Market structure, efficiency and performance of banking industry in Sri Lanka

Transcription

1 Market structure, effcency and performance of bankng ndustry n Sr Lanka AUTHORS ARTICLE INFO JOURNAL FOUNDER Lalth Seelanatha Lalth Seelanatha (2010). Market structure, effcency and performance of bankng ndustry n Sr Lanka. Banks and Bank Systems, 5(1) "Banks and Bank Systems" LLC Consultng Publshng Company Busness Perspectves NUMBER OF REFERENCES 0 NUMBER OF FIGURES 0 NUMBER OF TABLES 0 The author(s) Ths publcaton s an open access artcle. busnessperspectves.org

2 Banks and Bank Systems, Volume 5, Issue 1, 2010 Lalth Seelanatha (Australa) Market structure, effcency and performance of bankng ndustry n Sr Lanka Abstract Durng the perod of , reforms n the fnancal-servces sector, development n nformaton and communcaton technologes (ICT) and globalzaton of the ndustry have drastcally changed the market structure of bankng ndustry n Sr Lanka. Fnancal reforms commenced n late 1970s were the man drvng force of those changes. The reforms amed to enhance both the productvty and effcency and the degree of competton of bankng market as a way of mprovng overall operatonal performance of the fnancal servces sector n Sr Lanka. Ths paper revewed how the banks effcency and market structure affect the overall performance of the bankng frms measured n terms of proftablty and net nterest margn usng structure conduct performance lterature. The study fndngs suggest that tradtonal structure conduct performance argument s not held n the bankng ndustry n Sr Lanka and the banks performance does not depend on ether market concentraton or market power of ndvdual frms but on the level of effcency of the bankng unts. Keywords: effcency, market structure, performance, bankng. JEL Classfcaton: G20, G21, N20. Introducton Durng the last three decades, bankng ndustry n Sr Lanka has experenced a transton perod as a consequence of deregulaton of fnancal servces sector, development n ICT and globalzaton of the ndustry. Impacts of consequent changes were observable n areas such as the scope of bankng operatons, number of banks and bank branches, technologes used and qualty of human resources n the bankng ndustry as well as nsttutonal structure of the bankng ndustry. Ths paper examnes how the banks market structure and banks effcency have nfluenced the performance of the bankng frms n Sr Lanka usng structure conduct performance lterature. The bankng ndustry whch holds approxmately 60% of the total fnancal assets (World Bank, 2003), s the man ntermedary n the fnancal servces sector n Sr Lanka. Therefore, effcency and productvty of the bankng ndustry are an mportant requrement for the development of the sector. After nearly 30 years of nward-lookng economc polces and fnancal represson, the economc polcy reforms package whch was ntroduced n 1977 paved the way for structural transformaton of the overall economy (Dunham & Kelegama, 1996). The reform package ncluded some drastc polcy changes n relaton to deregulaton of the fnancal servces sector, along wth other economc reforms. Fnancal reforms n Sr Lanka commenced n late 1977 amed to mprove the performance of banks through enhancng compettveness and effcency of the ndustry. Intal reform measures have allowed some structural changes n fnancal servces sector by gvng greater freedom to the prvate sector. The Lalth Seelanatha, government encouraged new entrants to the fnancal servces market. Those changes were affected to expand the scope of the bankng ndustry as well as to ncrease the number of frms n the bankng ndustry. Structural changes n the ndustry amed to enhance competton antcpatng productvty and effcency mprovements n banks. Ultmately, polcy makers amed to mprove the performance of overall bankng ndustry. These trends n the bankng ndustry have led to creatng publc polcy concern wth the degree of concentraton of bankng market. Further, a polcy dlemma s arsen on how frms n the bankng ndustry mght allow to compete effectvely n a more lberal bankng market. Prevous researches have suggested two alternatve polcy drves (Byeongyong, Paul Cho and Mary A. Wess, 2005; Lloyd- Wllams, D. M., Phl Molyneux, and John Thornton 1994; Molyneux, Phl 1999; Moore, Robert R., 1998) for ratonalzng market structure n bankng ndustry. The frst one les n lmtng the number of bankng unts n the market through encouragng mergers among exstng banks. Ths wll help to ncrease the bank sze for perusng scale of economcs. The second strategy s the sharng common facltes such as ATM wth other banks n the ndustry. Both strateges may be useful n enhancng the competton n the market and mprovng the overall productvty and effcency of the market. On the other hand, the deregulaton, the ICT and the globalzaton have changed the way of competton n the bankng ndustry. The mproved level of competton has forced banks to be more effcent. As explaned n the effcent structure hypothess (ESH), there s no need to encourage mergers, snce the effcent enttes can mprove ther market share by provdng more economcal bankng servces n the market. Therefore, ESH suggests that publc

3 polcy makers should focuse on dentfcaton and mplementaton of polces leadng to enhancng productvty and effcency. Instead of encouragng bank mergers, the ESH supports polces whch may encourage sharng common facltes to avod duplcaton of captal cost. However, the frst alternatve polcy s very much algned wth the structure market paradgm whch prases the sgnfcance of the market power n takng operatonal decsons. Accordngly, ths paper emprcally revews what polcy opton s more approprate to the bankng ndustry n Sr Lanka usng the Berger and Hannan s (1993) framework. Ths paper s organzed as follows. The next secton presents a bref revew of lterature related wth market structure and bank effcency gvng specal reference to the bankng ndustry. The second secton detalzed the emprcal framework used. The thrd secton presents results and mplcaton of the analyss. The last secton presents conclusons of the analyss. 