Economies of Scale in the Banking Industry: The Effects of Loan Specialization

Transcription

1 Economes of Scale n the Bankng Industry: The Effects of Loan Specalzaton Y-Ka Chen Department of Busness Admnstraton and Educaton School of Busness Empora State Unversty Empora, KS E-mal: chenyka@empora.edu Joseph R. Mason Department of Fnance LeBow College of Busness Drexel Unversty Phladelpha, PA E-mal: joseph.r.mason@drexel.edu Erc J. Hggns Department of Fnance College of Busness Kansas State Unversty Manhattan, KS E-mal: ehggns@ksu.edu Draft, August 2001 Contact person: Y-Ka Chen Department of Busness Admnstraton and Educaton School of Busness - Box 4058 Empora State Unversty 1200 Commercal Street Empora, KS E-mal: chenyka@empora.edu Phone: Fax:

2 1 Economes of Scale n the Bankng Industry: The Effects of Loan Specalzaton Abstract Popular belef holds that bankng consoldaton wll result n a small number of large monopoly banks that can acheve substantal economes of scale and outperform small compettors. But research suggests that small banks are sometmes more effcent than ther large counterparts. One compettve strategy undertaken by numerous small banks s agrcultural lendng. Agrcultural lendng s thought to beneft from mcro-level montorng to gauge whether borrowers mplement scentfc farmng practces and exert requste care to maxmze yeld and repay bank loans. Ths essay uses county-level data from 1992 Census of Agrculture to classfy agrcultural banks and evaluate ther lendng actvtes n terms of cost effcency. Agrcultural producton, product-mx, crop yelds and product prces are key factors affectng the performance of agrcultural lendng. They are used to explan dspartes n bank effcency. Agrcultural lendng s found to be postvely correlated to bank effcency. The results also show that t s not necessary to be large to be effcent for the banks wth loan specalzaton. Nevertheless, both rural and urban banks tend to be susceptble to prce volatlty n the agrcultural sector.

3 2 I. Introducton Before nterstate bankng deregulaton, there were numerous small unt banks n the Unted States. After deregulaton, the number of banks decreased dramatcally (See Fgure 1). From 1988 to 1997, the total number of banks decreased 29%, whle the number of commercal banks decreased 32.6% (Fgure 1 and Table 1). Acceleraton of mergers and acqustons seems to be the man culprt for the declne. In terms of sze, large and medum sze banks actually grew n number, whle the number of small-szed banks declned by 46%. Nevertheless, there are stll 5818 small banks left, much more than the number of large and medum banks. Lookng at the data n Table 5, t appears that large banks are concentrated n metropoltan statstcal areas (MSAs), accountng for 9.71% of the number of banks, whle small banks congregate n rural areas, accountng for 42.89% of the number of banks. Do rural small banks have an nherent advantage assocated wth ther specalty and locaton over large metropoltan banks? As fnancal ntermedares, banks mtgate the problem of nformaton asymmetry between lenders and borrowers. Tmely and relable customer nformaton s essental for successful banks. Good nformaton n turn requres close relatons wth customers. It s alleged that small local banks dealng wth small customers are n a better poston to develop and mantan close customer contacts than large banks. Small customers, wthout establshed credt, also tend to be more loyal to small banks than bg banks. If small banks ndeed have a comparatve advantage n relatonshp bankng to large banks, ths may be key to ther contnued exstence. Relatonshp bankng, however, may sometmes lead to moral hazards. Large banks tend to adhere rgdly to rules and crtera n loan revew whle loan offcers n small banks may have some dscreton that they may msuse and thereby compromse loan qualty. Ths study tres to fnd whether small banks have a comparatve advantage n relatonshp bankng by comparng ther effcency wth that of larger banks. In agrculture, great uncertanty and gyratons n yelds and prces requre a more n-depth understandng of customer stuatons, local envronments, government polcy, and fluctuatons n agrcultural exports n order for sound lendng to occur. Examnng the

