Investigating the Market-Structure - Performance Relationship in the Commercial Banking Sector: Evidence from Jamaica

Transcription

1 Investgatng the Market-Structure - Performance Relatonshp n the Commercal Bankng Sector: Evdence from Jamaca Sherene A. Baley 1 Fnancal Stablty Department Bank of Jamaca January 2007 Abstract Ths paper employs a two-stage estmaton procedure to evaluate the mpact of bank concentraton on performance. In the frst stage of the estmaton process, a stochastc cost fronter s estmated for the domnant commercal banks n Jamaca over the perod , usng both translog and Cobb-Douglas cost functons. The translog cost fronter model was found to be a more approprate ft for the data. As such, effcency estmates from ths cost fronter model served as endogenous nputs n the second stage of the estmaton procedure, where a VAR framework was employed to nvestgate the relatonshp between effcency, concentraton, and performance n the ndustry. The fndngs from the paper suggest that, on average, domnant banks n the ndustry would only need to reduce costs by 7.0 per cent n order to operate as effcently as possble. Results from the VAR framework reject the structure-market-performance hypothess. Rather, mprovements n effcency contrbute to ncreased proftablty for the domnant banks. However, mprovements n effcency for these domnant banks may not be reflected n ther prcng polces due to the absence of strong competton n the sector. As such, there s further scope for ntatves geared at lowerng nterest margns and stmulatng growth n the wder economy. JEL Classfcaton Numbers: C61; G21; L25 Keywords: Market structure; stochastc fronter; effcency 1 The vews expressed are those of the author and do not necessarly reflect those of The Bank of Jamaca.

2 Table of Contents 1. Introducton Prevous Lterature The Relatonshp between Market Structure & Performance Developments n Concentraton & Performance n the Commercal Bankng Sector: Methodology & Data Measures of Effcency Techncal Effcency Data Estmaton Results Techncal Ineffcency Estmates Impulse response Analyss Concluson and Polcy Recommendatons References

3 1. INTRODUCTION The Jamacan commercal bankng sector underwent sgnfcant structural changes subsequent to the perod of fnancal sector dstress durng the late 1990s. Ths perod of reform engendered changes n the ownershp structure and the degree of concentraton wthn the sector. At the start of the crss perod n 1996, there were nne banks n operaton, however, ths number declned to sx at the end of the crss perod n A prevous study on the mpact of ncreased concentraton on competton n the bankng system, found that there was declnng competton over the perod 1988 to However, the nature of the relatonshp between concentraton and performance n the sector remans unexplored. Proponents of the structure-conduct-performance (SCP) hypothess post that market players n concentrated ndustres have the capacty to rase prces through monopolstc conduct and that these frms may gan large market shares and earn supernormal profts. At the same tme, ncreased proftablty may be evdence of ncreased effcency wthn an ndustry due to lower costs of producton, enablng some frms to compete more aggressvely and capture a bgger market share. Snce the consoldaton n the sector subsequent to the crss perod, ntermedaton margns have fluctuated wthn a band of 10.0 per cent to 15.0 per cent, despte the trend declne n polcy rates snce Ths has rased concerns for polcymakers and has resulted n varous efforts to ncrease competton n the sector. It s mportant to note that the alternatve hypotheses have opposng polcy mplcatons. If banks proftablty were drven by ncreased concentraton, regulators would fnd t benefcal to enforce anttrust regulaton to move prces towards compettve levels and allocate resources more effcently. Alternatvely, f greater proftablty s attrbutable to ncreased effcency, then regulaton geared at dsmantlng effcent frms and/or preventng mergers may rase costs and lead to less favourable prces for consumers. As such, the paper explores whether reductons n domestc nterest rates should be 2 See Duncan (2002) 2

4 accompaned by reforms such as the reducton of entry barrers n order to mprove the compettve structure n the commercal bankng sector. The lnk, f any, between market concentraton and bank performance receved consderable attenton durng the 1990s, gven substantal fnancal sector consoldaton across a large number of ndustralzed countres. Regulators n concentrated markets have been faced wth the ssue of determnng whether proftablty s nfluenced by ncreased effcency or the monopolstc behavour of the more domnant market players and the resultng mplcatons for consumer welfare. Cvelek and Al-Alam (1991) noted that the bankng sector s mportant to the economy and hence emprcal evdence on the concentraton-performance relatonshp can nform regulatory polces that govern the bankng system. Increasng bankng sector concentraton may rase the cost of funds, thereby reducng the level of ntermedaton n the economy and mparng economc growth. The ssues outlned above rases an mportant polcy queston: Dd the ncreased concentraton n the sector begnnng n the latter part of the 1990s lead to reduced effcency and ncreased proftablty n the sector? As such, the purpose of the emprcal nvestgaton s two-fold. Frst, effcency estmates are derved for domnant banks n the Jamacan bankng sector. 3 Second, these estmates are employed, through a VAR framework, n assessng the concentraton-performance relatonshp n the bankng sector. The results of the study wll nform the drecton of polcy for the commercal bankng sector as t relates to the safeguardng of consumer welfare and regulatng competton wthn the sector n order to engender lower costs for bank servces. 3 The domnant banks used n the study are the two largest commercal banks. 3