1. Lterature revew Prevous studes have used structural and nonstructural approaches to nvestgate behavor of the bankng market. Structural approaches are manly based on the tradtonal ndustral organzaton theory whch focuses on the structure-conductperformance (SCP) paradgm and on the effcent structure paradgm. Lterature based on structural approaches has nvestgated how the market concentraton weakens the market competton by fosterng collusve behavor among frms. Conversely, non-structural approaches assume that factors other than market structure and concentraton may affect compettve behavor, such as entry/ext barrers and the general contestablty of the market (Panzar, J.C. and J. N. Rosse, 1987; Rosse, J.N. and J.C Panzar, 1977). Non-structural approaches have been developed n the context of the new emprcal ndustral organzaton (NEIO) lterature. Ths study s based on structural approaches expectng to uncover the advantage of enhancng banks operatonal effcency aganst bank concentraton. The next secton presents basc arguments of structure conduct performance hypothess (SCH) and effcent structure hypothess (ESH), applcaton n bankng ndustry and ther fndngs. SCP was frst ntroduced by Mason n 1939 as a method of analyzng markets and frms (Worthngton, Ian, Chrs Brton, and Andy Rees, 2001). As explaned n the SCP, the market concentraton fosters colluson among large frms n the ndustry whch subsequently leads to hgher profts. Hence, SCP ponted out that changes n market concentraton may have a postve nfluence on a frm s fnancal performance (Goldberg, Banks and Bank Systems, Volume 5, Issue 1, 2010 Lawrence G. and Anoop Ra, 1996). Furthermore, SCP recognzed the consequent postve relatonshp between market concentraton and performance as a result of ant-compettve behavor of frms wth large market share (Berger, Allan N. and T. Hannan, 1989). The relatve market hypothess (RMPH) whch s a specal case of SCP posted that only frms wth large market shares and well dfferentated product lnes are able to exercse market power to gan superor proft on non-compettve prce settng behavor (Berger, Allan N., 1995). An alternatve theory, the effcent structure hypothess (ESH) states that aggressve behavor of effcent frms n the market leads to ncrease n those frms sze and market share. Ths behavor of the effcent frms allowed such frms to concentrate and earn hgher profts wth further enhancng ther market share. Those frms can maxmze profts ether by mantanng the present level of prce and frms sze or by reducng prce and expandng the frm sze (Lloyd-Wllams, D. M., Phl Molyneux, and John Thornton, 1994). Berger and Hannan (1989) stated that frms n markets wth a large dsperson of effcency wthn market create unequal market share and hgh level of concentraton. Accordngly, the ESH stated that the postve relatonshp between proft and concentraton results from the lower cost acheved through superor management and effcent producton process (Goldberg, Lawrence G. and Anoop Ra, 1996). However, proponents of ESH argued that effcency dfferences among DMUs wthn markets create hgh levels of concentraton. The hgh concentraton rato n the market creates greater than average effcency n these markets yeldng a postve proft concentraton relatonshp (Berger, Allan N. and T. Hannan, 1989). As explaned by Berger and Hannan (1989), ESH and SPC stand on smlar observaton on the relatonshp between concentraton and performance (proftablty). But the dfference n two theores conssted manly n ways of nterpretaton of the relatonshp. Some contemporary studes have challenged the acceptablty of the postve relatonshp predcted between market concentraton and proftablty. Smrlock (1985) posted that there s no relatonshp between concentraton and proftablty but between proftablty and market share. Hs study whch used 2,700 unt-banks n state found no evdence for the relatonshp between concentraton and proftablty. However, he found strong evdence for the relatonshp between market shares (whch are used as proxy for the frm s effcency) and frms proftablty. Smrlock (1985) showed that market concentraton s not a sgnal of collusve behavor but rather the superor effcency of the leadng frms. 21

4 Banks and Bank Systems, Volume 5, Issue 1, 2010 Berger and Hannan (1994) ponted out four sources of ant-compettve behavors may be arsen as a consequence of hgh market concentraton: 1. If a frm s enoyng a large share of market and t s able to set the prces n excess of compettve levels wth a lesser pressure on managers for mantanng operaton costs at or near ther compettve level. 2. Managers self-nterest behavor may lead to makng more rsky fnancng decsons (whch are above the shareholders expectaton) to reduce the varaton n earnngs to protect ther postons. 3. Increase n the poltcal cost assocated wth obtanng and dependng on the exstng market power. 4. The retenton of neffcent managers or the mantenance of neffcent practces allowng managers to lve a quet lfe to pursue other obectves or mantan market power gans. The above explanaton has led to usng the quet lfe hypothess as an alternatve hypothess to the SCP and HSH. Ths hypothess assumes that the managers of frms wth relatvely large market shares gve less attenton to the effcent use of resources snce they can make profts usng ther prce-settng power (Punt, L.W. and M.C.J. Van Roo, 1999). Ths hypothess predcts that large frms n the market use ther market power to be quet n the market and earn proft wthout mprovng productvty and effcency. The behavor of such frms creates economc rent to the market. Early ESH studes had not used drect effcency measures. They had used frms market shares as a proxy to a frm s effcency (Molyneux, Phl and Wllam Forbes, 1995). Berger and Hannan (1995) have frst ncorporated drect method of effcency measures to emprcal models. The man dsadvantage of usng a frm market share s that t does not represent overall productvty and effcency level of frms. Incorporaton of drect measures of effcency captures mpact of all factors affectng the frm s performance. SCP, n general, provdes two man benefts to studes whch nvestgate market behavor. Frst, t shows the way the market s operatng. Thus, t explans dfferent forces whch restrct or expand the scope of a frm s operatons n the market. Especally wth productvty and effcency studes, SCP helps to nterpret dfferent sources of productvty and effcency gans or losses. Second, SCP provdes a ratonal bass for analyzng the market behavor. Even after 50 years of emergng the theoretcal bass of market structure and performance, there was very lmted number of studes, nvestgatng the market structure of developng countres. Prevous emprcal studes manly focused on few developed countres n North Amerca and Europe. Glbert (1984) has summarzed 44 such studes whch were based on US