Minimum Maximum TNLOAN 220196.3517588.8858.62852443.0OEA

219578.0522489.0617.52803695.0TDEPOSITS 414750.9967247.61971.95292685.0FASSETS 4341.611162.215.753216.2EMPLOYEES

55367.4

117808.8

510.0

452464.0

A detailed description of the de ?nition of the variables is given in Table A1;S.D.represents standard deviation.Table 2

The r removed observations corresponding to a minimum value of R (r ).r Deleted observation R min

(r )1330.3600217330.07403161733

0.02104141617330.0120515141617330.00606615141617330.00357131211151617330.0019813121115141617330.00079613121115141617330.00041061312111815141617330.00021120131219111815141617330.000112

0.0001

The technical ef ?ciency estimated by the double bootstrap procedure of a two-stage semi-parametric DEA model is the Debreu –

Farrell output technical ef ?ciency.The conventional Shephard ef ?ciency estimates reported in Table 3are equal to the reciprocals of the Debreu –Farrell output technical ef ?ciency scores.8

The conventional technical ef ?ciency estimates computed by the traditional DEA model are not reported in this paper due to space limitations.

X.Hou et al./Emerging Markets Review 20(2014)75–88

b ^δi is the bias-corrected estimate of the Debreu –Farrell technical ef ?ciency;x i is the vector of the control variables de ?ned in this paper;βs are the parameters to be estimated;and εi is the error term.The descriptive statistics of the parameter estimates obtained by steps [5]and [6]of the double bootstrap procedure are shown in Table 4.9

All the control variables de ?ned in this paper are introduced in Model 1.Furthermore,we drop the control variable,OWNERSHIP ,in Model 2,as well as OWNERSHIP and TREND ,in Model 3.10The descriptive statistics of the parameter bootstrap estimates in Table 4suggest that even if we remove the variables denoting the in ?uence of the ownership and the unobserved time-varying environmental factors from the control variable set,the impacts of the explanatory variables on the technical ef ?ciency estimates of Chinese commercial banks remain virtually consistent.

According to Simar and Wilson (2007),the times that the loop runs in the double bootstrap procedure to obtain the bootstrap estimates are set to 100.10

The average bias-corrected ef ?ciency estimates corresponding to the smaller set of the control variables are not reported here due to space limitations and the research purposes of this paper.The average estimates of the various groups of bias-corrected ef ?ciency estimates,corresponding to the different sizes of the control variable sets,have no systematical difference.

Table 3

Average bias-corrected relative technical ef ?ciency estimates of Chinese commercial banks.Bank

Tech ef ?ciency Bank

Tech ef ?ciency ICBC

0.994Bank of Guangzhou 0.887China Construction Bank 0.989Bank of Chengdu 0.823Bank of China Limited 1Harbin Bank 0.690Agricultural Bank of China 0.986Hankou Bank

0.690Bank of Communications 0.940Bank of Chongqing 0.724China Merchants Bank

0.957Bank of Jinzhou 0.642Shanghai Pudong Development Bank 1Bank of Nanchang

0.642China CITIC Bank 1Guiyang Commercial Bank 0.679Industrial Bank

1Bank of Qingdao 0.722China Minsheng Banking Co.1Fujian Haixia Bank 0.765China Everbright Bank 0.975Bank of Dalian

0.783Hua Xia Bank

0.922Weihai City Commercial Bank 0.889China Guangfa Bank

0.831Bank of Wenzhou 0.927Shenzhen Development Bank 1Bank of Jiujiang 0.667Evergrowing Bank 0.827Bank of Weifang 0.873China Zheshang Bank 0.904Bank of Rizhao 0.933Bank of Beijing 1Bank of Shaoxing

0.985Bank of Shanghai 0.832Zhejiang Tailong Commercial Bank 0.933Bank of Ningbo 0.806Yantai Bank

0.881Bank of Nanjing 0.670Dongying City Commercial Bank 1Bank of Hangzhou 0.951Bank of Guilin 0.850Huishang Bank

0.894

Laishang Bank

0.8200.8400.860

0.8800.9000.9202006

2007

20082009

2010

Year

A v e r a g e E f f i c i e n c y

Fig.1.Annual average technical ef ?ciency of Chinese commercial banks.Source:Author computation based on our ef ?ciency estimates.

