EUROSYSTEM. June Working

Transcription

1 CENTRAL BANK OF CYPRUS EUROSYSTEM WORKING PAPER SERIES Measurng Market Power n the Bankng Industry n the Presencee of Oortunty Cost Antons Mchs June 2013 Workng Paer

2 Central Bank of Cyrus Workng Paers resent work n rogress by central bank staff and outsde contrbutors. They are ntended to stmulate dscusson and crtcal comment. The onons exressed n the aers do not necessarly reflect the vews of the Central Bank of Cyrus or the Eurosystem Address 80 Kennedy Avenue CY-1076 Ncosa, Cyrus Postal Address P. O. Box CY-1395 Ncosa, Cyrus E-mal ublcatons@centralbank.gov.cy Webste htt:// Fax Paers n the Workng Paer Seres may be downloaded from: htt:// Central Bank of Cyrus, Reroducton s ermtted rovded that the source s acknowledged. 2

3 Measurng Market Power n the Bankng Industry n the Presence of Oortunty Cost Antons Mchs 1 June 2013 Abstract A conectural varatons model s develoed to measure market ower n the bankng ndustry. Unlke revous studes, whch use comlete cost functon secfcatons n the modellng framework, ths study defnes margnal cost based on an oortunty cost, whch s reresented by the nterest rate on mnmum reserves offered by the monetary authortes n a country. Deosts wth the monetary authortes are consdered to be an alternatve use of avalable funds that are usually allocated to loans. The estmates of market ower n the bankng ndustry n Cyrus, usng the roosed model, reect the monooly hyothess. Keywords: market ower, conectural varatons, nterest rates, oortunty cost JEL classfcaton: G21, L11 1 Central Bank of Cyrus. The onons exressed n ths aer are those of the author and do not necessarly reflect the vews of the Central Bank of Cyrus or the Eurosystem. Corresondence: Antons A. Mchs, Statstcs Deartment, Central Bank of Cyrus, 80 Kennedy Avenue, P.O.Box 25529, CY-1395, Ncosa, Cyrus. Emal: AntonsMchs@centralbank.gov.cy 3

4 1. INTRODUCTION Research on the ndustral organsaton of the bankng sector has grown consderably durng the ast two decades as a result of arallel develoments n the ndustral organsaton lterature. The measurement of market ower and the develoment of comettve models for the bankng ndustry usng advances n the New Emrcal Industral Organsaton lterature (NEIO) have artcularly generated mortant nsghts and mroved understandng of several comlcated ssues that are of great mortance to olcy makers and regulators. Intal efforts to model cometton n the bankng ndustry have reled manly on the structure-conduct-erformance (S-C-P) model and the relatonsh between nterest rates and market concentraton. The lmtatons assocated wth ths aroach led to the develoment of more sohstcated methods of modellng cometton, most notably the Panzar-Rosse and conectural varatons aroaches. Research usng these aroaches s stll actve and also addresses several econometrc ssues assocated wth the emrcal measurement of market ower. More recent research has adoted alternatve aroaches that can rovde useful nsghts for comettve bankng markets, such as dscrete choce models and structural models of entry, whch are revewed along wth other methods n Secton 2 of ths artcle. In ths study, a conectural varatons model for loans s develoed, ncororatng oortunty cost n the form of the nterest rate on mnmum reserves offered by the monetary authortes of a country. The model s derved from a roft maxmsaton rogramme, and the eulbrum rce condton s ncororated nto the erceved margnal revenues functon roosed by Bresnahan (1982). Ths functon deends on the oortunty cost, as n Merel (2009), and not on a comlete cost functon secfcaton. Ths fndng s desrable for the bankng sector because most avalable bankng mcrodata (e.g., from fnancal statements) rovde only aggregated cost nformaton that encomasses all the oeratons and roducts offered by banks. The aggregated nature of most avalable cost data ntroduces several measurement roblems n emrcal alcatons and therefore comlcates the valdty of the results. The roosed conectural varatons model s used to estmate the degree of market ower n the bankng ndustry n Cyrus, and the results are comared to two alternatve 4

5 methods roosed n the lterature. The frst of these s based on the conectural varatons model roosed by Coccorese (2005), and the second s based on the Panzar-Rosse revenue test roosed by Dels et al. (2008). All estmates reect the monooly hyothess for the bankng ndustry n Cyrus. The model of Coccorese suggests a comettve eulbrum, the revenue test suggests a market structure close to monoolstc cometton and the roosed conectural varatons model suggests a market structure that s almost erfectly comettve. The remander of ths artcle s organsed as follows. Secton 2 rovdes a detaled lterature revew of bankng market ower models, and Secton 3 develos a conectural varatons model based on oortunty cost. Secton 4 addresses the emrcal alcatons of the study, resents the data and estmates the models usng data from the bankng ndustry n Cyrus. Fnally, Secton 5 summarses the man conclusons of the study. 2. EMPIRICAL MEASUREMENT OF MARKET POWER IN THE BANKING INDUSTRY Ths secton revews several methods of measurng market ower and cometton n the bankng ndustry that have been roosed n the lterature. Cabral (2000,. 6) defnes market ower as the ablty to set the rce of a roduct or servce above the margnal cost, suggestng that the lower the level of cometton faced by a bank n a country, the hgher ts market ower. Ths secton begns wth the S-C-P and effcent markets (EM) models, whch reresent the frst attemts at measurng cometton n the bankng lterature. It then roceeds to descrbe some of the more recent NEIO methods that have been used for the bankng sector, such as the Panzar-Rosse and conectural varatons aroaches to measurng market ower. Some alternatve methods, whch use more recent develoments n the ndustral organsaton lterature, are also revewed. The systematc study of market ower and ts determnants n the bankng ndustry began n arallel wth the develoment of the S-C-P model n the ndustral organsaton lterature. In a bankng context, ths model suggests that changes n the market structure of a geograhcal area (e.g., country or state) affect the way banks structure ther economc olces and conseuently ther erformance. Market structure s usually reresented by a concentraton measure, and the leadng hyothess s that hgher 5