5 The paper s organzed as follows: Secton 2 examnes the theoretcal underpnnngs of the lnk between market concentraton and performance. Secton 3 provdes a bref outlne of the structure-performance relatonshp n the commercal bankng sector. Secton 4 dscusses the methodology and the data employed. Secton 5 presents the fndngs of the model. The polcy mplcatons of the results and the concluson are outlned n secton Prevous Lterature 2.1 The Relatonshp between Market Structure and Performance Evaluatng bank performance s a complex process that nvolves assessng the nteracton between a bank s nternal operatons and external envronment. An mportant aspect of the assessng the mpact of the external envronment s by examnng the nfluence of market structure on bank performance. The relatonshp between market structure and bank performance has been examned at length n the lterature, partcularly by assessng the mplcatons of ndustry concentraton for operatonal effcency n the sector. The relatonshp between ndustry concentraton and commercal bank proftablty or performance s explaned by two contrastng hypotheses. Alternatve explanatons provded by the tradtonal SCP hypothess and effcent market hypothess (EMH), gve nsght on the drecton and nature of ths relatonshp. 4 Proponents of the tradtonal paradgm post that there s a one-way relatonshp between concentraton and proftablty, due to the fact that market concentraton lowers the cost of colluson between frms, nfluencng ncreases n prces and hgher than normal profts. 5 The key ssue here s that hgher seller concentraton fosters explct colluson on the part of frms, allowng all frms n the market to earn monopoly rents. Ths suggests that where markets are conducve to monopolstc conduct, due to barrers to entry and market concentraton, frms have greater autonomy to rase prces above ther costs, resultng n excess profts. As a consequence, the letmotv of the tradtonal 4 The structure-conduct-performance theory s also known as the colluson hypothess. 5 See Chrwa (2003). 4

6 paradgm s that market structure nfluences bank performance through the nsttuton s prcng polcy or prcng behavour. The theoretcal mplcaton of the tradtonal framework s that n concentrated markets consumers face less favourable prces due to the non-compettve behavour that generally exsts n these markets. Hence, lower levels of proftablty are antcpated n markets where there s perfect demand elastcty compared to all other markets. Berger and Hannan (1989) post that f the SCP hypothess reflects ant-compettve prcng, then banks wll be able to charge lower depost rates and/or hgher loan rates. 6 Ths relatonshp has been wdely tested n the feld of ndustral economcs. Ban (1951) found evdence of a postve and sgnfcant relatonshp between concentraton and proftablty based on US manufacturng data. However, later studes usng bankng sector data have revealed mxed results. The specfcaton of the SCP model n bankng s based on a proft-concentraton relatonshp that s defned as follows 7 : π = f CR, X ) (1) ( where, π, s a proft measure of the th bank, ths can be captured by the rate of return on captal or assets (ROC or ROA), CR, s the banks ndex of concentraton and denotes a vector of control varables that account for frm-specfc and market-specfc characterstcs whch affect prces through market or cost consderatons. The mxed emprcal results on the SCP model have undermned the mportance of the model n explanng bank behavour. X, The EMH emerged from crtcsm of the tradtonal paradgm. Rather than beng a random event, the effcency theory bases ncreased market concentraton on the superor productvty or effcency of the leadng frms. Ths occurs because ncreased productvty and proftablty enables frms to offer compettve rates on loans and/or deposts, nducng them to gan larger market shares and further ncrease proftablty. 6 See Goldberg & Ra (1996) 7 See Chrwa (2001) 5