bankng ndustry. The fndngs n those studes have less emprcal valdty n relaton to an emergng developng country. However, those fndngs are of greater mportance to understand the theory behnd the market structure. Therefore, a bref account of prevous applcaton s presented below. One man ssue to be addressed n market structure and performance research s selectng an approprate measure to represent a frm s performance. In prevous studes there has been used prce nformaton (Berger, Allan N. and T. Hannan, 1989) or proftablty nformaton to proxy the frms performance (Molyneux, Phl and Wllam Forbes, 1995). In a mult-product envronment use of sngle measure of prce to represent a frm s overall performance may be napproprate. Proftablty measures can be used as a comprehensve performance ndcator snce t ntegrates both cost and revenue nto one measure. In some studes, ncreased market concentraton was found to be assocated wth hgher prces and greater than normal profts. Smrlock (1985) stated that hgher profts n concentrated markets could be the result of greater productve effcency. Berger (1995) fnds some evdence that the effcency hypothess holds n US bankng. A postve relatonshp between bank concentraton and ROE was found by Short (1979) n a study whch was based on a sample of banks from Canada, Western Europe and Japan. Moore (1998) examned the mpact of advanced communcaton technology on the ablty of banks to serve dstant customers. Advanced technology helped bank managers to serve dstant customers usng alternatve bankng methods, for example, telebankng and nternet bankng. Moore examned the changes n relatonshp between concentraton rato and proftablty usng both unvarate and multvarate regresson tests and found that, even though the technology had changed, the bank concentraton had postvely affected the performance. Molyneux and Forbes (1995) found evdence to support tradtonal SCP from a study n European Bankng. Lloyd-Wllams, Molyneux and Thornton (1994) examned the applcablty of the SCP and effcent market paradgm to analyze the Spansh bankng structure usng a three frms concentraton rato and market share of an ndvdual frm to represent frms effcency. The regresson result ndcated a postve relaton between the concentraton and the return on assets (proxy for performance), thus, supportng the SCP hypothess for the Spansh bankng ndustry. 22

5 Varous studes have employed dfferent methodologes to test SPC theory. Berger and Hannan s (1993) research framework provded a comprehensve methodology for testng potental relatonshps between market structure and performance under both SCP and effcent market hypotheses. They proposed to test four hypotheses: namely, the tradtonal SCP hypothess, relatve market hypothess, X-effcency hypothess and scaleeffcency hypothess. These hypotheses have been used to nvestgate whether market concentraton affects performance or effcency affects market concentraton. Goldberg and Ra (1996) examned the structure-performance relatonshp of banks n European countres usng the Berger and Hannan approach. Ther study dd not fnd a sgnfcant postve relatonshp between concentraton and proftablty. However, there was evdence n favor of the relatve market hypothess for all banks located n hghly concentrated ndustres. Usng a smlar approach Fu and Heffernan (2005) examned market structure of Chnese bankng market. Ther results found evdence for the RMPH. Even though, Fu and Heffernan have found postve sgnfcant coeffcent for effcency varables, they were not able to fnd postve relatonshp between market share and effcency whch were one of the necessary condtons to acheve. Byeongyong and Wess (2005) provded supportve evdence to ESH. They suggest that regulators should focus on measures whch may affect to enhance the frms effcency rather than market power. Further to that, Yu and Neus (2005) found evdence to support a postve scale effcency verson of the ESH and SCH. These results suggest that frms can mprove ther performance by havng an optmum scale of operaton and enhancng the market concentraton. The SCP framework has been wdely used n the lterature to examne market structures. However, t does not account for other factors whch nfluence frms proftablty and concentratons. Further, SCP studes gnore the long-run equlbrum n the market. Therefore, the evdence from market concentraton studes may be nsuffcent to support frm conclusons about the relatonshp between market behavor and competton. 2. Methodology To analyze the nfluence of market structure and effcency on bank performance, ths study uses smlar framework to the emprcal framework proposed by Berger and Hannan (1993) Hypotheses. Berger and Hannan (1993) ntroduce a seres of tests to ncorporate effcency and market structure varables drectly nto the reduced revenue equaton to unvel the nfluence of market structure and effcency on frm Banks and Bank Systems, Volume 5, Issue 1, 2010 performance. They develop four hypotheses relyng on tradtonal market structure paradgm and the effcent structure paradgms. The estmated coeffcents of the reduced form revenue equaton have been used to test the hypotheses. H1: Structure-conduct-performance (SCP). Followng the same theoretcal bass whch used the tradtonal SCP, ths hypothess predcts that the collusve behavor of domnatng frms n the ndustry nfluences the prce settng process n the market whch allowed those frms to gan superor proft over the other frms. Accordngly, SCP predcts there s postve relatonshp between market concentraton and frm performance. Ths hypothess used concentraton rato to proxy collusve market power of domnatng frms. H2: Relatve market power (RMP). Frms wth relatvely bgger market shares and dfferentated product lnes have a superor market power and use t to set market prces and, thereby, earn an above average proft. Therefore, market share and frm s performance mght have a postve relatonshp. The frst two hypotheses test the nfluence of two market structure varables on frms performance. The frst one examnes how the collusve behavor nfluences frm performance and the second one examnes how ndvdual frms are usng ther market power. To accept ether of the hypotheses, the estmated coeffcent should be postve and sgnfcantly dfferent from zero. The next two hypotheses examne the valdty of the effcent structure paradgm. ESH argues both superor performance and hgh market share result n operatonal effcency of ndvdual decsons makng unts n the market. Accordngly, effcency varables are ncorporated as ndependent varables to the revenue equatons. Consequently, these hypotheses predct that the nfluence of market structure on frm performance s nsgnfcant and economcally meanngless. Berger and Hannan proposed other two hypotheses to examne the relatonshp between frms performance and effcency. H3: X-Effcency (XEFF). Techncally effcent frms whch have superor management and/or producton process are able to operate at a lower cost and subsequently gan hgh profts and market share. The hgh techncal effcency allowed respectve frms to get a hgher market share at the expenses of less effcent frms. Therefore, t s expected to have a postve relatonshp wth proftablty and the varables such as techncal effcency, market share and concentraton. H4: Scale-effcent frm (SEFF). Ths hypothess predcts that the dfferences n performance among frms exst not because of dfferences n the 23