6 concentraton (structure) s assocated wth more market ower for banks (Cabral, 2000,. 157). As a result, banks are able to set hgher nterest rates for loans and lower nterest rates for deosts (conduct), thereby ncreasng ther rofts (erformance). The exstence of market ower ensures that frms can be neffcent wthout beng forced out of the market, as would be the case n a comettve settng. An underlyng assumton of the S- C-P model s that there are barrers to entry n the market (whch generally characterse the bankng ndustry), and few banks are able to rase lendng nterest rates as n a Cournot olgooly settng (see, Heffernan, 2005,. 495). An alternatve model for measurng the nfluence of market ower on nterest rates s the effcent markets (EM) model, and ts assocated effcency hyothess, roosed by Demsetz (1973). Ths model assumes that some banks are more effcent than others (e.g., due to sueror management or technology) and, as a result, are able to offer lower lendng nterest rates and hgher deost rates than comettors. These banks are able to rogressvely ncrease ther market shares (thereby ncreasng concentraton) and rofts. In contrast to the S-C-P model, ths model redcts that lendng nterest rates wll decrease wth ncreasng market concentraton (Heffernan, 2005, 495). Emrcal tests of the two models and ther assocated hyotheses are usually conducted by consderng econometrc models n whch the deendent varables are ether rofts or nterest rates and the ndeendent varables nclude a measure of market concentraton such as the Herfndahl-Hrschman ndex along wth other control varables. The followng s a smle regresson reresentaton of these tests: K t = β 0 + β1 CONC t + β k χ tk + ε t. (1) k =2 r The deendent varable ( r t ) s the nterest rate on loans or the nterest rate offered on deosts, the frst ndeendent varable reresents a measure of market concentraton ( CONC t ) and the model ncludes addtonal control varables ( χ tk ) that can vary across dfferent banks () and erods (t). 6

7 Table 1.1: Exected concentraton coeffcents: S-C-P and EM models Model Concentraton - Lendng nterest rates Concentraton - Deost nterest rates S-C-P Postve Negatve EM Negatve Postve If the deendent varable reresents loan nterest rates, then, accordng to the S-C-P model, the coeffcent of the concentraton varable ( β 1 ) wll be ostve, mlyng that ncreasng market concentraton s assocated wth hgher nterest rates. If the deendent varable reresents deost nterest rates, then the coeffcent wll be negatve. Postve and negatve coeffcents should be exected f the EM model holds. There wll be a negatve coeffcent between concentraton and loan nterest rates and a ostve coeffcent between concentraton and deost nterest rates. These cases are summarsed n Table 1.1. Several emrcal studes have tested these two models n the bankng context, rmarly usng data from the Unted States (US). Berger and Hannan (1989) used the uarterly deost data of 470 banks belongng to dfferent metrooltan areas from 1983 to 1985 to estmate the relatonsh between deost nterest rates and market concentraton n the US. The Herfndahl-Hrschman ndex and the three frm concentraton ratos were used as measures of market concentraton n each metrooltan area. The authors were able to fnd a negatve and statstcally sgnfcant relatonsh between market concentraton and deost nterest rates n the maorty of the cases; ths relatonsh s consstent wth the S-C-P hyothess. The study by Berger and Hannan was crtcsed by Jackson (1992), who dvded the same samle nto three grous based on concentraton level: low, medum and hgh. By estmatng dfferent coeffcents for each grou, the author found a negatve and statstcally sgnfcant coeffcent (at the 1% level) only n the frst grou. In the second grou, the coeffcent had the same sgn but was statstcally nsgnfcant, whereas the coeffcent was ostve and sgnfcant n the thrd grou. These results are more 7

8 consstent wth the EM hyothess because the hgher levels of market concentraton realsed by the ncreasng share of effcent frms are assocated wth hgher deost nterest rates. Berger and Hannan (1992) ublshed a resonse to Jackson that ncluded new results, usng the same market segmentaton aroach adoted by Jackson. They found a negatve relatonsh between concentraton and deost nterest rates at lower levels of concentraton, but they were not able to fnd a statstcally sgnfcant ostve or negatve relatonsh for hgh levels of concentraton. Molyneux and Forbes (1995) estmated a dfferent model n whch the deendent varable referred to rofts nstead of nterest rates. Ther dataset covered the years from 1986 to 1989 and many Euroean countres. The results of ther analyss were consstent wth the S-C-P hyothess, whch redcts a ostve relatonsh between market concentraton and rofts. Berger (1995) used both fnancal erformance (return on euty and return on assets) and nterest rate data as deendent varables to test for the valdty of the S-C-P and EM hyotheses but addtonally ncluded measures of X- effcency and relatve scale effcency n hs analyss to dentfy the drvng factors behnd the EM hyothess f t were to hold. He used an extensve dataset of 4800 US banks from the 1980s. Econometrc estmaton dd not fnd any suort for the S-C-P hyothess and found only weak evdence for the X-effcency hyothess. The scale effcency hyothess was reected. Goldberg and Ra (1996) used the market segmentaton aroach of Jackson (1992) and an econometrc methodology smlar to Berger (1995) to examne the EM hyothess n eleven Euroean countres usng data for the erod from 1988 to Unlke revous studes, ther results rovded evdence for the exstence of scale effcences n markets charactersed by low levels of concentraton, whch suorts the EM hyothess. A smlar study that also controlled for dfferences n effcency among frms was conducted by Berger and Hannan (1997), who found a negatve relatonsh between market concentraton and lendng nterest rates. Berger s modellng framework was also used by Goddard, Molyneux and Wlson (2001) n a study that ncluded data for 15 Euroean countres for the erod from 1980 to 1996 and whose results suorted the S- C-P hyothess. 8