4 3 effcency of agrcultural banks could shed lght on the future vablty of small banks n copng wth ncreasng competton from consoldated super banks. Also, examnng agrcultural banks gves us some nsght on the effect of specalzaton on a bank s effcency and, hence, ts ablty to survve. In contrast to earler studes, the effcency of small, rural, agrcultural banks s studed by smultaneously examnng the effect of sze, locaton and lendng specalty. In addton, the effect of the local agrcultural envronment on bank effcency s studed n detal to see f the degree of specalzaton requred n rural agrcultural areas affects bankng effcency. Fgure 2 and Fgure 3 show the number of specalty banks and managed assets of specalty banks n the Unted States n Most commercal banks are classfed as general commercal banks. There are no specfc target customers for those banks. However, there are stll sgnfcant numbers of banks wth loan specaltes n the bankng ndustry. Fgure 2 shows that 14.31% of U.S. commercal banks are classfed as agrcultural banks. Although the number of mortgage banks, wth resdental, nonresdental, and redevelopment loans specaltes, s the second largest n terms of the number of banks n the commercal bankng ndustry n 1996, the nformaton asymmetry n the mortgage market s low. Banks wth specaltes n agrcultural lendng are specalty banks wth hgh nformaton asymmetry. Although Fgure 3 shows agrcultural lendng s only 1.08% of all lendng n the specaltes banks, the number of agrcultural banks s stll hgh n the bankng ndustry. Therefore, agrcultural banks, whch have hgh nformaton asymmetry, are the target banks examned n ths study. Because of the hgh nformaton asymmetry n agrcultural banks, relatonshp bankng mght be one of the mportant ssues of bank operatng effcency. Generally, economes of scale do exst n the bankng ndustry. However, n banks wth loan specalzaton, ths mght not be the case. In term of operatng cost effcency, specalty banks wth superor nsde nformaton mght not need to be large n order to be cost effcent. Thus, one of the objectves of ths study s to examne whether economes of scale exst n the bankng ndustry wth loan specalzaton. Due to the hgh nformaton

5 4 asymmetry n the agrcultural lendng sector and the relatvely hgh percentage of the agrcultural banks n the U.S. bankng ndustry (Fgure 2), agrcultural banks are examned n ths study. Whether the smaller banks wth agrcultural loan specalzaton have survval value under the trend of bank consoldaton s an mportant ssue that we examne n ths study. Banks specalzng n agrcultural lendng mght be nfluenced by local agrcultural producton. Thus, the fluctuaton of agrcultural producton and prce mght mpact bank operatng effcency. In order to capture the mpact of local economc actvtes on bank effcency, agrcultural factors at the county level are used to proxy for local economc condtons and are examned to determne whether there s a scale of economy effect n banks wth loan specalzatons n agrculture. II. Lterature Revew Bank performance s found to vary wth sze, tme, locaton, loan portfolo-mx and locaton. Small banks appear best at lendng to small local busness. Small banks are better at relatonshp bankng than large banks due to superor nformaton and greater dscreton n applyng nformaton. Nakamura (1994) and Udell (1989) concluded that loan offcers at large banks tend to follow bank rules and crtera more rgdly n loan revew than ther counterparts at small banks. 1 Furthermore, Brckley, Lnck, and Smth (2000) found that small local owned banks have a comparatve advantage over branch 1 Nakamura (1994) ndcates that small busnesses are defned as busnesses that have less than $10 mllon n annual recepts and borrow less than $3 mllon from all sources.

6 5 banks of large banks n some envronments. 2 survve under the trend of bank consoldaton. Thus, banks wth smaller szes mght Neely and Wheelock (1997) and Zmmerman (1996) showed that tme plays an mportant role n bank effcency. As the busness envronment vares from regon to regon, Neely and Wheelock and Zmmerman ndcate that local economc factors affect the performance of local banks sgnfcantly. Bank locaton s a strong factor n determnng loan portfolo-mx. For example, farm loans are lkely to be made by rural banks due to ther proxmty to farmers. Levonan (1996) showed that banks headquartered n metropoltan areas, even through they have branches n agrcultural areas, are less lkely to engage n agrcultural lendng. The study of the effect on bank effcency of structural and envronmental factors by Neff, Dxon, and Zhu (1994) found that a hgher agrcultural loan rato reduces cost neffcency but ncreases proft neffcency. Large banks n metropoltan areas, through ther many local branches, can reap economes of scale. The exstence of other large rval banks n an urban area fosters competton and drves banks to operate more effcently. Weber and Devaney (1998) ndcated that local banks n rural areas usually make only local loans and tend to be less techncally effcent due to small-scale operatons. After the deregulaton of nterstate branchng n 1997, Glbert (1997, 2000) asserts that large banks wth more resources are lkely to expand nto rural areas leadng to greater competton and effcency. However, there s another vew by Glbert and Belonga (1988) that argues that large banks, beng able to dversfy, may be less nclned to nvest n agrcultural loans. The entry of large banks nto rural areas, therefore, may actually lower credt avalablty n rural areas. 2 Ths concluson s based on a county-by-county comparson of two sets of banks operatng n Texas. One set comprses small locally chartered banks. The second set comprses branches of large banks that hold out-of-state charters.