6 Banks and Bank Systems, Volume 5, Issue 1, 2010 superorty of management and producton approaches but because of the dfference n level of economcs of scale. It predcts, frms whch are operatng under an optmum scale produce goods and servces at a relatvely lower cost and are able to gan a hgh proft whch leads to a hgh market share Emprcal model. The followng reduced form proft equaton has been employed to determne whch of the stated hypotheses best explaned frms performance. Coeffcents for the unknown varables are to be estmated usng smple lnear regresson approach. p 24 CONC con ms MS EFF TE SEFF SE n 1 where p = measures of performance (proftablty/net nterest margn), = estmated coeffcent for concentraton, market share, techncal effcency and scale effcency, TE = techncal effcency, SE = scale effcency, Z = vector representng the control varables, = estmated coeffcent for control varables, MS = market share of the th bank, CONC = concentraton of the market whch s measured usng HHI, = random error, I = th bank Testng SCP and RMP hypotheses. SCP and RMP hypotheses are used to test the relatonshp between market structure and conduct. These two hypotheses assume market power as the domnant varable n determnng frm proftablty. As explaned before, collusve power of domnatng frms sets the prces on the market and leads to hgh profts. If SCP holds, then the expected coeffcent of Equaton (1) for varables representng concentraton should be greater than zero wth a postve sgn. If RMP holds, the varable representng market share should have a statstcally sgnfcant postve coeffcent. If ether of these hypotheses remans n effect, other control varables ncludng effcency varables may have sgnfcant effect on proftablty Testng ESH hypothess. ESH states that cost advantage enoyed by effcent frms leads them to have a hgher proft than neffcent frms. Effcent frms pass cost advantages to ther customers through adustng prces whch lead to have a hgher market share. Therefore, t s expected to have followng sgns for estmated coeffcent of Equaton (1) f ESH holds. TE>0, SE>0, CON=0 and MS=0. Snce effcent frms are expected to have relatvely low cost advantage leadng to hgher proft, a statstcally sgnfcant postve relatonshp between frm performance and effcency s assumed. One of the necessary condtons to hold the effcent structure hypothess s the postve relatonshp between effcency and market structure. Hence, parameter for the followng functonal forms s estmated. MS CON EFF TE EFF SEFF TE SE SEFF n 1 SE Z n 1 Z. (2) 2. (3) 3 Z, (1) 7 If the above models are able to provde a statstcally postve coeffcent for effcency varables, t could be regarded as that the relatonshp between market structure and effcency s uncondtonally accepted Model varables. Regressons models used n ths study have ncluded ROA, market concentraton (HHI), relatve market share, techncal effcency and scale effcency as man varables under study. We used market concentraton and relatve market share to proxy the market power for examnng the nfluence of market structure on banks performance. DEA estmated techncal effcency and scale effcency scores are used to represent the banks effcency. Other control varables are used to represent rsk (loans to total assets), sze (ln[total assets]), economc growth (GDP growth) and nflaton. A dummy varable has been ncorporated to proxy the mpact of government ownershp on banks performance. Dfferent measures of performance have been used n pror studes n the area of market structure and frm performance. Glbert (1984) revealed that both proftablty ndcators and prce ndcators have been appled n pror studes to proxy bank performance. Followng pror studes (Goldberg, Lawrence G. and Anoop Ra, 1996; Smrlock, Mchael, 1985; Yu, Pey and Werner Neus, 2005), ths study used both proftablty and net nterest margn (NIM) to proxy banks performance. Prevous studes have used rato of profts to assets (.e. the return on assets (ROA)) (Goldberg, Lawrence G. and Anoop Ra, 1996; Yu, Pey and Werner Neus, 2005) and the profts to equty rato (.e. the return on equty (ROE)) (Smrlock, Mchael, 1985; Yu, Pey and Werner Neus, 2005) to represent banks performance. In prncple, ROA reflects the ablty of a bank s management to generate profts from the bank s assets, although t may be based due to off-balance-sheet actvtes. ROE ndcates the return to shareholders on ther equty and equals ROA tmes the total assets-to-equty rato. The latter s often referred to as the bank s equty multpler, whch measures fnancal leverage. Banks wth lower leverage (hgher equty) wll generally report hgher ROA, but lower ROE. Snce an analyss of ROE dsregards the greater rsks assocated wth hgh