9 Unlke revous emrcal studes, Covoser and Gro (2002) develoed a theoretcal Cournot model to exlan nterest rate margns (the dfference between lendng and money market rates) and roceeded to emrcally test the relatonsh between margns and market concentraton wth an econometrc model. One dstnct characterstc of ther study s that dfferent Herfndahl-Hrschman ndexes were calculated for the dfferent roducts offered by banks (short-term loans, mortgages, tme deosts, etc.). Ther emrcal results for a samle of Euroean countres were consstent wth the S-C-P hyothess n the case of loans and demand deosts (hgher concentraton was found to be assocated wth hgher margns) but not for savngs and tme deosts. Another choce for a deendent varable n testng the S-C-P and EM hyotheses s the Lerner Index, whch s the rce mark-u over margnal cost dvded by the rce ( MC) /. An advantage of usng the Lerner ndex s that t rovdes a more drect measure of market ower that s also free of the accountng measurement roblems (e.g., derecaton and catal valuaton) that can dstort the relatonsh between market concentraton and rce. A dsadvantage of usng the Lerner ndex s the nherent dffculty n accurately measurng margnal cost n ndustres that roduce multle roducts, as s the case n bankng. Ths aroach was followed by Angeln and Cetorell (2003) n a study of 900 Italan banks that covered 14 years of data begnnng n A regresson analyss of the Lerner ndex on the number of banks and the Herfndahl- Hrschman Index dd not fnd evdence n favour of the S-C-P hyothess because ncreases n market concentraton were found to be assocated wth lower values for the Lerner Index and, therefore, wth lower rce mark-us. Ths tye of relatonsh s consstent wth the EM hyothess. A common lmtaton among the studes examnng rce-concentraton relatonshs s the ossble endogenety roblem assocated wth measures of market concentraton (see, Evans et al., 1993), whch can bas econometrc results. Endogenety bas can result ether from the ossble nverse effects of rces on concentraton or from errors n the measurement of both rce and concentraton, whch are dffcult to measure n the bankng ndustry. These roblems led to the develoment of new methods of measurng bankng market ower that are dstnctly charactersed by the formulaton of roft maxmsaton models for banks and the endogenous estmaton of market ower 9

10 along wth the other arameters of the assocated systems of econometrc models. These methods are generally referred to as New Emrcal Industral Organsaton methods (see, Bresnahan, 1989 for a revew), and two basc aroaches have been extensvely used n the bankng lterature: the Panzar-Rosse and conectural varatons methods. The Panzar-Rosse (1987) method measures market ower by examnng how changes n nut rces affect the eulbrum revenues of frms n an ndustry. The senstvty of revenues to changes n nut rces deends on the level of cometton n the market and s calculated as the elastcty of revenues ( R ) wth resect to a vector of nut rces (see, Dck and Hannan, 2010,. 410): H R = k = 1 R W W R. The revenue functon s defned as R = f ( W1,..., W, Z, Y ) and ncludes an exogenous vector Z of cost determnants, an exogenous vector Y of demand determnants and k nut rces W. In the context of a monooly, a change n nut rces (and therefore n margnal costs) wll reduce the revenues of the monoolst; t should therefore be exected that H < 0. In contrast, n a erfectly comettve market, revenues wll R ncrease roortonally wth margnal cost and t should be exected that the Panzar- Rosse statstc wll eual unty: H = 1. R Shaffer (1982) estmated the Panzar-Rosse statstc n a study of New York banks and obtaned a value of 0.318, whch suggests that ths secfc market s nether a monooly nor a erfectly comettve market but somethng between these two extremes. Nathan and Neave (1989) examned a cross secton of Canadan banks and estmated the Panzar-Rosse statstcs for 1983 and They obtaned ostve values of the statstc that were dfferent from unty, whch led them to reect the monooly ower hyothess. Smlar values were also obtaned for Euroean markets. Molyneux et al. (1994) analysed bankng data from Germany, the UK, Italy and San for the erod from 1985 to 1989 and concluded that the market structure n these countres s comatble wth monoolstc cometton. An exceton to the revous fndngs s the study by k 10

11 Molyneux et al. (1996) on the Jaanese bankng market, where the monooly hyothess was not reected. More recently, Claessens and Laeven (2004) estmated the Panzar-Rosse statstc for ffty countres and then roceeded to examne the regulatory and countrysecfc characterstcs that nfluence comettveness. Ther results suggest that countres wth fewer restrctons on entry and a greater resence of foregn banks tend to be assocated wth more ntense cometton. The Panzar-Rosse aroach, although useful, s also charactersed by some mortant lmtatons. Frst, values that are close to unty ( H = 1) should be exected to hold only n the long-run (see, Dck and Hannan, 2010,. 410). In contrast, most studes assume that there s no lag n the adustment of nterest rates (they are assumed to change n arallel wth changes n revenues) and that entry and ext by other frms n the market occurs wthn the same erod. Second, negatve values of the statstc can also arse n cases that are not monoolstc. Heffernan (2005,. 508) descrbes one such case n the context of a erfectly contestable market. Fnally, addtonal lmtatons are related to ossble endogenety ssues n the estmaton rocedure and the dentfcaton of arorate nstrumental varables as well as correctly accountng for all the nut factors n the estmaton of the revenues elastctes. The conectural varatons aroach s based on the emrcal conectural varatons model, roosed by Bresnahan (1982) and Lau (1982), that allows for strategc nteractons among frms (see, Varan, 1992, ). Ths model consders the assumtons (conectures) that a frm mght have concernng the reacton of other frms to ts rce or uantty decsons. For examle, n the case of a uantty ( A ) change decson by bank A, attenton s ad to the artal dervatve B R /, whch uantfes the reacton of any comettve bank B. The frst order condton for roft maxmsaton s more comlex n ths framework relatve to when there are no strategc nteractons among frms ( / = 0 ), and takes the followng form (Dck and Hannan, 2010,. 411): B A A P = MC Q, Z) D ( Q, Y) Qλ. ( 1 11