7 Accordng to the EMH, frm-specfc effcences encourage unequal market shares and hgh levels of concentraton. Molyneux & Forbes (1995) and Smrlock (1985) posted that the EMH s specfed as a market share proftablty relatonshp. The market share varable captures frms superor effcency n obtanng a larger porton of the market and equaton (1) s respecfed as follows: π = f CR, MS, X ) (2) ( Tests of both hypotheses are captured n equaton (3): m 0 + β1cr + β1ms + α1 = 1 π = β X (3) where the sgnfcance of MS n equaton (3) would mply support for the effcent structure hypothess; whle the sgnfcance of the concentraton rato would support the tradtonal hypothess. However, Berger and Hannan (1993) have argued that the valdty of the market share varable n testng for effcency s dependent on whether the varable can be consdered as a proxy for the effcency of larger frms rather than as measure of market power. Gven the spurous nature of ths relatonshp, emprcal fndngs based on these two market structure-performance hypotheses have been nconclusve. The Berger & Hannan Model Berger & Hannan (1993) developed a model to drectly nclude effcency n examnng the relatonshp between market structure and performance. The model captures effcency from two perspectves: Scale effcency and X-effcency. A smultaneous test of the competng hypotheses s presented n equaton (4). P = f ( X EFF, S EFF, CONC, Z ) + e (4) m where, P, represents measures of frm performance, X EFF and S EFF, represent measures of techncal effcency and scale effcency, respectvely, and, Z, s a vector of 6

8 control varables. The scale effcency verson of the effcent structure hypothess ndcates that some frms produce at a more effcent scale than others, whch could result n lower nterest margns and ncreased market shares. Ths measure captures whether the bankng frm operates at an optmal scale of producton at whch the per unt average producton cost s mnmzed. Under the X-effcency hypothess, t s mportant to determne whether a frm s producng as effcently as t can, gven ts sze. Hence, frms wth superor management or producton processes can operate at lower costs and subsequently reap hgher profts. Ths encourages growth n market share and fuels ncreased ndustry concentraton. Generally, emprcal studes on bank effcency revealed that X-effcency s more mportant n explanng dfferences n bank costs than scale effcences (see Berger, Hunter, and Tmme 1993). As such, the paper focuses solely on X-effcency estmates n nvestgatng the relatonshp specfed n equaton (4). The methodology for dervng effcency estmates s dscussed n secton Developments n Concentraton & Performance n the Commercal Bankng Sector: The process of economc lberalzaton n the Jamacan economy durng the late 1980s and early 1990s engendered structural changes nvolvng the deregulaton of the fnancal sector. Ths process of reform led to rapd expanson n the number of commercal banks n the ndustry due to the lberalzaton of lcensng requrements n the sector. Aganst ths background, there was a marked declne n concentraton n the sector, as measured by the Herfndahl - Hrschman ndex (HHI) and the two-frm concentraton rato (see Fgure 1). 8 8 The Herfndahl-Hrschman ndex s a wdely used measure of concentraton. It s computed as the sum of all the banks squared market shares, where market share s based on ether deposts or assets, whle the two-frm concentraton rato s represented as the per cent of deposts held by the two most domnant banks. Return on assets s utlzed as an overall measure of performance and s captured by net profts as a proporton of average total assets. 7

9 Fgure 1 Concentraton n the Commercal Bankng Sector HHI(d) HHI(a) 2-frm concen The concentraton ndces were computed on a quarterly bass over the perod March 1988 December 2005 and dsplayed a smlar pattern durng ths perod (see Fgure 1). The movement n these ndces was prmarly drven by the performance of the two most domnant banks, as the two-frm concentraton rato exceeded ndustry concentraton for most of the sample perod (see Fgure 1). Ths mples that ndustry concentraton was largely nfluenced by the duopolstc behavour present n the ndustry. per cent The declne n the HHI that was observed durng the lberalzaton era contnued nto the post-lberalzaton perod (see Fgure 1). The HHI (deposts) fell to a low of at end June 1994, from approxmately at the end of 1989, representng a 15.0 per cent declne over ths perod. The proftablty of the domnant market players was nversely related to ndustry concentraton durng ths perod. 8

10 Fgure Concentraton & Return on Asset per cent HHI(d) domnant banks Other banks Fgure 2 shows that for the smlar perod, the ROA of the two most domnant banks ncreased from 0.6 per cent to a hgh of 2.0 per cent. Ths contradcts the fndngs of the structuralst theory, whch s suggestve that a declne n concentraton s assocated wth reduced proftablty for the domnant banks. Rather, ths mprovement n performance s suggestve of reductons n the costs of producton and ncreases n productvty of the domnant banks due to the ncreased competton n the sector. Fgures 3 & 4 also show that the mprovement n the ROA of these banks was drven by strong ncreases n net nterest margn and non-nterest ncome as a proporton of total assets durng the post-lberalzaton perod. Ths occurred relatve to much weaker growth n operatng expenses as a proporton of assets. In fact, operatng costs as a proporton of average total assets were relatvely stable averagng 4.0 per cent over the perod 1992 to

11 Fgure Net Interest Margn to Total Assets per cent domnant banks other banks Fgure Non-nterest Income to Total Assets 2.0 per cent domnant banks Other banks The contanment n operatng costs as a proporton of total assets was reflectve of a moderaton n the growth of nterest costs (see Fgures 5 & 6). 10