7 leverage and fnancal leverage s often determned by regulaton, ROA emerges as the key rato for the evaluaton of bank proftablty (Sundararaan, Vasudevan, Charles Enoch, Armda S. San Jose, Paul Lous Cerel Hlbers, and Russell C. Morett Krueger, Marna Slack, Graham L., 2002). Thus, we use ROA to proxy the banks performance n ths study. We used NIM as an alternatve measure of banks performance whch s drectly lnked to the market condton. NIM can be regarded as a drect measure of performance whch may result n the market power of domnatng frms as well as the resdual of nterest ncome resulted from effcent decson makng of management. Ths study estmates NIM dvdng net nterest ncome (the dfference between total nterest ncome and nterest expenses) by total asset. We measure frms effcency (techncal and scale) usng data envelopment analyss (DEA) 1. Prevous studes whch used DEA to estmate techncal effcency have been lmted to the CCR 2 model formulated by Charnes, Cooper and Rhodes (1978) and the BCC model formulated by Banker, Charnes and Cooper (1984). Effcency scores estmated usng CCR and BCC models have been used for dervng Scale effcency scores. The CCR model assumes constant return to scale and BCC relaxes ths assumpton by beng based on the varables returns to scale. CCR and BCC DEA formulatons are appled to estmate the TE and PTE, respectvely. Prevous studes have employed an MPI lke ndex 3 to decompose scale effect on a DMU s neffcency. A frm s TE s a functon of PTE and SE. Therefore, PTE should be separated from TE to dentfy SE (Coell, Tm, D.S. Prasada Rao, and George E. Battese, 1998). The DEA models power of dscrmnatng neffcent unts from effcent unts depends on the number of unts under observaton and the number of nputs and outputs n the model. Snce ths study s based on a relatvely small sample, three-year movng wndows are used to construct producton fronters for measurng effcency (Charnes, Abraham, Wllem W. Cooper, Are Lewn, and Lawrence M. Seford, 1997). Effcency measures how effectvely banks use ther nputs to produce a gven level of output. Pror studes have used fve approaches of nput and output 1 Few reasons have nfluenced on selecton of DEA as the preferred method for estmatng effcency n banks n Sr Lanka. They are: (1) non requrement of pre-specfed functonal form, (2) ablty to ncorporate a combnaton of nput and output varables, and (3) ts ablty to measure the effcency even usng a small number of observatons. 2 Please, see the appendx for DEA model used n ths paper. 3 Techncal effcency (TE) s the product of pure techncal effcency (PTE) and the scale effcency (SE). Thus, SE can be estmated by dvdng the estmated effcency scores of CCR (whch measures TE) by the estmated effcency scores of BCC (Coell, Rao and Battese, 1998). Banks and Bank Systems, Volume 5, Issue 1, 2010 specfcaton, namely ntermedaton, producton, user-cost, value-added and assets (Avkran, N.K., 2000). However, there s no apparent consensus n the lterature concernng the most approprate approach for defnng banks nput and outputs. Based on ntermedaton approach (Sealey, C.W. and Jemes T. Lndley, 1977), ths study dentfed nterest expenses, personnel costs and premses and establshment expenses as banks nputs and loans and other advances, nterest ncome and other ncome as banks outputs. Snce banks are mult-products frms, fndng a sngle varable to represent the bankng market s a dffcult task. Total depost held (Goldberg, Lawrence G. and Anoop Ra, 1996; Smrlock, Mchael, 1985), total loans granted and total assets held are some of such varables that can be used as proxy for the market capacty. Ths study prefers to use total assets as a proxy for the sze of market n bankng ndustry snce the total assets represented combne outcome of whole bankng actvtes. Prevous studes have used Herfndahl-Hrshman ndex (HHI) 4 (Bkker, Jacob A. and Katharna Haaf, 2002; Byeongyong, Paul Cho and Mary A. Wess, 2005; Goldberg, Lawrence G. and Anoop Ra, 1996; Molyneux, Phl, 1999; Yu, Pey and Werner Neus, 2005) and k banks concentraton rato (CR k ) 5 to proxy the collusve power n a gven market. CR k takes the total market shares of the k th largest banks n the market. Contemporary bankng studes have gven more preference to HHI as proxy of market concentraton snce, t consders the market shares of all frms n the market and k bank concentraton rato takes nto account only domnatng banks market shares. Accordngly, ths study stcks nto HHI as a better proxy for collusve power domnatng n the bankng market. Together wth prevously specfed varables, sze, operatonal rsk, ownershp structure, GDP and nflaton have been ncorporated as other control varables. Followng pror studes (Goldberg, Lawrence G. and Anoop Ra, 1996; Smrlock, Mchael, 1985) bank s sze s ncorporated to represent bank s dversfcaton ablty. If large banks were able to capture sgnfcant cost advantages over small banks, banks sze should be postvely related to the proftablty. To control the rsk takng behavor of the proft seekng banks, loan to total 4 HHI defnes market shares as weghts, and stresses the mportance of larger banks by assgnng them a greater weght than smaller banks. It accounts all banks and weghts them accordng to ther market share, N thereby avodng an arbtrary cut-off v HHI, N = number 1 V of frms, v = market share of th frm, V = total market share. 5 k bank concentraton rato ndcates the percentage of a market or an ndustry accounted for by domnatng frms. 2 25

8 Banks and Bank Systems, Volume 5, Issue 1, 2010 assets rato s ncluded. Further, factors such as nflaton and GDP growth rate may greatly nfluence the proft makng opportuntes n bankng frms, these two varables are ncluded to control general economc envronment. Ownershp structure of the banks may lmt the decson makng capabltes of bankng nsttuton especally n state-owned banks (Goldberg, Lawrence G. and Anoop Ra, 1996; Molyneux, Phl, 1999). Most prevous studes have shown that prvately owned banks have relatvely more freedom to set frms operatonal polces and procedures. Therefore, t s expected a postve nfluence of prvate ownershp on frms proftablty Data and samples. The nformaton about all varables except banks effcency s gathered from an unbalanced panel data set spread over a sxteenyear cross secton of tme perod from publshed fnancal statements of local commercal banks. Snce the bank effcency scores are estmated based on 3-year mowng wndows, all data collected were adusted accordngly. 