12 In ths euaton, the rce ( P ) of the roduct s eual to the margnal cost ( MC ) mnus the dervatve of the nverse demand ( D 1) multled by the uantty (Q ) and the conduct arameter ( λ ). The conduct arameter (or conectural dervatve) can be endogenously estmated by a system of euatons that ncludes the demand and cost functons assocated wth a fnancal roduct n addton to the frst order condton mentoned above. The conduct arameter wll have a value of zero ( λ = 0 ) n a erfectly comettve market (because the margnal cost wll be eual to the rce of the roduct) and a value of one ( λ = 1 ) n a monooly. Varyng degrees of merfect cometton wll roduce values between one and zero ( 0 < λ < 1). Ths aroach was used by Shaffer (1989, 1993) to study market ower n the US and Canadan bankng ndustres, reectng the ossblty of colluson n any of the markets. Although ths s a romsng aroach for measurng market ower n bankng, t also resents some lmtatons. Frst, t s senstve to the correct secfcaton of the demand functon. Second, t s senstve to the geograhc defnton of a market, whch can often be mrecse n the bankng ndustry (see, Dck and Hannan, 2010,. 412). Other more recent aroaches to measurng bankng cometton have reled on develoments n the ndustral organsaton lterature. One famly of models that has been used wth romsng results s that of the dscrete choce models wth random coeffcents as roosed by Berry et al. (1995). These models have the advantages of allowng for the structural estmaton of demand wth aggregated data whle addressng endogenety roblems by mosng moment condtons (and not usng dffcult to fnd nstrumental varables). The foundaton of these models s the consumer utlty functon, whch ncororates nteractons between consumer and roduct characterstcs to allow for varyng margnal utltes deendng on the characterstcs of the roduct. The dscrete choce framework was used by Dck (2008) n an effort to estmate a structural model for deosts and to dentfy how the deregulaton of branchng networks n the US durng the nnetes affected consumer welfare. The emrcal results dd not fnd any evdence of a reducton n consumer welfare as a result of the deregulaton. Another alcaton of these models was rovded by Adams et al. (2007), who comared the deost demand of banks and thrft nsttutons and found that they cannot be consdered to be close substtutes. 12

13 Another area of ndustral organsaton alcatons n bankng nvolves structural entry models, such as those used n Bresnahan and Ress (1987). In these models, the emhass s on the relatonsh between rce, the number of frms n the market and the sze of the market necessary to accommodate the entry of new frms. The level of cometton n the market s ndrectly dentfed by estmatng entry thresholds, whch refer to the sze of the market er bank when an addtonal bank enters the market (Dck and Hannan, 2010,. 416). The dea s that when a new bank enters the market, rce cometton wll ncrease and banks wll be exected to realse lower varable rofts as a result. Conseuently, the sze of the market er bank (entry thresholds) must ncrease to rovde the oortunty for varable rofts to cover fxed costs (whch are assumed to reman constant) and allow the new bank to survve n the market. If estmates ndcate that entry thresholds have ncreased, then t can be nferred that rce cometton also has ncreased. Structural entry models have been used n a bankng context by Cetorell (2002) and Cohen and Mazzeo (2007). The former study examned a large samle of US local bankng markets and concluded that the degree of cometton had ncreased n these markets snce the ncrease of entry thresholds to accommodate the entry of new banks. The latter study focused on cometton between dfferent tyes (grous) of nsttutons, such a multle-market banks, sngle-market banks and thrft nsttutons. The results suggest that wthn-grou cometton has ncreased more than between-grou cometton. In a romsng artcle, Boone (2008) ntroduced a new way of measurng cometton that combnes elements of the structural entry and effcency hyothess models. Based on ths measure, cometton can be ntensfed n two ways: (1) through a fall n the barrers to entry that characterse an ndustry and (2) through more aggressve market nteractons between the artcatng frms. In the frst case, lower entry barrers n an ndustry are assocated wth hgher frm artcaton and therefore wth more ntense cometton. In the second case, more aggressve nteractons between frms force neffcent frms to ext the market and market concentraton ncreases as a result. By caturng both effects, ths comettve measure s shown to be more robust than the rce-cost margn, whch s often emloyed n cometton studes. 13

14 To llustrate ths comettve measure, consder three frms wth effcency levels n ** > n * > n, an entry cost functon of γ = γ εζ and a varable roft functon of π ( n, N, I, θ ). Ths functon deends on the effcency level of each frm ( n ), the aggregated effcency ndex ( N ), the number of frms that enter the market ( I ), and a arameter that reresents the aggressveness of frm nteracton n the market (θ ). Based on ths roft functon, Boone (2008) defned the followng relatve roft dfferences metrc: π ( n Φ = π ( n ** *, N, I, θ ) π ( n, N, I, θ ), N, I, θ ) π ( n, N, I, θ ) > 0. Increases n cometton ncrease ths metrc as follows: d Φ / dθ > 0 (ncreases n frm nteracton ncrease cometton) and d Φ / dε > 0 (decreases n entry costs ncrease cometton). In emrcal alcatons, the roft elastcty rovdes a way of caturng both effects and can be estmated from log-lnear models as follows (see, Degryse et al. 2009,. 36): logπ = a + β log. c In ths model, the rce elastcty ( β ) s estmated as the coeffcent of the margnal cost varable ( c ). More effcent frms wth lower margnal cost wll be exected to realze hgher rofts ( π ) and therefore β < 0. Conseuently, hgher absolute values of ths negatve coeffcent wll be assocated wth stronger bank cometton. Van Leuvensten et al. (2007) used the Boone ndcator to measure bankng cometton n several countres and the results suggested consderable varablty among countres. As suggested by Degryse et al. (2009), more research s needed to evaluate the usefulness of the Boone ndcator. 14