12 Durng the crss perod of the late 1990s, the return on assets for the domnant banks averaged 0.6 per cent relatve to an average ROA of negatve 1.5 per cent for the other banks over the perod. There was also a slower pace of growth n operatng expenses as a proporton of average total assets relatve to other banks durng the perod of dstress. Nonetheless, ths performance was largely due to greater effcency n the operatons of one domnant bank. Fgure Operatng Costs to Total Assets 12.0 per cent domnant banks other banks 80.0 Fgure 6 Interest Expense as a proporton of Operatng Costs per cent other banks domnant banks

13 Subsequent to the dstress perod, there was substantal consoldaton n the sector followng the restructurng of a number of faled banks. 9 Aganst ths background, the HHI ncreased to an average of for the post-crss perod of 2000 to 2005 relatve to an average of durng the crss perod (see Table 1). At the same tme, the ROA and NIM of the domnant banks reflected respectve averages of 0.8 per cent and 1.7 per cent durng the post-crss perod relatve to 0.6 per cent and 1.2 per cent durng the crss perod. Addtonally, there was a postve, albet margnal, correlaton between the HHI ndex and the NIM of the domnant banks durng 2000 to 2005 (see Table 1). Ths performance s somewhat consstent wth the structuralst theory, whch ndcates a postve assocaton between concentraton and proftablty. Nonetheless, the mprovement n performance of the domnant banks could be reflectve of mprovements n effcency relatve to the crss perod, whch would mply support for the EMH. Therefore, despte the postve relatonshp between concentraton and performance durng ths perod, the exact nature of ths relatonshp remans uncertan. Ths underscores the relevance of explctly accountng for effcency, n order to ascertan a more precse relatonshp between concentraton and performance. Table Mean Performance Ratos (per cent) & Concentraton Measures: Sub-Perods Concentraton HHI - deposts Performance ROA NIM to total assets Correlaton HHI & ROA HHI & NIM to total assets The restructurng process was spearheaded by the Fnancal Sector Adjustment Company (FINSAC), whch was establshed by Government n 1997 to resolve the dffcultes n the fnancal sector. 12

14 4. Methodology & Data 4.1 Measures of Effcency Techncal Effcency Techncal effcency or X-effcency s a measure of how effectvely banks utlze nputs to produce a gven level of output. The bank s effectveness n achevng the optmal mx of cost-mnmzng nputs can be specfed by an effcent cost fronter. Gven the lkelhood of bank-by-bank devatons n the effcent cost fronter, t s necessary to specfy a stochastc cost functon. As such, ths paper employs the stochastc cost fronter proposed by Agner et al. (1977), where devatons from the effcent cost fronter are captured by a random nose, v, and an neffcency component, u. 10 The cost functon s represented as: ln tc = f ( y, p ) + ε (5) where ε = u + v, y, s the output of each bank, p, s the cost of nput and, v, s statstcal nose dstrbuted normal 0, σ ). 11 U s a an neffcency measure whch can ( 2 follow a truncated or half-normal dstrbuton and measures the ndvdual frm s devaton from the effcent cost fronter, due to non-optmal employment of the quantty or mx of nputs gven ther prces as a result of management errors. Ths varable s referred to as techncal neffcency. However, t relates to both techncal neffcences from usng too much of the nputs to produce the same output and allocatve neffcences from falng to react optmally to the relatve prces of nputs. The varable v s an exogenous component whch s due to data or measurement error or unexpected and uncontrollable factors such as labour strkes and war that are not under the nfluence of management. 10 Agner et al. utlzed the approach to nvestgate cost effcences. 11 Maxmum lkelhood estmaton technques are utlzed n estmatng the coeffcents. 13

15 Estmates of u or techncal neffcency are derved from the stochastc fronter for each bank. In nvestgatng the concentraton-performance relatonshp outlned n equaton 4, u s substtuted for the effcency varable, X-EFF. The predcted relatonshp between u and proftablty s negatve when the performance measure s defned by the return on assets, snce lower neffcency s assocated wth larger profts. Whle for the net nterest margn, the relatonshp s postve snce hgher neffcency mples a larger net nterest margn. The log-lkelhood functon for equaton (5) s specfed as: N N N λ ln L = ln N lnσ + ln ψ π σ (6) = 1 = 1 σ wth N denotng number of banks and ψ s the standard normal cumulatve dstrbuton. Equaton (8) shows that the rato of varablty, σ, can be used to measure a frm s mean neffcency by: σλ φ( λ / σ ) λ E( u / ) = + (8) 2 1+ λ ψ ( λ / σ ) σ where σ [ σ u + σ v ] functon. =, λ = σ u / σ v, and φ( ) s the standard normal densty Ths paper employs two varatons of the model specfed n equaton (5), n estmatng a cost fronter for banks n Jamaca. A translog cost functon s consdered because of ts flexblty n allowng for nput substtutablty (see equaton (9)) as well as a smpler Cobb-Douglas cost functon whch s specfed n equaton (10). 14