82X.Hou et al./Emerging Markets Review 20(2014)75–88

After obtaining the set of bootstrap estimates,we construct the estimated 95%asymptotic con ?dence intervals for each element of βand for σεusing the bootstrap values and the original estimates in step [5]of the double bootstrap procedure,and obtain the P values for the test of signi ?cance.11Table 5presents the empirical results of the double bootstrap procedure.

Model 1is of primal importance in this paper.Models 2and 3are viewed as the robustness checks for the signs of the estimated parameters because several important environmental control variables are dropped in them (Fried et al.,2008).To some extent,the P values of the Chi-square tests show that the constructed empirical truncated regression models are signi ?cant as a whole.

Allen et al.(2007)comprehensively examine all aspects of China's ?nancial system.They principally conclude that China's ?nancial system is dominated by a large,inef ?cient banking sector.Despite the entry and growth of many domestic and foreign banks in recent years,China's banking sector is still mainly controlled by large state-owned banks.Furthermore,continuing efforts to control the risks of the major banks within normal levels,thereby avoiding a banking crisis,are the most important aspects of reforming China's banking system.Against this background,we investigate the impact of market structure and risk taking on the ef ?ciency of Chinese commercial banks.

In Model 1,the DHHI has a signi ?cant impact on the Debreu –Farrell ef ?ciency,which is the reciprocal of the conventional Shephard ef ?ciency,indicating that ceteris paribus,the decrease in China's banking market concentration in ?uences bank behavior,which in turn,improves the Shephard technical ef ?ciency of Chinese commercial banks.

The intense market competition reduces the market power of the banks,which compels them to develop advanced technical experience and skills so that they might raise their outputs,ultimately resulting in a higher technical ef ?ciency.This ?nding supports the Structure-Conduct-Performance (SCP)hypothesis of the banking sector,which is inconsistent with Su ?an and Habibullah (2011).The signi ?cant positive estimates of the DHHI parameters in Models 2and 3also con ?rm the stimulative effect of banking competition on technical ef ?ciency.

The empirical results suggest that CAPRISK has no signi ?cant in ?uences on the technical ef ?ciency of Chinese commercial banks in any empirical equations from Models 1to 3.However,CREDRISK is associated with the Debreu –Farrell ef ?ciency signi ?cantly.In any case,the in ?uence directions of the capital and credit risk indicators are consistent with those of Su ?an (2009)'s and Kosmidou (2008)'s.

On the one hand,the ?ndings reveal that the capital risk underlying guarantees by the Chinese government,12providing additional strength to withstand ?nancial crisis and increasing the safety for

The normal-based con ?dence intervals rather than percentile-t intervals are constructed based on the suggestions of Simar and Wilson (2007)and Cameron and Trivedi (2009).12

In the process that the Chinese banking sector has undergone waves of restructuring,the Chinese government has injected fresh capital into the big state-owned commercial banks for non-performing loan clean-up funding from the ?scal revenue and foreign exchange reserve in the name of the “Banking Restructuring Plan ”(Allen et al.,2013).

Table 4

Descriptive statistics of the parameter bootstrap estimates.Variable

Model 1Model 2Model 3Mean

S.D.Mean S.D.Mean S.D.

DHHI 2.208

1.663

2.179 1.609 2.091 1.603CAPRISK 0.5240.7220.4750.6880.4990.667CREDRISK ?4.925

3.321?

4.876 3.235?