15 3. A CONJECTURAL VARIATIONS MODEL WITH OPPORTUNITY COST In ths secton, a conectural varatons model that ncororates oortunty cost n the form of nterest ncome forgone by the rovson of loans s develoed. Ths s a reasonable assumton n bankng markets because the total deosts receved by banks n a erod are usually allocated nto two channels. The frst s the rovson of loans to economc agents n the market and the second s the lacement of deosts wth the monetary authortes of the country. At a mnmum, these deosts should be eual to the mnmum (or reured) reserves reuested by the monetary authortes. As emhassed by Mshkn and Eakns (1998,.325), banks also hold addtonal amounts of reserves wth the monetary authortes because these deosts (called excess reserves) are consdered to be the most lud assets n a bank s balance sheet. In ths way, these deosts are readly avalable to meet unexected oblgatons that may arse, esecally n case of large wthdrawals. The dea of ncororatng oortunty cost n models that examne cometton n bankng was also roosed by Heffernan (2002), who used the Lbor rate n generalsed rcng econometrc models. In these models, the Lbor rate s used as a roxy for the deost rate forgone by a bank when t rovdes an addtonal monetary unt of loans nstead of wthholdng t as art of ts deosts n another nsttuton. In ths study an alternatve oortunty cost concet s used: the nterest rate that s offered by monetary authortes on mnmum reserve deosts (see, Heffernan, 2005,.290). In Euroe, banks n countres that are members of the Eurosystem (as s the case for Cyrus) are reured to kee a mnmum reserves rato as defned by the Euroean Central Bank. Ths rato s calculated based on ther short-term labltes and banks earn nterest on these reserves that s eual to the average rate of the weekly tenders over the mantenance erod (Matthews and Thomson, 2008,. 54). The nterest rate on mnmum reserves s used n a conectural varatons model that ncororates comettve resonses. Mérel (2009) used a smlar modellng framework for a commodtes market by ncororatng the oortunty cost of the revenue forgone by not roducng a substtute roduct n the context of a roft maxmsaton rogramme. However, hs study dd not consder comettve resonses, as s common n conectural varatons models, by estmatng the relevant cross-rce elastctes. Ths secton 15

16 develos one such model for bankng that also ncororates an oortunty cost measure n the form of the nterest rate on mnmum reserves. An advantage of the roosed model s that t does not reure a comlete cost functon secfcaton to estmate the market ower arameter n a system of euatons. Ths s a freuently encountered roblem n emrcal studes because fnancal statements and other bankng ndustry data sources do not searate exenses by roduct or servce category (e.g., oeratng cost attrbuted to loans and deosts). As a result, t s dffcult to accurately measure bank outut (see, for examle, Heffernan, 2005,. 476), whch ntroduces several roblems for econometrc estmaton. Consder the followng roft maxmsaton rogramme for bank : maxπ = + k ( D ) C (, ω 1, ω2, ω3, z, D )., The total loan revenues are eual to the uantty demanded = f,, χ ) multled ( by the rce charged for the fnancal roduct. The uantty demanded s also a functon of a rce ndex for cometton (e.g., comettors weghted average nterest rate) and a vector of exogenous factors χ. The total loans rovded by the bank are derved from the deosts t receves ( D = s + ) after subtractng the amount laced by the bank as an nterest-bearng deost (eual to or n excess of the mnmum reserves) wth the monetary authortes ( s ). Smlarly, the cost functon C (, w, z, D ) deends on several factors: the loan outut ; nut factors such as nterest ad on deosts ( ω ), labour wages ( ω ) and the cost of hyscal catal ( ω ); and the deosts 1 2 held by other nsttutons ( D ), whch also nvolve a managng cost. The deosts at other nsttutons are assocated wth an nterest rate ( 16 3 k ) that also contrbutes to roftablty. Ths wll consttute the oortunty cost concet of the analyss and wll be measured by the nterest rate on mnmum reserves. Ths roft maxmsaton rogramme s smlar to that used n Coccorese (2005), whch also estmated a conectural varatons model. However, ths model ncludes the oortunty cost varable, whch further comlcates the frst-order roft maxmsaton condtons.

17 17 The artal dervatve of the roft functon wth resect to rce s as follows: C C k k + + = Π. Settng ths exresson eual to zero and lettng MC C = rovdes the frst order condton [ ] 0 = + + MC k. Ths can be wrtten n a rce-cost margn format as MC k + =. To obtan an eulbrum condton n terms of own and cross-rce elastctes, as s common n conectural varatons models, both sdes of the above euaton are multled by /. In addton, the second term n the denomnator of the rght sde s multled by 1 / =. The euaton then becomes MC k + = 1. When wrtten n terms of elastctes, the euaton s eual to

18 k MC =. (1) ε + ε n The own-rce elastcty s ε =, the cross-rce elastcty s ε = and the comettve rce-reacton elastcty s η =. Ths elastcty shows how rce changes by bank generate a comettve rce reacton by comettor (whch also nfluences the market rce, as s common n conectural varatons models). In Coccorese (2005), the eulbrum condton (1) does not nclude the nterest rate on mnmum reserves on the left sde and only two elastctes are ncluded on the rght sde. Instead of the elastcty η, a conectural varaton arameter λ = / s defned and s multled by /. However, estmaton of ths arameter n the context of euaton (1) reures a system of euatons wth a cost functon secfcaton, an aroach not adoted by ths study due to the roblems assocated wth cost measurement n bankng. Leavng only margnal cost on the left sde and rearrangng rovdes the followng eulbrum condton: MC = k. (2) ε + ε η To ntroduce a market ower arameter n the modellng framework, a erceved margnal revenue functon s defned, as roosed by Bresnahan (1982) and used by Shaffer (1989) and Mérel (2009) n the bankng and commodtes markets, resectvely. The erceved margnal revenue functon s PMR = ( 1 h) + h MR. (3) Euaton (3) ncludes the market ower arameter h (also referred to as the conduct arameter n the lterature). Values of ths arameter that are close to zero ( h = 0) ndcate 18

19 a comettve ndustry whle values that are close to one ndcate monooly ower ( h = 1 ). Accordngly, ntermedate values reresent eulbra between these two extremes. Because roft maxmsaton n eulbrum (see, Peall et al., 2008,. 247) mles eualty between margnal revenue and margnal cost ( MR = MC ), exresson (2) can be substtuted nto (3): PMR = h h h k. ε + ε η By cancellng out eual terms, the euaton becomes PMR = h h k 1 + ε + ε η. (4) Another exresson for margnal revenues can be obtaned by takng the artal dervatve of the roft functon wth resect to the uantty demanded: Π C (,.) C (., D ) = + k +. The frst artal dervatve of the cost functon C (,.) refers to the frst uantty exresson n the cost functon C, ω, ω, ω, z, D ) and, as s the case of the ( second artal dervatve, whch refers to the deosts wth the monetary authortes D that are also ncluded n the cost functon. The frst two terms n the above frst order condton are eual to the margnal revenue MR = + (see, for examle Cabral, 2000,. 25). Euatng the frst order condton to zero and substtutng for the margnal revenues exresson gves 19