3. Results and dscusson Table 1. Descrptve statstcs of test data Ths secton presents the fndngs of the emprcal analyss. Frst, t revews descrptve statstcs and correlaton coeffcent of data related to varables used n the analyss. Next subsecton presents the results of the analyss Descrptve statstcs of test data. Table 1 and 2 summarze the descrptve data and Pearson correlaton coeffcent of test data. The standard devaton shows small statstcal dsperson n data used for estmatng equatons. Recorded low standard devatons ponted out that the data ponts are not hghly varable. Further, these statstcs show that there are no outlers n the data set. Varable Defnton Mean Std. dev. Proftablty Return on assets (ROA) Net nterest margn (NIM) Interest expenses to nterest ncome Techncal effcency (TE) CCR DEA estmated scores n frst phase Scale effcency (SE) BCC DEA estmated scores n frst phase Concentraton rato HHRF Total assets Market power Total assets share n the market Rsk Loan and advances to total assets Sze Ln[Total assets] GDP growth Natonal accounts Inflaton Change n Colombo consumer prce ndex Ownershp A dummy varable (prvate banks = 0 and state banks = 1) The estmated correlaton coeffcents show that there s a very lttle correlaton among varables ncluded nto the model. As explaned by Guarat (2003), f the par-wse correlaton coeffcent between two regresses exceeds 0.8, a serous problem of multcollnearty wll arse. Estmated par-wse correlaton coeffcent for explanatory varables shows two such relatonshps between SE and TE, and ownershp and market power. Exstence of multcollnearty lmts the explanatory power of the ndependent varable even f the Varables Proftablty Techncal effcency Table 2. Pearson correlaton coeffcent Techncal effcency Scale effcency Scale effcency Concentraton rato regresson has shown a hgh R 2 value. However, the reported regresson results n Table 4 show a relatvely hgh R 2 value wth ndvdually sgnfcant regresson coeffcents for the model varables. Guarat (2003) states that n one stuaton multcollnearty may not pose a serous problem, when R 2 s hgh and the regresson coeffcents are ndvdually sgnfcant as revealed by the hgher t value. Therefore, t s clear that there s no serous multcollnearty problem related to the model. Concentrato n rato Market power Market power Ownershp Ownershp Rsk Sze GDP Rsk Sze GDP Inflaton Inflaton Interest margn

9 Table 3 exhbts the aggregated effcency scores on dfferent bankng segments n Sr Lanka. Overall, the statstcs shows that there s a very lttle varaton n estmated effcency scores. Further, the savngs bankng sector ndcates relatvely hgher effcency than the commercal bankng sector. However, n ths paper greater emphass has been put on the nature of ownershp. The estmated effcency scores ndcate that state-owned banks are less effcent than the prvate banks. Further, these results ndcated that the neffcency of state-sector has resulted n the excessve scale of operatons. Table 3. Descrptve statstcs of estmated effcency scores ( ) Form of DMUs (banks) Techncal effcency Pure techncal effcency Scale effcency All [0.090] [0.026] [0.079] Commercal [0.095] [0.028] [0.085] Savng [0.028] [0.008] [0.026] State commercal [0.135] [0.037] [0.118] Prvate commercal [0.051] [0.023] [0.042] Old commercal [0.117] [0.028] [0.103] New commercal [0.055] [0.026] [0.045] 3.2. Regresson results. The man research queston rased n ths paper s whether market power whch results from hgh market concentraton and frms relatve market share or the frms effcency s mportant n determnng overall frm performance. The frst two regressons have used reduced-form-revenue equaton to nvestgate factors nfluencng the banks performance. Both ROA and NIM based regressons have gven smlar evdence for the nfluence of market structure and effcency on the frm s performance. The estmated coeffcent for market power and concentraton varables n both models were not statstcally dfferent from zero. Therefore, results found that nether market concentraton nor market power have sgnfcant assocatons wth banks proftablty and NIM. Therefore, ths study reects tradtonal MSH and concludes that nether collusve power enoyng by large banks nor hgh market power enoyng by ndvdual banks have a sgnfcant nfluence on frm performance n bankng ndustry n Sr Lanka. The frst regresson provded statstcally sgnfcant evdence that the man source of superor performance s the manageral effcency and not the collusve power. Further, scale effcency varables have shown a statstcally sgnfcant negatve relaton wth the performance by reectng the scaleeffcency verson of ESH. These emprcal fndngs show that scale of operaton s not a pre-condton to have superor performance. The second regresson provded statstcally sgnfcant evdence for assocatons of SE and TE wth NIM. However, the Banks and Bank Systems, Volume 5, Issue 1, 2010 sgn of the estmated parameter of the TE s not smlar to the predcted. Therefore, the evdence derved on the second model reects the TE verson of ESH. Further, the recorded negatve sgn mpled that the managerally effcent banks are chargng relatvely small nterest margn. Nevertheless, the statstcally sgnfcant postve coeffcent of SE fals to reect the SE verson of ESH. Equaton (2) also reects the SCP arguments havng statstcally nsgnfcant coeffcent for both varables whch represent market structure. The results have supported the scale effcency verson of ESH. The regresson result ponted out a statstcally sgnfcant negatve relatonshp between techncal effcency and NIM ndcatng that effcent banks charge lower net nterest margn than less effcent banks. Furthermore, statstcally sgnfcant postve coeffcent dentfed for SE ndcates that n relaton to NIM, SE s the man determnant. It shows that scale effcent frms can gan hgher NIM supportng the scale effcency verson of ESH whch says frms n optmum scale produce goods and servces at relatvely lower cost. Table 4. Estmated regresson coeffcent Dependent varables Independent varables Proftablty Gross nterest Concentraton margn Techncal effcency Scale effcency Concentraton Market power Ownershp Rsk Sze GDP growth Inflaton Intercept R-squared Adusted R 2 S.E. of regresson F-statstc *** [4.82] *** [-3.44] [-1.58] [1.51] *** [2.28] [1.08] [1.08] *** [4.26] *** [2.67] *** [-2.96] *** [-2.53] *** [2.28] [1.65] [-1.51] [-1.77] *** [3.24] [0.27] [-1.16] [-0.59] [0.52] *** [2.23] *** [-1.82] *** [-2.78] [1.81] *** [-2.44] *** [7.64] *** [7.93] *** [3.57] Market power *** [2.97] *** [-2.82] *** [-15.17] *** [5.75] *** [6.28] *** [2.05] *** [2.82] [-0.71] Estmated parameters for equatons (1) and (2) ponted at some nterestng fndngs on the relatonshp of market power and concentraton wth ROA and NIM. The varable representng market power has shown a postve relatonshp wth ROA and a negatve relatonshp wth NIM. These results predct the frms wth hgh market power are capable to earn hgher proft wth the lowerng nterest 27