5.197 2.940LTODEP ?1.6770.159?1.5940.156?1.5830.151LSIZE

?0.0350.008?0.0340.005?0.0340.005OWNERSHIP 0.0240.035––––DUMMY20080.2360.2290.2380.2220.2730.182TREND

0.0090.0260.0090.025––GDPGROWTH 0.0430.0420.0440.0400.0480.036ROA

?0.1020.040?0.1010.039?0.1010.039Constant 1.7810.751 1.7230.716 1.6670.693^σ?ε

0.124

0.009

0.121

0.009

0.121

0.009

A detailed description of the de ?nition of the variables is given in Table A1;S.D.represents standard deviation.

X.Hou et al./Emerging Markets Review 20(2014)75–88

depositors when the macro economy is unstable,greatly ease the capital constraints on banks'credit expansion,and offset the positive in ?uence of the capital risk indicator on the banks'technical ef ?ciency.Thus,the parameter estimate of CAPRISK is not signi ?cant enough.On the other hand,CREDRISK has a signi ?cant positive impact on the Shephard technical ef ?ciency of Chinese commercial banks since the Chinese government actually provides the underlying deposit insurance.However,it is worth to mention that in this case the bank outputs and therefore the technical ef ?ciency are improved at the cost of taking more credit risk,although this risk is hidden by the government temporarily.

Table 5shows that LTODEP and LSIZE are signi ?cantly associated with the reduced Debreu –Farrell ef ?ciency,indicating that they have signi ?cant positive impacts on the conventional Shephard technical ef ?ciency of Chinese commercial banks.The positive effect of bank size on technical ef ?ciency is also supported by Kosmidou (2008)and Su ?an and Habibullah (2011).Hauner (2005)argues that the positive impact of bank size on technical ef ?ciency may be due to the presence of the IRS over a higher volume of services,or the ef ?ciency gains from a specialized workforce.

From our viewpoint,according to Berger et al.(2010),the bank with a higher loan-deposit ratio is more active in expanding its business,and meanwhile has to take more liquidity risk.In China's current ?nancial supervision system,both relatively large bank size and aggressive business expansion represent the stronger relationship with the government at various levels,as well as the more government support and higher bargaining power of banks against the state monetary authority to obtain more ?exible credit lines,which ultimately allows the banks to produce more outputs,therefore improving their technical ef ?ciency.Finally,the signs of the control variable estimates are also substantially consistent with the relevant empirical results in the literature.

Furthermore,the marginal effects of changes in the main regressors at the sample mean are also evaluated using the calculus method,as shown in Table 6.Again,Model 1is of primal importance in this paper.

Table 6shows that a 1%decrease in DHHI is signi ?cantly associated with a 0.317%increase in the conventional Shephard technical ef ?ciency of Chinese commercial banks;a 1%increase in CAPRISK or in CREDRISK is associated with a 0.005%or 0.032%increase in technical ef ?ciency,respectively,but the former correlation is not signi ?cant;a 1%increase in LTODEP is signi ?cantly associated with a 0.673%increase in technical ef ?ciency;and a 1%increase in LSIZE is signi ?cantly associated with a 0.031%increase in the technical ef ?ciency of Chinese commercial banks.The results of Models 2and 3are virtually similar to the empirical evidence of Model 1.

To sum up,the intense market competition compels banks to develop advanced technical and managerial skills,ultimately resulting in a higher technical ef ?ciency.In addition,the technical ef ?ciency is positive associated with the risk taking.Considering that more risk taking implies a credit expansion of

Table 5

Determinants of the technical ef ?ciency of Chinese commercial banks.Variable

Model 1Model 2Model 3Parameter

Conf.interval Parameter Conf.interval Parameter Conf.interval DHHI 3.219??(0.205,6.233) 3.107??(0.197,6.018) 3.009??(0.168,5.850)CAPRISK ?0.096(?1.516,1.324)?0.103(?1.466,1.260)?0.075(?1.435,1.286)CREDRISK ?5.888?(?12.778,1.001)?5.532?(?12.105,1.042)?5.843?(?11.837,0.150)LTODEP ?1.437???(?1.809,?1.065)?1.364???(?1.738,?0.991)?1.355???(?1.726,?0.983)LSIZE