20 MR = k + μ. (5) C (,.) C (., D ) The term μ = n euaton (5) s the dfference between the two artal dervatves of the cost functon, and accordng to Mérel (2009), reresents the dfference n margnal cost assocated wth the two alternatve uses of nuts. In the context of ths study, the deosts collected by bank are drected to two alternatve uses: (1) for the roducton of loans and (2) as deosts held wth the monetary authortes. Conseuently, accordng to euaton (5), n eulbrum, the margnal revenue from roducng an addtonal unt of loans s eual to the margnal revenue forgone f the unt was deosted wth the monetary authortes (the oortunty cost k ) lus the dfference n margnal cost ( μ ) between roducng an addtonal unt of loans nstead of managng an addtonal unt of deosts. By euatng the PMR n (4) wth the MR n (5), the followng rce euaton s obtaned, whch s to be used n a nonlnear system of euatons to estmate the market ower arameter: h 1 + = h k + + μ ε ε η k. (6) + In addton to euaton (6), the nonlnear system of euatons wll also nclude a demand functon for bank based on the revously defned secfcaton for the uantty demanded = f,, χ ). Followng Coccorese (2005), a log lnear model wll be used as follows: ( log = a0 + a1 log + a2 log + a3 logy + a4 log log + ε. (7) However, n addton to the nteracton term ( ln ln ), whch s to be used as an elastcty shfter n the estmaton rocedure, euaton (7) also ncludes the exogenous varable Y that reresents the level of ncome n the economy. Bresnahan (1982) 20

21 emhasses that elastcty shfters are necessary to change (rotate and shft) the demand sloe n such a way that dstngushes between the hyotheses of cometton and monooly. Rotaton s acheved through the new exogenous varable (comettors rce ndex) n the demand euaton, whch enters nteractvely wth the rce varable, whereas shfts n the demand functon are acheved through the ncome varable Y. Jont movements n the two exogenous varables enable the estmaton of the degree of market ower. In a erfectly comettve market, the rotatons wll not change the eulbrum rce, but the ooste s true f market ower exsts, as n the case n a monooly. Euatons (6) and (7) can be estmated as a nonlnear system of euatons. However, unlke the studes by Coccorese (2005) and Mérel (2009), the eulbrum rce condton (6) addtonally ncludes the comettve rce elastcty η = that catures the rce reacton of cometton when bank changes the rce of ts roduct. To estmate ths elastcty, an addtonal euaton wll be ncluded n the system of euatons, n log-lnear format, usng the comettors rce ndex as the deended varable: log γ 0 + γ 1 log + γ 2 = log EURIB. (8) t The comettve rce elastcty wll be estmated usng arameter γ 1. Euaton (8) also ncludes the Eurbor rate ( EURIB ) as an ndeendent varable, followng the analyss of t Heffernan (2002), whch ncluded the Lbor rate n econometrc models for lendng nterest rates n the UK. It s to be used as a roxy for the term k + μ n euaton (6), whch s the nterest rate on mnmum reserves lus the dfference between the margnal cost of roducng an addtonal unt of loans and the margnal cost of managng an addtonal unt of deosts laced wth the monetary authortes. 4. ECONOMETRIC ANALYSIS 21

22 4.1. Data and relmnary analyss accordng to the S-C-P model The emrcal testng of the modellng framework roosed n the revous secton was conducted usng data from the bankng ndustry n Cyrus. The data were collected from the fnancal statements of twenty fnancal nsttutons that offer bankng servces n the country. Commercal banks can be searated n two grous: (1) those wth an ultmate controllng afflate n Cyrus and (2) those wth an ultmate controllng afflate abroad. In addton to tradtonal banks, whch rmarly offer retal, cororate and nvestment bankng servces, the samle ncludes cooeratve socetes (COOPs), whch have develoed smlar roducts durng the last decade and can be consdered drect comettors of tradtonal banks n Cyrus. For smlcty, n the remander of the analyss, all fnancal nsttutons n the samle wll be referred to as banks. Table 1.2 rovdes nformaton on the lendng market shares of the to sx banks n Cyrus, whch account for 80% of the market. Two maor commercal banks n the market have market shares rangng from 16% to 24%. The total market share of COOPs ranges from 19% to 20%. Most COOPs are local n character, and, accordng to ther consttutons, are allowed to rovde servces only wthn the boundares of the muncalty n whch they are establshed. However, they have a central authorty (the Cooeratve Central Bank), whch takes over ther economc and fnancal olces towards cometton and admnsters a common network of automatc teller machnes (ATMs). For examle, the ATM network of a artcular COOP can be used by the subscrbers of any other COOP rresectve of geograhcal locaton. Table 1.2: Lendng shares of maor fnancal nsttutons n Cyrus Fnancal Insttuton Dec 2008 Dec 2009 Dec 2010 Dec 2011 Bank of Cyrus Ltd 24% 24% 24% 24% Cooeratve Socetes 19% 19% 20% 20% Marfn Poular Bank Publc Co Ltd 16% 16% 17% 16% Hellenc Bank Ltd 7% 7% 7% 7% Alha Bank Cyrus Ltd 8% 8% 8% 7% Eurobank EFG Cyrus Ltd 2% 5% 4% 3% Total 76% 79% 80% 77% Source: Central Bank of Cyrus 22

23 To estmate market ower n the bankng ndustry of Cyrus based on the modellng method outlned n Secton 3, several varables were collected or constructed from the fnancal statements of the banks used n the samle: the total value of loans for each bank n year t ( Q t ); the nterest rate on loans for each bank n year t ( P t ), obtaned by dvdng total nterest ncome from loans by total loans, as suggested by Coccorese (2005); the (loan value based) weghted average lendng nterest rate of the comettors of bank n year t ( CP ); and the total loans ncluded n the assets of the comettors of bank n year t ( CQ ). t t Table 1.3: Summary statstcs Varable Samle Mean St dev Mn Max Loans (mllon EUR) Interest rate on loans (%) Comettors nter. rate on loans (%) Interest rate on deosts (%) Cost of hyscal catal (EUR) Eurbor (%) Interest rate on mnmum reserves (%) Gross domestc roduct (mllon EUR) Total cost (mllon EUR) Average cost (EUR) ,03 0,214 Labour cost (EUR) Total Revenues (mllon EUR) Deosts to total assets rato (%) Assets (mllon EUR) Herfndahl-Hrschman Index (%) Banks wth ultmate controllng afflate n Cyrus: 4 Banks wth ultmate controllng afflate abroad: 5 Cooeratve socetes: 11 Unbalanced anel for the erod: For comarson uroses, measures of market ower n the bankng ndustry of Cyrus were also obtaned by usng two addtonal models. The frst s the conectural varatons model roosed by Coccorese (2005). Because ths method reures the full 23