10 Banks and Bank Systems, Volume 5, Issue 1, 2010 margn. On the other hand, the concentraton varable ndcated completely opposte relatonshps. Snce, any of the estmated coeffcents were not statstcally dfferent from zero, those fndngs were not able to make use of any concluson. The results of auxlary regresson whch addresses the multcollnearty problem are not dfferent from the orgnal regresson results. It ndcates sgnfcant negatve relatonshps of proftablty wth concentraton and market power reectng the both hypotheses. As explaned n the ESH verson of market structure hypotheses say that effcency nfluences market share of the frm and concentraton. Therefore, t s a necessary condton that techncal effcency should have a postve relatonshp wth both market power and concentraton. Testng two supplementary regressons has been run. The statstcal evdence derved from two regressons has not gven supportve evdence to reect the ESH. Further, both regressons have provded the condtonal support for the exstence of ESH n Sr Lankan bankng market. The coeffcents of control varables seem to be mxed wth the results of two regressons. The regresson results ponted out that prvately-owned banks have earned relatvely hgher proftablty and lower nterest margn than state-owned banks. It mples that prvately-owned banks are n a better poston of managng operatonal cost other than the nterest expenses and subsequently earn hgher proft than state-owned banks. Both regressons have exhbted postve sgn for the estmated coeffcent for varable represented operatonal rsk. Ths varable measured the sgnfcance of the bank s lendng portfolo n relaton wth the total assets. Keepng a large porton of assets on rsk lendng portfolo ncreases the rskness of the operaton. On the other hand, lowerng ths rato may lead to reducng the total earnngs assets of the frm. The regresson results confrmed the postve relatonshp between rsk and the bank s proftablty ndcatng that ncurrng hgher rsk forces banks to earn superor proft. Keepng more assets on lendng portfolo also may assst to earn hgh nterest ncome lowerng the nvestment n dle assets. Therefore, postve relatonshp between NIM and rsk can be expected. The results provded statstcally nsgnfcant evdence to support the above relatonshp. The control varable used to ncorporate sze was not able to provde evdence for sgnfcant assocaton wth ether varable. GDP growth and nflaton rate have been ncorporated to the regresson model to control the nfluence from macroeconomc envronment. Results ndcated that proftablty has statstcally sgnfcant postve relatonshps wth GDP growth and rate of nflaton. On the other hand, both varables were not able to provde statstcal evdence to prove that, nfluences of varables are not sgnfcantly dfferent from zero. The postve coeffcent of nflaton both wth NIM corroborates that lower nflaton has more pronounced downward effect on long-term nterest rates leadng to declnng nterest margn. Consequently, ths result forces reducton n proftablty. 4. Polcy mplcaton of emprcal fndngs On ths phase, the effect of contemporary polcy on bank concentraton s addressed. It s beleved that hgh market power and concentraton mprove the bank concentraton. Polcy makers predcted the hgh concentraton together wth few frms wth large market share may help to reduce the unnecessary overhead cost n the ndustry. The both performance measures used clearly reect that ether market share or concentraton are drectly related to the market share. The results show whle proftablty s postvely related wth the techncal effcency, t s negatvely related wth scale effcency. Further, the results show the hgher the techncal effcency bankers ganed, the lower the NIM they charged. The results also evdence scale effcency s the man factor nfluencng the banks nterest margn. Ganng optmum scale banks can mnmze the overhead cost and enoyed hgher NIM. Therefore, these results suggest polcy makers to focus on polcy reforms whch enhance the banks effcency. Concluson Ths paper has examned man structural and performance features of bankng ndustry n Sr Lanka. The study used four hypotheses proposed by Berger and Hannan (1997) and two performance measures, namely ROA (proftablty) and NIM. Emprcal results are not consstent wth both market power hypothess and structure-conductperformance hypothess. It s appeared that hgh market concentraton wth small number of large banks n the ndustry has ntensfed the competton. Confrmng the maor arguments rased by Molyneux (1999) aganst the proft-concentraton relatonshp, ths study totally reects the tradtonal SCP hypothess. However, the study s fndngs reected Goldberg and Ra (1996) s fndngs whch showed sgnfcant proft-market power relatonshp. Emprcal results ponted out that effcent operaton of bankng frms are vtal for havng hgher proftablty wth better NIM. 28

11 References Banks and Bank Systems, Volume 5, Issue 1, Asmld, Mette, Joseph C. Parad, Vanta Aggarwall, and Clare Schaffnt (2004). "Combnng DEA Wndow Analyss wth the Malmqust Index Approach n a Study of the Canadan Bankng Industry". Journal of Productvty Analyss, 21: 1, pp Avkran, N.K. (2000). "Decomposton the Techncal Effcency of Tradng Banks n the Deregulated Perod". Department of Fnance, Unversty of Queensland: Brsbane. 3. Banker, R.D., A. Charnes, and W.W. Cooper (1984). "Some Models for Estmatng Techncal and Scale Effcences n Data Envelopment Analyss". Management Scence, 30: 9, pp Berger, Allan N. (1995). "The Proft-Structure Relatonshp n Bankng-Tests of Market-Power and Effcent- Structure Hypotheses". Journal of Money, Credt and Bankng, 27: 2, pp Berger, Allan N. and T. Hannan (1989). "The Prce-Concentraton Relatonshp n Bankng". Revew of Economcs & Statstcs, 71: 2, pp Berger, Allan N. and T. Hannan (1993). "Usng Effcency Measures to Dstngush Among Alternatve Explanatons of the Structure Performance Relatonshp". Workng Paper Federal Reserve Board: Washngton, DC. 7. Berger, Allan N. and T. Hannan (1994). "The Effcency Cost of Market Power n the Bankng Industry: A Test of the 'Quet Lfe' and Related Hypotheses". Workng Paper 94-29, Vol Fnancal Insttutons Center, The What School, Unversty of Pennsylvana. 8. Bkker, Jacob A. and Katharna Haaf (2002). "Competton, Concentraton and Ther Relatonshp: An Emprcal Analyss of the Bankng Industry". Journal of Bankng & Fnance, 26: 11, pp Byeongyong, Paul Cho and Mary A. Wess (2005). "An Emprcal Investgaton of Market Structure, Effcency, and Performance n Property-Lablty Insurance". Journal of Rsk & Insurance, 72: 4, pp Charnes, A., W.W. Cooper, and E. Rhodes (1978). "Measurng Effcency of Decson Makng Unts". European Journal of Operaton Research, 2: 6, pp Charnes, Abraham, Wllem W. Cooper, Are Lewn, and Lawrence M. Seford (1994). Data Envelopment Analyss: Theory, Methodology and Applcaton. Boston/ Dordrecht, London: Kluwer Academc Publshers. 