?0.038???(?0.054,?0.022)?0.035???(?0.045,?0.024)?0.035???(?0.045,?0.024)OWNERSHIP 0.007(?0.070,0.083)–

–

DUMMY20080.427??(0.030,0.823)0.414??(0.039,0.788)0.450???(0.139,0.760)TREND

0.009(?0.041,0.060)0.009(?0.040,0.058)–

–

GDPGROWTH 0.083??(0.012,0.154)

0.081??(0.013,0.148)

0.086???(0.024,0.147)

ROA

?0.123???(?0.210,?0.037)?0.118???(?0.201,?0.034)?0.119???(?0.202,?0.035)Constant 1.051(?0.291,2.393)0.996(?0.319,2.311)0.942(?0.347,2.232)^σ?ε

0.117???(0.101,0.133)0.114???(0.098,0.130)0.113???(0.097,0.129)P value of χ2test

0.000

–

0.000

–

0.000

–

*,**and ***denote signi ?cance at the 10%,5%,and 1%levels,respectively.A detailed description of the de ?nition of the variables is given in Table A1.

84X.Hou et al./Emerging Markets Review 20(2014)75–88

Chinese commercial banks based on the soft risk constraint derived from underlying guarantees by the government,the improvement of technical ef ?ciency may accompany an accumulation of banking risks in the current ?nancial system of China.4.3.Additional robustness tests

We run an additional set of robustness checks of the impacts of market structure and bank risk taking that are not shown for reasons of brevity.We particularly run additional robustness tests of our main ?ndings in which we specify alternative market structure variables and alter the data sample.

First,we de ?ne the market structure variable of China's banking sector as the Her ?ndahl Hirschman Index in loans and CR5,13respectively,and rerun the ef ?ciency analysis using a two-stage semi-parametric DEA model.We still ?nd that our main results hold despite a slight difference in the marginal effects of changes in the main regressors at the sample mean.Moreover,we re-estimate our model exclusive of the local commercial banks in our data sample.These banks are small relative to the other banks that may be characterized by different production technologies.The results are consistent with the reported results for the combined sample.5.Conclusion

The ?nancial system is often considered as the economy's blood circulatory system.Up to now,commercial banks are still the most important ?nancial sectors in China's ?nancial system,although the capital market has achieved great development.Therefore,the health of commercial banks is crucial to China's economy.

This paper aims to ?ll the gap in the literature,estimate the technical ef ?ciency of Chinese commercial banks,and investigate the effects of market structure and risk taking on the ef ?ciency estimates employing a two-stage semi-parametric DEA model after the outliers are removed from the sample observations.The empirical results suggest that the average technical ef ?ciency of Chinese commercial banks in the sample period is 0.886;the annual mean technical ef ?ciency has shown a declining trend.

Our most important ?ndings concern the impacts of market structure and bank risk taking.We ?nd that the intense market competition compels banks to develop advanced technical and managerial skills,ultimately resulting in a higher technical ef ?ciency.Besides,the technical ef ?ciency is positive associated with the risk taking.Considering that more risk taking implies a credit expansion of Chinese commercial banks based on the soft risk constraint derived from underlying guarantees by the government,the improvement of technical ef ?ciency may accompany an accumulation of banking risks in the current ?nancial system of China.These results are also robust to our additional robustness checks.