24 secfcaton of a cost functon, addtonal cost nformaton was collected from fnancal statements: the total cost of each bank n year t ( C ); the average cost of each bank n year t ( AC t ), calculated by dvdng total cost by the total value of loans; a roxy varable for the er unt cost of hyscal catal of bank n year t ( ω ), calculated by dvdng oeratng cost (excludng nterest exenses and labour costs) by funds under management, as n Coccorese (2005); the average labour cost of each bank n year t ( ω 2t ), calculated by dvdng total labour exenses (salares, socal securty and enson fund contrbutons lus any other staff exenses) by the total number of emloyees; and the nterest rate ad on deosts ( ω ), calculated by dvdng total nterest exenses on 3t deosts by the total amount of deosts reorted by each bank n the fnancal statements of year t. The second model consdered n Secton 4 s based on a revenue model roosed by Dels et al. (2008) to estmate the Panzar-Rosse statstc, whch addtonally ncludes the followng varables: total assets ( t t 1t Assets t ); the rato of total deosts to total assets for each bank and year t ( Deosts / Assets ) and total revenues ( TR t ); Addtonal varables obtaned from other sources ncluded the nterest rate on mnmum reserves n year t ( R ), calculated from data n the annual reorts of the Central t Bank of Cyrus by dvdng the nterest exenses of the Central Bank by the total deosts laced wth the Central Bank by banks; the gross domestc roduct of the country n year t at current market rces ( Y t ), ublshed by the Cyrus Statstcal Servce; the Herfndahl-Hrschman ndex of market concentraton ( HHI ) for the bankng ndustry of Cyrus, ublshed by the Euroean Central Bank (ECB), based on the assets of the banks oeratng n the country from 1999 to 2010; and the Eurbor rate (euro nterbank offered rate) n year t ( EURIB ), ublshed by the Euroean Bankng Federaton and uon whch t banks n the Euroean Unon (EU) exchange funds. The fnancal statements used n the dataset covered the erod from 1999 to 2011; however, the data were not avalable for all the erods n the analyss for several banks. In several cases, ths was because some banks began oeratng n Cyrus after The avalable annual observatons for each t t 24

25 bank are ncluded n Table 1.3 along wth summary statstcs for the varables used n the analyss. To nvestgate the relatonsh between lendng nterest rates and market concentraton accordng to the S-C-P model, a log-lnear fxed effects model was estmated as follows: log P t = β 0 + δ d = 1 + β logt + ε β logcp 1 t + β log EURIB 2 t + β log HHI 3 t + β logy 4 t (9) The lendng nterest rate of bank n year t s a functon of comettve nterest rates, the Eurbor rate, whch s used as a roxy for the average comettve nterest rate n the EU, the Herfndahl-Hrschman ndex of market concentraton, the gross domestc roduct, and a tme trend varable (T ). In addton, the model ncludes ndcator varables for each bank n the samle to be used as fxed effects. Table 1.4: Log-lnear nterest rate model Coeffcent Estmate St error t test Intercet * log HHI * log Y * log EURIB ** log CP * log T R suared: Samle sze: 206 Sgnfcance level: *5%; **10% The coeffcent estmates from the OLS method are ncluded n Table 1.4. All the coeffcents are statstcally sgnfcant at the 5% or 10% levels and have the exected sgn excet for the coeffcent of the trend varable. The coeffcent for the Herfndahl- Hrschman ndex of market concentraton s ostve and suggests that an ncrease n market concentraton of an average of 1% s assocated wth a 0.565% ncrease n lendng 25

26 nterest rates. Ths fndng s consstent wth the S-C-P model, accordng to whch ncreasng market concentraton s assocated wth ncreasng market ower, resultng n hgher rces (lendng nterest rates) and hgher rofts. Conversely, ncreasng cometton n the market (decreasng market concentraton) s assocated wth declnng nterest rates. 16% 14% 12% 10% 8% 6% 4% 2% 0% Source: Euroean Central Bank Fgure 1.1: Herfndahl-Hrschman ndex for the bankng sector n Cyrus (assets) Fgure 1.1 exhbts the trend of the Herfndahl-Hrschman ndex durng the erod from 1999 to The overall market concentraton level s low and has not changed sgnfcantly durng the decade covered by the avalable data, remanng mostly around the 10% level. Small ncreases are observed after 2005, artcularly n 2007 and Ths analyss, smlarly to most emrcal S-C-P tests, cannot offer clear conclusons wth regards to the level of cometton n the ndustry. It s therefore necessary to estmate a market ower arameter as exlaned n Secton Estmaton of the conectural varatons model To measure market ower n the bankng ndustry of Cyrus, ths secton roceeds wth the estmaton of the nonlnear system consstng of euatons (6)-(8). Usng the varable 26