16 ln tc = α / 2 α ln( y ) j= 1 h= 1 3 j= 1 β j ln( p β jh ln( p j )ln( ph) + j ) + 1/ = 1 j= = 1 k= 1 αk ln( y )ln( y δj ln( y )ln( p j k ) + ) + 1/ j= 1 h= 1 β jh ln( p j )ln( ph) (9) ln tc = β + β1 ln( y1t ) + β2 ln( y2 t ) + β3 ln( p1 t ) + β4 ln( p2 t ) + β5( p3 t ) + ( ut + v 0 t ) (10) where the subscrpts and t refer to the th bank and the t th perod. In both models, the term, U, represents non-negatve random varables that are assumed to be ndependently dstrbuted as truncatons of the N( δ, σ ) dstrbuton, where Z t are factors that affect frm effcency, whle δ s a vector of parameters to be estmated. As a consequence, neffcency s modelled usng the auxlary model specfed n equaton (11). U t = Ztδ = δ 0 + δ1z1t + δ 2z2t + wt Z t 2 u (11) where z 1 = total assets and z 2 = return on assets Addtonally, for the two models consdered, TC represents total operatng and nterest costs. Two outputs are employed n the model; loans, whch s the prmary output, y 1, and all other earnng assets, y 2, s ncluded as a secondary output. There are three nputs, wth prces defned as the prce of labour, p 1, prce of fxed captal, p 2, and borrowed funds, p 3. P 1 s defned as personnel expenses dvded by total assets, P 2 s defned as captal and occupancy expenses dvded by fxed assets, and P 3 s defned as total nterest expenses dvded by nterest bearng labltes. Consstent wth lnear homogenety condtons, TC and the prces of all nputs are normalzed by the prce of labour ( p 1). Therefore, the transformed varables are denoted as TC *, p 2 * and p 3 *. 15

17 The cost functons defned by equatons (9) and (10) are estmated usng FRONTIER, an econometrc software package desgned to provde maxmum lkelhood estmates of a varety of stochastc fronters. 12 FRONTIER follows a three-step estmaton procedure. Frst, ordnary least squares (OLS) estmates of the functon are obtaned as startng values. Next, a two-phase grd search s conducted to refne the startng values. 13 Fnal estmates are obtaned teratvely usng the Davdson, Fletcher, and Powell Quas-Newton Method. Cost effcency estmates derved from the model range over the nterval [ 1, ], wth a score of one ndcatng full effcency, whch means that the frm s operatng on ts effcent cost fronter. The amount by whch the score devates from 1 s a measure of techncal neffcency. 4.2 Data The model utlzes quarterly commercal bankng system data coverng the perod March 1989 to December Average neffcency measures are derved for domnant banks for each quarter over the sample perod. The bank performance measures utlzed n the emprcal analyss are return on assets and the net nterest margn. Return on assets s computed as the rato of net ncome to average total assets whle net nterest margn s the nterest revenue less nterest expenses, as a proporton of total assets and s used as a proxy for the prcng ablty of banks. The HHI ndex computed for the bankng sector s used as a measure of concentraton. A dummy varable s ncluded to capture the fnancal crss that occurred durng the second half of the 1990s. A dynamc VAR framework s utlzed n analysng the nature of the concentraton proftablty relatonshp by ncorporatng neffcency estmates, performance and concentraton measures and the dummy varable as endogenous varables See Coell (1996) for a complete dscusson of FRONTIER. 13 These estmates (except for the ntercept) are unbased. 14 Ineffcency estmates ncluded n the VAR model are based on the cost functon whch best fts the data. 16