Our work could be extended in future studies by examining the empirical impacts of other sophisticated indicators measuring banking market structure and bank risk taking such as Boone indicator and Z-score on the bank ef ?ciency.Our study also makes an important contribution by broadening the

Table 6

Marginal effects at the sample mean on the conventional Shephard technical ef ?ciency.Variable

Model 1Model 2Model 3Marginal effect

Std.err.Marginal effect Std.err.Marginal effect Std.err.DHHI ?0.317??0.136?0.310??0.133?0.300??0.133CAPRISK 0.0050.0300.0050.0290.0040.029CREDRISK 0.032??0.0130.030??0.0130.032???0.012LTODEP 0.673???0.0760.649???0.0810.646???0.080LSIZE

0.031???

0.006

0.029???

0.003

0.029???

0.003

**and ***denote signi ?cance at the 5%,and 1%levels,respectively.A detailed description of the de ?nition of the variables is given in Table A1.

CR5is the asset concentration ratio of the ?ve largest banks.

X.Hou et al./Emerging Markets Review 20(2014)75–88

scope of the research on the determinants of bank ef ?ciency in an emerging market economy.The main ?ndings could be applied to investigate the effects of market structure and bank risk taking on the bank ef ?ciency in other developing countries whose banking sectors are undergoing marketization transition and subject to the soft risk constraint.

In terms of policy implications,the positive impact of the intense market competition suggests that reducing the market concentration and breaking the monopoly of China's banking sector are likely to improve the technical ef ?ciency of Chinese commercial banks.China needs to accelerate market-oriented reforms in the banking sector,and break the monopoly of large state-owned commercial banks by means of supporting the development of small and medium banks.

On the other hand,the improvement of bank technical ef ?ciency may accompany an accumulation of banking risks in current Chinese ?nancial system.Chinese commercial banks still need to pay close attention to the risks they take to avoid the serious negative impact on their technical ef ?ciency stemming from the banking crisis in the long run.They should commit themselves to increase the technical ef ?ciency by continuously improving their management skills and production experiences.Acknowledgments

Financial support from Humanity and Social Science Youth foundation of the Ministry of Education of the People ’s Republic of China (project no.12YJC790059)is gratefully acknowledged.We greatly appreciate the comments and suggestions given by the anonymous referee,Professor P.Wilson,Professor Cheng Li,and Professor Honggang Xue.All errors remain our responsibility.Appendix A

Appendix B

The steps of double bootstrap procedure of a two-stage,semi-parametric DEA model are as follows:[1]Using the original input x and output y data,compute the Debreu –Farrell output ef ?ciency

estimates ^δi ?^δx i ;y i j ^?

eT?i ?1;…;n ,where ^?is the estimator of the production set.[2]Use the method of maximum likelihood to obtain an estimate ^β

of the vector of parameters βand an estimate ^σεof the standard deviation of the error term σεin the truncated regression of ^δi on

environmental variables z i in ^δi ?z i βtξi ≥1using the m b n observations when ^δi N 1.

[3]Loop over the next four step L 1times to obtain n sets of bootstrap estimates B i ?^δ?ib

n o L 1

b ?1

:[3.1]For each i ,draw εi from the N 0;^σ2ε

distribution with left-truncation at 1?z i ^β .

Table A1

Variables used in a two-stage,semi-parametric DEA model.Variables Mean Description

TNLOAN 220,196.300Total net loans,outputs of Chinese commercial banks

OEA

219,578.000Other earning assets,outputs of Chinese commercial banks TDEPOSITS 414,750.900Total deposits,inputs of Chinese commercial banks FASSETS 4341.600Fixed assets,inputs of Chinese commercial banks

EMPLOYEES 55,367.400

Number of employees,inputs of Chinese commercial banks

DHHI 0.127Her ?ndahl –Hirschman Index in deposits,the explanatory variable CAPRISK 0.060Ratio of equity to total assets,the explanatory variable CREDRISK 0.007Ratio of LLP to total loans,the explanatory variable

LTODEP 0.598Ratio of total loans to deposits,the explanatory variable LSIZE

11.069Natural logarithm of total assets,the control variable OWNERSHIP 0.912State ownership dummy variable,the control variable DUMMY20080.6122008dummy variable,the control variable TREND

2.022Time trend,the control variable

GDPGROWTH 11.174GDP growth rate,the control variable

ROA

0.990

Return on assets of the banks,the control variable

86X.Hou et al./Emerging Markets Review 20(2014)75–88

[3.2]Again for each i ,compute δ?i ?z i ^βtεi .