7 6 Jayartne and Strahan (1996) found the lberalzaton of branch restrctons stmulates the rural economy by ncreasng competton. Featherstone (1996) observed that small rural banks are lkely to be acqured by large banks specalzng n agrcultural loans. As a result, mergers and acqustons do not affect the rural economy. Although large banks n the future are lkely to domnate rural areas, recent changes n bankng regulaton are favorable to small banks n reducng ther regulatory burden. They are now allowed to expand nto new busnesses. Glbert (1997) showed evdence that competton from new entrants of large banks would also compel small banks n the rural areas to operate more effcently. As small rural banks specalzng n agrcultural loans had not been the prmary targets of nterstate mergers and acqustons, Neff and Ellnger (1996) argued that those small rural agrcultural banks can contnue to retan the old agrcultural busness as well as expandng nto consumer loans. 3 The above-cted studes suggest that small banks mght have comparatve advantages, especally n the rural fnancal markets. Because of the proftablty of small banks n the local fnancal markets, large banks are also attracted to enter those markets. The compettveness of the local fnancal market wll ncrease. Under these crcumstances, operatng effcency wll be an mportant ssue for the survval of small banks. If small banks take advantage of ther superor nformaton n the local market and ther ncreased flexblty, small banks may operate more effcently than large banks. However, small banks are ted more closely to local economy. Thus, small banks operatng n rural envronments may be senstve to agrcultural producton and prces wthn ther local area. Fgure 2 shows the managed assets of specalty loans of specalty banks. Loan specalzaton and local economc condtons are examned n ths study. Therefore, ths study nvestgates whether sze matters to bank operatng effcency and survval value. The results wll provde evdence on whether small banks wll be vctms 3 Belonga and Glbert (1990) ndcated that agrcultural bankng s proftable n general. They fnd that agrcultural banks performed effcently n agrcultural lendng, except durng agrcultural shocks.

8 7 of recent bank consoldaton. The study wll also examne the effect that specalzaton has on bank effcency by examnng agrcultural banks. Ths examnaton of the mpact of specalzaton has not been done before. III. Methodologes, Data, and Results Ths essay examnes the survval values of small banks by lookng at ther operatng effcency. Operatonal effcency s examned for dfferent types of bank structure, such as bank sze, agrcultural loan specalzaton, and charter locaton. We employ the X-effcency methodology to measurement of the operatng effcency. Prevous studes show that there are several ways to dfferentate banks by sze. We examne bank sze usng a new sortng crtera, whch wll be dscussed n the followng sectons. The specfcaton of the bank specalty and charter locaton wll also be dscussed n the followng sectons.. Estmaton of Bank Effcency Berger, Hunter, and Tmme (1993) surveyed dfferent approaches to estmatng bank effcency. Ths study uses the measure of X-effcency of Berger (1993) that measures a bank's operatng effcency relatve to the most effcent bank. 4 To derve X- effcency, a total cost functon s frst estmated for the average bank n a gven year. Agrcultural shocks mght cause some problems for those agrcultural banks that are not well dversfed n ther loan portfolos. 4 Berger, Hunter, and Tmme (1993), and Mester (1997) also apply the same methodology.

9 Total cost s expressed as a functon of output, and nput prces. A standard translog functon s used to estmate to total costs and derve the X-effcency measure. 8 ln TC t 4 = α 4 l = 1 h = 1 0 α lh 6 k = 1 l 1 k = 1 j = 1 ln( ) ln( ht ) + where x t = X-effcency factor p lt p 4 + β k ln( y kt ) + α l ln( p lt ) β 6 4 k = 1 l = 1 δ lk ln( y kt 6 6 ) ln( p lt kj ) + ln( ln( y x t kt ) ln( y ) + u t jt ), (6 ) u t = random error TC t = total costs of bank at tme t y k = bank outputs (1: real estate loans, 2: agrcultural loans, 3: commercal and ndustral loans, 4: personal loans, 5: depost lablty for transacton accounts, and 6: depost lablty for non-transacton accounts ). p l = prce nputs (1: total nterest expenses 2: prce of captal, 3: labor, and 4: federal funds rate). Note that the jont effect of any two ndependent varables s captured wth the product terms. The effcency factor s captured n the resdual term x t when ndvdual bank data are appled to the regresson results. Data used n equaton (6) are taken from the Reports of Condton and Income Report Gude (Call Report). Commercal banks of charter number 200, 210, 250, or 340 and havng ssuer code and total assets greater than zero are ncluded n the sample. Ths defnton ensures that only commercal banks are ncluded n the sample. Cost-share equatons of each nput prce can then be derved as follows: 4 6 ln TC t = S lt = α l + α lh ln( p ht ) + ln( p ) lt h = 1 k = 1 δ lk ln( y kt ), ( 7 )