18 5.0 Estmaton Results 5.1 Techncal Ineffcency Estmates Translog Cost Functon Estmates The maxmum lkelhood estmates of the translog cost functon defned by equaton (9) are presented n Table 2. The value of the γ parameter, whch captures the varablty related to the techncal neffcency component, s sgnfcant. Ths ndcates that the translog model s a sgnfcant mprovement over the standard OLS functon. Addtonally, output varables, y1 and y2, and the prce varable p3 * along wth varous nteracton varables are postve and sgnfcant at the 10.0 per cent level (see Table 2). However, asset sze and ROA were found to be statstcally nsgnfcant ndcatng that ncreases n these varables were not nfluental n determnng techncal neffcency. Table 2. Model parameters of the stochastc Translog cost functon Varable Coeffcent t-rato Dependent varable: ln (cost) constant -48.2* ln(y1) 7.6* 6.50 ln(y2) -2.1* ln(p2*) ln(p3*) -1.4* ln(y1)ln(y1) ln(y1)ln(y2) ln(y2)ln(y2) 0.1* 5.10 ln(p3*)ln(p3*) 0.5* 4.80 ln(p3*)ln(p2*) 0.3* 2.40 ln(y1)ln(p3*) ln(y1)ln(p2*) -0.2* ln(p3*)ln(p3*) ln(y2)ln(p3*) 0.2* 4.40 ln(y2)ln(p2*) Gamma 0.98* 5.70 constant (delta0) asset sze ROA "*" Sgnfcance at 10% level 17

19 Fgure 7 Based on the results from the estmated translog cost fronter, techncal neffcency for the domnant banks averaged 6.8 per cent over the sample perod. Ths ndcates that on average, domnant banks were ncurrng 6.8 per cent hgher costs than necessary. Addtonally, Fgure 7 shows that for most of the perod, these banks operated close to ther effcent cost fronter. Nonetheless, the results ndcate that there was a marked ncrease n techncal neffcency durng the pre-lberalzaton perod due to absence of strong competton n the sector. However, ths performance mproved substantally n the post-lberalzaton perod. Not surprsngly, techncal neffcency also ncreased sharply to 23.0 per cent durng the crss perod before moderatng to fluctuate n a band of 3.0 to 7.0 per cent over the rest of the sample perod. 18

20 5.1.2 Cobb-Douglas Cost Functon Estmates Table 3. M odel parameters of the stochastc Cobb-Douglas cost functon Varable Coeffcent t-rato Dependent varable: ln (cost) constant 2.1* ln(y1) 0.43* ln(y2) 0.56* ln(p1*) 0.61* ln(p2*) 0.02* Dependent Varable: Ineff constant 2.92* 8.04 asset sze -0.11* ROA -8.52* Gamma 0.99* "*" Sgnfcance at 5% level The results n Table 3 are based on the Cobb-Douglas cost fronter and ndcate that all coeffcents are sgnfcant at the 5.0 per cent level. In partcular, gamma s statstcally sgnfcant ndcatng that the proporton of the one-sded error component n the total varance of the error terms n the model s approxmately 99.0 per cent. As such, techncal neffcency s the man source of random error n the model. The coeffcents on asset sze and ROA are negatve and statstcally sgnfcant ndcatng that ncreased asset sze and proftablty contrbute to declnes n techncal neffcency. Fgure 8 shows that the pattern of the techncal neffcency estmates are smlar to that observed based on the translog functon. Nonetheless, techncal neffcency estmates based on the translog and Cobb-Douglas functons devated sharply pror to the crss perod but were more closely related subsequent to ths perod (see Fgure 9 & Table 4). Estmates for both cost functons are compared n order to ascertan whch model s a more approprate ft for the data. The lkelhood rato test statstc was compared to the Ch-square crtcal value at the approprate degrees of freedom. 15 The null hypothess of the more restrctve Cobb-Douglas cost functon was rejected at the 5.0 per cent level. As 15 The test statstc was computed based on the lkelhood functon values for the Cobb-Douglas and translog cost functons. Snce the test statstc follows a mxed Ch square dstrbuton, crtcal values were based on Kodde Palm (1986). 19

21 such, neffcency estmates from the translog cost functon were used as nputs n the VAR framework. Fgure 8 Fgure Cost Effcency Estmates Mar-89 Mar-90 Mar-91 Mar-92 Mar-93 Mar-94 Mar-95 Mar-96 Mar-97 Mar-98 Mar-99 Mar-00 Mar-01 Mar-02 Mar-03 Mar-04 Mar-05 Translog Cobb Douglas 20

22 Table 4 Average X-neffcency Estmates over dfferent Sub-perods Sub-perods Average X-neff Estmates (per cent) Translog Cobb-Douglas Pre-Lberalzaton Post Lberalzaton Crss perod Post Crss Entre Sample Impulse Response Analyss Fgure 10 plots the mpulse responses of ROA, HHI and NIM wth respect to a one standard devaton ncrease n techncal neffcency over a horzon of thrty-sx months. The VAR coeffcent and standard errors from the model are computed by the Monte Carlo method wth 1000 repettons (of 2 standard devatons). 21