[3.3]Set x ?i ?x i ;y ?i ?y i ^δi =δ?i for all i .

[3.4]Compute ^δ?

i ?δx i ;y i j ^??eT?i ?1;…;n ,where ^??is obtained by replacing (x i ,y i )by (x i ?,y i ?).

[4]For each i ,compute the bias-corrected estimator b

^δi using the bootstrap estimates in B i and the

original estimate ^δi .

[5]Estimate by maximum likelihood the truncated regression of b ^δi on z i ,yielding b ^β;b ^σ

.[6]Loop over the next three step L 2times to obtain a set of bootstrap estimates ??^β?;^σ?ε

b n o L 2

b ?1

:[6.1]For each i ,draw εi from the N 0;b ^σ

distribution with left-truncation at 1?z i b ^β .[6.2]Again for each i ,compute δ??i ?z i b ^βtεi .

[6.3]Use the maximum likelihood method to estimate the truncated regression of δi

??on z i ,yielding b ^β?;b ^σ?

.[7]Use the bootstrap values in ?and the original estimates b ^β

;b ^σto construct estimated con ?dence intervals for each element of βand σε.References

Allen,F.,Qian,J.,Qian,M.,2007.China's ?nancial system:past,present,and future.In:Rawski,T.,Brandt,L.(Eds.),China's Great

Economic Transformation.Cambridge University Press,Cambridge,pp.506–568.

Allen,F.,Qian,J.,Zhang,C.,Zhao,M.,2013.China's ?nancial system:opportunities and challenges.In:Fan,P.H.J.,Morck,R.(Eds.),

Capitalizing China.University of Chicago Press,Chicago,pp.63–143.

Ariff,M.,Can,L.,2008.Cost and pro ?t ef ?ciency of Chinese banks:a non-parametric analysis.China Econ.Rev.19,260–273.

Banker,R.,Natarajan,R.,2008.Evaluating contextual variables affecting productivity using data envelopment analysis.Oper.Res.56,

48–58.

Barnett,V.,Lewis,T.,1995.Outliers in Statistical Data.Wiley,Chichester.

Berger,A.N.,Humphrey,D.B.,1997.Ef ?ciency of ?nancial institutions:international survey and directions for future research.Eur.J.

Oper.Res.98,175–212.

Berger,A.N.,Mester,L.J.,1997.Inside the black box:what explains differences in the ef ?ciencies of ?nancial institutions?J.Bank.

Financ.21,895–947.

Berger,A.N.,Hunter,W.C.,Tinume,S.G.,1993.The ef ?ciency of ?nancial institutions:a review and preview of research past,present

and future.J.Bank.Financ.17,221–249.

Berger,A.N.,Hasan,I.,Zhou,M.M.,2009.Bank ownership and ef ?ciency in China:what will happen in the world's largest nation?J.

Bank.Financ.33,113–130.

Berger,A.N.,Molyneux,P.,Wilson,J.O.S.,2010.The Oxford Handbook of Banking.Oxford University Press Inc,New York.Bogetoft,P.,Otto,L.,2011.Benchmarking with DEA,SFA,and R.Springer Science +Business Media,LLC,New York.Cameron,A.C.,Trivedi,P.K.,2009.Microeconometrics Using Stata.Stata Press,College Station,Texas.

Charnes,A.,Cooper,W.W.,Rhodes,E.,1978.Measuring the ef ?ciency of decision-making units.Eur.J.Oper.Res.2,429–444.

Chen,X.,Skully,M.,Brown,K.,2005.Banking ef ?ciency in China:application of DEA to pre-and post-deregulation eras:1993–2000.