27 notaton ntroduced n Subsecton 4.1, the system (model 1) to be estmated can be reresented as follows: log Q t = a 0 + a 6 + a log P 1 Foregn t t + a + a 7 2 logcp t logt + ε + a logy Q 3 t + a 4 log P logcp t t + a Coo 5 t P t = β + hr 0 t h ε P t + β EURIBt ε η β logt + ε 2 P (10) log CP t = γ + γ log P 0 1 t + γ log EURIB 2 t + γ logcq 3 t + γ logt + ε 4 CP The system s subect to the followng arameter restrctons: a = ε a = ε = γ 1, 2, 1 n and 0 < h < 1, and all nterest rate varables have been deflated as suggested by Coccorese (2005). The frst euaton s smlar to euaton (7) excet that t ncludes a tme trend varable (T ) and two ndcator varables n the lst of exlanatory varables. The frst ndcator varable takes the value of 1 when the observatons n the samle refer to a cooeratve socety ( Coo ). Smlarly, the second ndcator varable takes the value of 1 when the observatons n the samle refer to a bank wth a foregn ultmate controllng afflate ( Foregn ). t Based on these secfcatons, the control grou, whch refers to banks wth an ultmate controllng afflate n Cyrus, s catured by the ntercet. Because we are usng a anel dataset n the estmaton rocedure, t s ossble to use searate ndcator varables (fxed effects) for each bank n the samle. However, due to the lmted number of observatons and the unbalanced nature of the anel dataset, t was not ossble to obtan ndvdual coeffcent estmates for all the banks n the samle because the generalsed method of moments (GMM) estmator dd not converge. The classfcaton based on the control status of the banks s sensble because the banks wthn each grou have hstorcally been smlar n terms of sze and strategc orentaton. In the second euaton, the Eurbor rate has been ncluded as a roxy for the term k + μ n euaton (6), whch s eual to margnal revenues. It conssts of the nterest 27

28 rates on mnmum reserves ( k ) and the dfference n margnal cost ( μ ) assocated wth the two alternatve uses of deosts (rovson of loans or deosts at the Central Bank). In a comettve market, the margnal revenues of a bank should reflect movements n the Eurbor rate, as suggested by Heffernan (2002) for the relatonsh between lendng rates and the Lbor rate n the UK economy. The thrd euaton has been ncluded n the system n log-lnear form to estmate the comettve rce elastcty η = / )( / ) that enters the eulbrum rce ( condton n the second euaton. The weghted average ndex of the comettve nterest rates s consdered to be a functon of four ndeendent varables: the lendng rate offered by bank, whch s not ncluded n the calculaton of the ndex (for each cross secton of observatons n the anel dataset); the Eurbor rate, whch s agan used as a roxy for the comettve nterest rate n the EU; and the total loans offered by all banks ncluded n the calculaton of the ndex excet for bank. The euaton also ncludes a trend varable. The terms ε Q, ε P and ε CP reresent the errors n the system of euatons. It s ossble that the CQ t varable s correlated wth the error term of the model n the thrd euaton because a general declne n the total loans n the market mght generate a reducton n the lendng nterest rates charged by banks to recover. However, because the total amount of loans comrses data collected from nneteen dfferent banks, t s unlkely that t wll exhbt varaton that reresents a common olcy reacton by all banks. To test for the ossblty of endogenety bas n the results, a Haussman endogenety test (see Cameron and Trved, 2005,.271) was constructed and a second verson of the model (model 2) was estmated wthout the use of ths varable. In addton to the exogenous varables ncluded n each euaton, the er unt cost of hyscal catal was used as an nstrumental varable for CQ t n the estmaton of model 1 and was also used as an addtonal nstrumental varable n the estmaton of model 2. Ths varable s exected to be correlated wth the total value of loans n the market because t s one of the nut factors used by banks n the roducton rocess. 28

29 Table 1.5: GMM estmaton of nonlnear conectural varatons models Model 1 Model 2 Coeffcents Estmate St error t test Estmate St error t test Euaton 1: Demand for loans Intercet * * log P * * log CP * * log Y * * Coo * * Foregn bank * * log P log CP log T R suare Euaton 2: Interest rate Intercet ** * log Eurbor * * log T * * R suare Euaton 3: Comettors' nterest rate ndex Intecet * * log P * * log CQ * log Eurbor * * log T * h (market ower) * * R suare Ch-suare test Samle sze:216 Sgnfcance level: *5%, **10% The coeffcent estmates for the two versons of the model usng the GMM method are ncluded n the Table 1.5. All coeffcents have the exected sgn and are statstcally sgnfcant at the 5% or 10% levels, excet the coeffcents for the 29 CQ t and EURIB t varables n the thrd euaton. However, because the varables n ths model concern average movements n 19 dfferent banks, the sgns of the coeffcents cannot be known wth certanty. The demand for loans s elastc, and the elastctes that enter the rce

30 euaton have the followng values: ε = , ε = and η = The market ower arameter ( h ) s statstcally sgnfcant at the 5% level n both models and s close to zero. Based on these values, the bankng ndustry of Cyrus can be consdered to have a nearly erfectly comettve market structure, and the monooly hyothess s reected. Based on the value of the Ch-suare test for over-dentfyng restrctons, the valdty of the addtonal nstrument (the er unt cost of hyscal catal) used n the GMM estmaton rocedure cannot be reected at the 5% sgnfcance level n ether of the two models (the assocated -values of the tests n models 1 and 2 were and 0.227, resectvely). Fnally, a Hausman endogenety test for the CQ t varable n model 1 was constructed based on the euvalence of the OLS and 2SLS estmators under the null hyothess that the two estmators are asymtotcally euvalent and that there s no endogenety bas. The estmates rovded a value of 6.75 wth a corresondng -value of 0.748, accordng to whch the eualty of the two estmators s acceted at both the 5% and 10% sgnfcance levels Results wth alternatve models As already noted n Secton 2, most emrcal studes of market ower n bankng have concentrated on the conectural varatons and Panzar-Rosse aroaches. To comare the results of the revous subsecton wth alternatve models that have been roosed n the bankng lterature, two further models were estmated and comared wth model (10) usng smlar varable defntons as n the revous secton. The frst s the conectural varatons model roosed by Coccorese (2005), who estmated a nonlnear system of three euatons usng a translog cost functon as shown below (model 3). The frst euaton n ths system reresents loan demand and s smlar to the frst euaton n system (10) n whch the nterest rate varables n the system have been deflated. The translog-cost euaton ncludes three nut factors: the nterest rate ad on deosts ω t = DRt, the average rce of labour ω t = LBt and the er unt cost of hyscal catal ω 1 3t = PC t 2. To dentfy arameter J (the conectural dervatve), fve restrctons that are consstent wth lnear homogenety n nut rces were ncororated 30