23 Fgure 10 Generalzed mpulse response analyss s utlzed, as t s nvarant to the orderng of the varables n the VAR, resultng n a unque soluton. A lag order of one was selected based on the Schwarz nformaton crteron and the Hannan-Qunn nformaton crteron. 16 The mpulse response suggests that shocks n neffcency nfluences declnes n overall proftablty wthn the frst three months, based on the ROA. Thus, proftablty s nversely related to ncreases n neffcency. 16 See Table 5 n Appendx 22

24 Ths fndng supports the EMH, whch ndcates that ncreases n effcency are assocated wth ncreases n proftablty. It s also observed that ncreases n neffcency contrbute to declnes n concentraton, suggestng that effcent frms are lkely to gan larger market shares. However, ncreases n neffcency result n a declne n NIM, mplyng that lower neffcency s assocated wth ncreases n NIM. Ths mples that declnes n neffcency are not translated nto lower net nterest margns beng appled by the domnant banks. A possble explanaton for ths fndng s that domnant banks have lmted ncentve to translate productvty mprovements nto lower nterest margns because of the weak competton n the sector. As such, the relatvely hgh nterest margns n the domestc bankng sector are ndcatve of the absence of strong competton rather than unfavourable prcng due to the presence of a large degree of neffcency n the operatons of domnant market players. Fgure 11 shows the response of techncal neffcency (neff), ROA and NIM to a 1.0 unt shock n concentraton. The fndngs do not support the SCP hypothess. The results ndcate that ncreases n concentraton are assocated wth strong declnes n both the proftablty measures wthn the frst three months. The results also show that ncreased concentraton contrbutes to lower neffcency. 23

25 Fgure 11 24

26 7.0 Concluson and Polcy Recommendatons The results suggest that domnant banks are operatng wth a hgh degree of effcency. Based on fndngs from the translog cost functon, domnant banks would only need to reduce costs by 7.0 per cent to acheve ther optmal level of effcency. Ths estmate compares favourably to fndngs n the US bankng ndustry. 17 The pattern of techncal neffcency estmates for the domnant banks ndcates that there was ncreased neffcency pror to the lberalzaton perod as well as durng the crss perod. Both cost fronter approaches ndcate that techncal neffcency has remaned relatvely stable subsequent to the crss perod, fluctuatng n narrow band of 3.0 per cent to 5.0 per cent. Results from the VAR framework reject the SCP hypothess, ndcatng that proftablty of the domnant banks n the ndustry s not drven by monopolstc conduct. Whle the proftablty of the domnant banks s drven by the effcency n ther operatons, ths performance s not reflected n the prcng polces of these banks. Ths suggests that despte gans n productvty, domnant banks are unlkely to admnster lower net nterest margns n the absence of greater competton n the bankng system. There s further need for regulatory authortes to engneer contnued ncremental reductons n polcy rates n an effort to ncrease competton n the bankng sector and avod destablzng the system. The lower nterest rate envronment could stmulate ncreased competton n the sector, as commercal banks reduce ther relance on earnngs from Government debt ssues and realgn ther asset portfolos towards core busness actvtes. Addtonally, contnued mprovement n the fscal poston and Government s debt sustanablty should further reduce the bankng sector s relance on earnngs from debt nstruments. 17 See Kwan (2001). 25

27 Reductons n domestc nterest rates should be accompaned by moral suason by regulatory authortes. Polcymakers can promote a compettve envronment by encouragng domnant banks to ncrease performance by expandng busness volumes by narrowng ther nterest margns. The ncreased compettveness of the domnant market players could nfluence greater competton n the supply of loanable funds n the sector. There s some debate as to the compettve structure that optmzes effcency and fnancal stablty. Market power can contrbute to stablty by mtgatng rsk-takng behavour and provdng ncentves to screen and montor loans, whch can mprove the qualty of banks portfolos. In ths context, rather than pursung efforts to mnmze market power, regulatory authortes should seek to facltate an envronment that promotes compettve behavour. Regulatory authortes can stmulate competton by promotng the establshment of ndependent credt ratng agences, as the lack of accurate, relable nformaton about borrowers ablty to pay ncreases credt rsk and hence lendng rates. It s also mportant to ensure that the regulatory framework s accommodatve of product nnovatons, as strong restrctons on new products and servces can decrease compettveness n the bankng ndustry. Claessens and Laeven (2003), estmated compettveness ndcators for a large cross-secton of countres and found evdence that fewer actvty restrctons n the bankng sector can ncrease compettveness n the system. Addtonally, regulatory authortes must carefully evaluate the costs and benefts of mergers and acquston n order to safeguard consumer welfare and promote competton n the sector. 26