China Econ.Rev.16,229–245(16).

Drake,L.,Hall,M.J.B.,2003.Ef ?ciency in Japanese banking:an empirical analysis.J.Bank.Financ.27,891–917.

Fried,H.O.,Lovell,C.A.K.,Schmidt,S.S.,2008.The Measurement of Productive Ef ?ciency and Productivity Growth.Oxford University

Press Inc,New York.

Fungacova,Z.,Pessarossi,P.,Weill,L.,2013.Is bank competition detrimental to ef ?ciency?Evidence from China.China Econ.Rev.27,

121–134.

Hauner,D.,2005.Explaining ef ?ciency differences among large German and Austrian banks.Appl.Econ.37,969–980.

Hughes,J.P.,Mester,L.J.,2010.Ef ?ciency in banking:theory,practice,and evidence.In:Berger,A.N.,Molyneux,P.,Wilson,J.O.S.(Eds.),The

Oxford Handbook of Banking.Oxford University Press Inc,New York,pp.463–485.

Kosmidou,K.,2008.The determinants of banks'pro ?ts in Greece during the period of EU ?nancial integration.Manag.Financ.34,

146–159.

Kumbhakar,S.C.,Wang,D.,2007.Economic reforms,ef ?ciency,and productivity in Chinese banking.J.Regul.Econ.32,105–129.Luo,D.,Yao,S.,Chen,J.,Wang,J.,2011.World ?nancial crisis and ef ?ciency of Chinese commercial banks.World Econ.34,667–852.Naceur,S.,Omran,M.,2011.The effects of bank regulations,competition,and ?nancial reforms on bank's performance.Emerg.Mark.

Rev.12,1–20.

Pastor,J.M.,1999.Ef ?ciency and risk management in Spanish banking:a method to decompose risks.Appl.Financ.Econ.9,371–384.Pessarossi,P.,Weill,L.,2013.Do capital requirements affect bank ef ?ciency?Evidence from China.BOFIT Discussion Papers,p.28

(http://www.suomenpankki.?/bo ?t_en/tutkimus/tutkimusjulk aisut/dp/Documents/2013/dp2813.pdf).

Simar,L.,Wilson,P.W.,2007.Estimation and inference in two-stage,semi-parametric models of production processes.J.Econom.

136,31–64.

Simar,L.,Wilson,P.W.,2011.Two-stage DEA:caveat emptor.J.Prod.Anal.36,205–218.

Su ?an,F.,2009.Determinants of bank ef ?ciency during unstable macroeconomic environment:empirical evidence from Malaysia.

Res.Int.Bus.Financ.23,54–77.

X.Hou et al./Emerging Markets Review 20(2014)75–88

88X.Hou et al./Emerging Markets Review20(2014)75–88

Su?an,F.,2010.The impact of risk on banks'technical and scale ef?ciency:empirical evidence from the Chinese banking sector.IUP J.Financ.Econ.8,82–102.

Su?an,F.,Habibullah,M.S.,2011.Navigating the impact of globalization on bank ef?ciency in China.China World Econ.19,85–101. Wheelock, D.,Wilson,P.W.,2009.Robust nonparametric quantile estimation of ef?ciency and productivity change in U.S.

commercial banking,1985–2004.J.Bus.Econ.Stat.27,354–368.

Wilson,P.W.,2008.FEAR1.0:a software package for frontier ef?ciency analysis with R.Socio Econ.Plan.Sci.42,247–254.

Yao,S.,Han,Z.,Feng,G.,2008.Ownership reform,foreign competition and ef?ciency of Chinese commercial banks:a non-parametric approach.World Econ.31,1310–1321.

Ye,Q.,Xu,Z.,Fang,D.,2012.Market structure,performance,and ef?ciency of the Chinese banking sector.Econ.Chang.Restruct.45, 337–358.