12. Charnes, Abraham, Wllem W. Cooper, Are Lewn, and Lawrence M. Seford (1997). Data Envelopment Analyss: Theory, Methodology and Applcaton. Boston/ Dordrecht, London: Kluwer Academc Publshers. 13. Coell, Tm, D.S.Prasada Rao, and George E. Battese (1998). An Introducton to Effcency and Productvty Analyss. London: Kulvwer Academc Publsher. 14. Fu, Xaoqng (Magge) and Shelagh Heffernan (2005). "Chna: The Effects of Bank Reform on Structure and Performance". Cass Faculty of Fnance Workng Paper WP-FF Cass Busness School, Cty Unversty: London. 15. Glbert, R. Alton (1984). "Bank Market Structure and Competton". Journal of Money, Credt & Bankng, 16: 4 (Part II), pp Goldberg, Lawrence G. and Anoop Ra (1996). "The Structure-performance Relatonshp for European Bankng". Journal of Bankng & Fnance, 20: 4, pp Guarat, Demodar N (2003). Basc Econometrc. Boston: McGraw Hll. 18. Lloyd-Wllams, D.M., Phl Molyneux, and John Thornton (1994). "Market Structure and Performance n Spansh Bankng". Journal of Bankng & Fnance, 18: 3, pp Molyneux, Phl (1999). "Increasng Concentraton and Competton n European Bankng: The End of Ant-trust?", Vol European Investment Bank, Cahers Papers. 20. Molyneux, Phl and Wllam Forbes (1995). "Market Structure and Performance n European Bankng". Appled Economcs, 27: 2, pp Moore, Robert R. (1998). "Concentraton, Technology, and Market Power n Bankng: Is Dstance Dead?", Vol Fnancal Industry Studes, Federal Reserve Bank of Dallas. 22. Panzar, J.C. and J.N. Rosse (1987). "Testng for Monopoly Equlbrum". Journal of Industral Economcs, 35: 4, pp Punt, L.W. and M.C.J. Van Roo (1999). "The Proft-structure Relatonshp, Effcency and Mergers n the European Bankng Industry: An Emprcal Assessment". WO Research Memoranda, Vol Research Department, Netherlands Central Bank: Amsterdam. 24. Rosse, J.N. and J.C Panzar (1977). "Chamberln vs Robnson: An Emprcal Study for Monopoly Rents". Bell Laboratores Economc Dscusson Papers. 25. Sealey, C.W. and Jemes T. Lndley (1977). "Inputs, Outputs and a theory of Producton and Cost at Depostary Fnancal Insttutons". The Journal of Fnance, 32: 4, pp Short, Brock K. (1979). "The Relaton Between Commercal Bank Proft Rates and Bankng Concentraton n Canada, Western Europe, and Japan". Journal of Bankng & Fnance, 3: 3, pp Smrlock, Mchael (1985). "Evdence on the (Non) Relatonshp between Concentraton and Proftablty n Bankng". Journal of Money, Credt and Bankng, 17: 1, pp Sundararaan, Vasudevan, Charles Enoch, Armda S. San Jose, Paul Lous Cerel Hlbers, and Russell C. Morett Krueger, Marna Slack, Graham L. (2002). Fnancal Soundness Indcators: Analytcal Aspects and Country Practces. Washngton: Internatonal Monetary Fund. 29

12 Banks and Bank Systems, Volume 5, Issue 1, Worthngton, Ian, Chrs Brton, and Andy Rees (2001). Economcs for Busness: Blendng Theory and Practce. London: Prentce Hall. 30. Yu, Pey and Werner Neus (2005). "Market Structure, Scale Effcency, and Rsk as Determnants of German Bankng Proftablty". Vol Faculty of Economcs and Busness, Unversty of Tübngen. Appendx 1. Measures of effcency Data envelopment analyss Ths paper utlzes data envelopment analyss (DEA) technques (a non-parametrc approach) to estmate effcency n Sr Lankan banks 1. Both the CCR model formulated by Charnes, Cooper and Rhodes (1978) and the BCC model formulated by Banker, Charnes and Cooper (1984) are used to estmate techncal and scale effcency of banks n Sr Lanka. The CCR model assumes constant return to scale and BCC relaxes ths assumpton by beng based on the varables returns to scale. Basc CCR formulaton (dual problem/envelopment form) Mn Subect to x x 0 y y r r0 0, for for r 1,2,... m 1,2,... k (1) where, y r s the amount of r th output produced by DMU usng x amount of I nput. denotes the CCR effcency of DMU. Both y r and x are exogenous varables and represents the ntensty varables assgned to each DMU under observaton. The soluton to the above mnmzaton problem determnes the values of ntensty of the varables. The frst constrant of the above lnear problem mples that the combnaton of the nputs of frms, such as banks, on the fronter must be less than or equal to the nput of the frm. The second constrant restrcts the observed output of frm s less than or equal to the lnear combnaton of the output of frms n the fronter. The orgnal CCR model assumed that all DMUs under consderaton were operatng n an optmum scale. The BCC- DEA formulaton relaxed the assumpton of optmum scale. The CCR model estmated TE. BCC accommodates the scale effect by relaxng the constant return to scale n CCR by ncorporatng a thrd constrant to the effcency evaluaton model. Generally, t reles on the convex combnaton of the effcent unts nstead of the lnear combnaton, as n the case of the CCR. The effcency estmaton of these two models can be used to dentfy the three components of effcency: techncal, pure techncal (PTE) and scale effcency. The BCC-DEA formulaton s gven below. Basc BCC formulaton (dual problem/envelopment form) Mn z 0 Subect to x x 0 y y r r0 0. PTE for r 1,2,... k 1 for 1,2,... m (4.1.5) The DEAs power of dscrmnatng neffcent unts from effcent unts depends on number of unts under observaton and the number of nputs and outputs of the model. Snce ths study s based on a relatvely small sample, vz., three year movng wndows (Charnes, Abraham, Wllem W. Cooper, Are Lewn, and Lawrence M. Seford, 1994) are used to construct producton fronters for measurng effcency, followng the prevous smlar studes (Asmld, Mette, Joseph C. Parad, Vanta Aggarwall, and Clare Schaffnt, 2004). CCR and BCC DEA formulatons are appled to estmate TE and PTE, respectvely. Prevous studes have employed an MPI lke ndex to decompose scale effect on a DMU s neffcency. A frm s TE s a functon of PTE and SE. Therefore, PTE should be separated from TE to dentfy SE (Coell, Tm, D.S. Prasada Rao, and George E. Battese, 1998). TE CCR = PTE BCC SE, (4.2.1) SE = TE CCR PTE BCC, (4.2.2) 1 Few reasons nfluenced on selectng DEA as the preferred method for estmatng effcency n banks n Sr Lanka. They are (1) non requrement of pre-specfed functonal form, (2) ablty to ncorporate a combnaton of nput and output varables, and (3) ts ablty to measure the effcency even usng a small number of observatons. 30