28 8.0 References Agner, D., lovell, C.A.K., and Schmdt, P. (1977), Formulaton and Estmaton of Stochastc Fronter Producton Functon Models, Journal of Econometrcs, 6, pp Ban, J.S. (1951), The Relaton of Proft Rate to Industry Concentraton: Amercan Manufacturng, Quarterly Journal of Economcs, 65, pp Berger, A.N. (1995), The Proft-Relatonshp n Bankng-Tests of Market-Power and Effcent-Market Hypotheses, Journal of Money, Credt and Bankng, 27, pp Berger, A.N., and Hannan, T.H. (1989), The Prce-Concentraton Relatonshp n Bankng, Revew of Economcs and Statstcs, 71, pp Berger, A.N., Hunter, W.C., and Tmme, S.G. (2003), The Effcency of Fnancal Insttutons; A Revew and Prevew of Research Past, Present, and Future, Journal of Bankng and Fnance, 17, pp Cebenoyan, A.S., Cooperman E.S., Regster C.A., and Hudgns S.C., The Relatve Effcency of Stock versus Mutual S&Ls: A Stochastc Cost Fronter Approach, Journal of Fnancal Servces Research. Chrwa, Ephram, W.T. (2003), Determnants of Commercal Banks Proftablty n Malaw: A Contegraton Approach, Appled Fnancal Economcs, 13, pp Chrwa, Ephram, W.T. (2001), Market Structure, Lberalzaton and Performance n the Malawan Bankng Industry, AERC Research Paper 108, Narob: Afrcan Economc Research Consortum. Cvelek, M.M., and Al-Alam, M.W. (1991), An Emprcal nvestgaton of the Concentraton-Proftablty Relatonshp n the Jordanan Bankng System, Savngs and Development, 15, pp Claessens, Stjn, and Laeven, L. (2003), What Drves Bank Competton? Some Internatonal Evdence, World Bank Polcy Research Paper No Coell, T.J.(1996), A Gude to Fronter Verson 4.1: A Computer Program for Stochastc Fronter Producton and Cost Functon Estmaton, Centre for Productvty and Effcency Analyss, Unversty of New England, Armdale. Coell, T.J., Sngh, S. and Flemng, E. (2000), Measurement of Techncal and Allocatve Effcency n Indan Dary Processng Plants: An Input Dstance Functon, CEPA Workng papers, No. 3/

29 Duncan, D. (2002), Testng for Competton n the Jamacan Bankng Sector: Evdence from Bank Level Data, Avalable va the Internet: Frame, W.S. and Coell, T.J. (2001), U.S. Fnancal Servces Consoldaton: The Case of Corporate Credt unons, Revew of Industral Organzaton, 18, pp Katb, M. Nasser (2004), Market Structure and Performance n the Malaysan Bankng Industry: A Robust Estmaton, Avalable va the Internet: Kodde, D.A., and Palm, F.C. (1986), Wald Crtera for Jontly Testng Equalty and Inequalty Restrcton, Econometrca, 54, pp Kwan, Smon, H. (2001), The X-Effcency of Commercal Banks n Hong Kong, Federal Reserve Bank Workng Paper. Lmam, Imed, Measurng Techncal Effcency of Kuwat Banks, Avalable va the Internet: Molyneux, P. and Forbes, W. (1995), Market Structure and Performance n European Bankng, Appled Economcs, 27, pp Ra,A. and Goldberg, L.G.(1996), The Structure-Performance Relatonshp for European Bankng, Journal of Bankng and Fnance, 20, pp Ra, A. (1996), Cost Effcency of Internatonal Insurance Frms, Journal of Fnancal Servces Research, 10, pp Robnson, J. (2002), Commercal bank Interest Rate Spreads n Jamaca: Measurement, Trends and Prospects (unpublshed; Kngston: Bank of Jamaca) Smrlock, M. (1985), Evdence on the (Non) Relatonshp between Concentraton and Proftablty n Bankng, Journal of Money, Credt and Bankng, 17, pp Worthngton, A. (1998), The Determnants of Non-bank Fnancal Insttuton Effcency: A Stochastc Cost Fronter Approach, Journal of Econometrcs, 8, pp

30 APPENDIX Fgure Intermedaton Margns 20.0 per cent Spread - loans & depost rates Spread - loan & tme depost rates Table 5 VAR Lag Order Selecton Crtera Lag LogL LR FPE AIC SC HQ NA 1.11E e-18* * * * E E E E * 1.47E * ndcates lag order selected by the crteron LR: sequental modfed LR test statstc (each test at 5% level) FPE: Fnal predcton error AIC: Akake nformaton crteron SC: Schwarz nformaton crteron HQ: Hannan-Qunn nformaton crteron 29