10 9 Equatons (6) and (7) are estmated subject to homogenety and symmetry restrctons usng the method of seemngly unrelated regressons (SUR). If a frm systematcally ncurs relatvely hgher costs than the other frms n a compettve envronment, t s consdered X-neffcent. In the survey conducted by Berger, Hunter, and Tmme (1993), several econometrc and lnear programmng technques have been proposed for estmatng X-effcency, ncludng the econometrc fronter approach (EFA), the thck fronter approach (TFA), data envelopment analyss (DEA), and the dstrbuton-free approach (DFA). In ths study, assumng effcency dfferences are stable, and the random error averages out over tme, the dstrbuton-free approach of Berger (1993) and DeYoung (1997) s used to estmate the effcency of the banks. Perstan (1997) shows that when we let e t = ln(x t ) + u t and transform e t such that the mnmum becomes 0; we arrve at the followng: ε ˆ t = mn{ eˆ t } eˆ t. ( 8 ) By takng the exponental of equaton (8), X-effcency s obtaned. XEFF t = exp( εˆ t ). ( 9 ) The X-effcency of bank at tme t (XEFF t ) s now normalzed to fall between zero and one. Snce XEFF t s not robust to outlers, Berger (1993) modfed the X- effcency measure such that observatons fallng below the p-th percentle are set to the p-th percentle value ( ε ˆ ( ε ˆ (1 p) t ( p) t ), and observatons exceedng the (1-p)-th percentle are set to ). The modfed X-effcency can be defned as: XEFF t ( p) p ) ( p ) = exp[ ε ˆ max{ εˆ, mn{ εˆ, εˆ ( (1 p ) t t t t }]. (10)

11 10. Bank Sze Specfcaton Banks are classfed accordng to sze, specalty and locaton based on the Federal Fnancal Insttuton Examnaton Councl (FFIEC) form 031, 032, 033, and Banks wth assets over $300 mllon are classfed as large banks, $100 mllon to $300 mllon as medum banks and less than $100 mllon as small banks. However, accordng to the sample statstcs n Panels A and B of Table 1 and Table 2, the sample s not normal dstrbuted durng the examnaton perod. Because of the unque dstrbuton of all U.S. commercal banks samples, t s hard to categorze banks by sze. More than 2/3 of banks n the U.S. are below $150 mllon n total assets n the frst quarter of 1988 and On average, over the sample perod, around 70% of the banks are small banks, 20% are medum banks, and 10% are large banks. In the largest sze category, we also notce a large dscrepancy. There are a cluster of extremely large banks wth total assets n excess of $3 bllon. Several recent studes employed dfferent approaches to classfy dfferent banks by sze. Alam(2000) and Akhgbe and McNulty(2000) categorzed large and small banks by usng a cutoff of $500 mllon n total assets. Stroh(2001) classfed banks usng more detaled categores. 7 In ths study, two dfferent approaches are used to dfferentate dfferent banks by sze. The frst uses FFIEC categorzaton and the second uses a modfed verson of the scale employed n Stroh(2001). If only few partcular cutoffs of bank sze are appled, bank growth mght result n problems. For example, f steady growth of the bankng ndustry s assumed, there wll 5 The call report forms s splt nto four forms, , representng banks wth domestc and foregn offces, banks wth domestc offces only and total assets of $300 mllon or more, banks wth domestc offces only and total assets of $100 mllon or more but less than $300 mllon, and banks wth domestc offces only and total assets less than $100 mllon. Thus, n ths study, the boundary of large, medum, and small banks are total assets of $300 mllon and $100 mllon. The cutoff of the larger banks, $300 mllon, s also consstent wth the one that Jayaratne and Wolken (1999) employ. 6 There are 86.95% and 76.95% of commercal banks are below $150 mllon n total assets n the frst quarter of 1988 and 1997 respectvely.

12 11 be an upward trend n bank sze. Inflaton ssue mght also be another ssue that wll affect bank sze. However, Table 3, Panels A and B show, usng two major ndcators of nflaton, that the nflaton rate s not very hgh durng our examnaton perod. Thus, nflaton mght not be an mportant ssue n dscussng the bank sze durng our examnaton perods. Addtonally, an upward bas of bank sze mght not exst because of the nformaton exposure n the call report. Banks n dfferent sze categores have dfferent reportng requrements from the FFIEC on the call report. For example, n the schedule RI-B of the call report, large banks, whch have more than $300 mllon, have to release the detals and temze the charge-offs and recoveres on loans and leases and changes n allowance for credt losses. 8 Small banks wth less than $100 mllon n total assets only have to dsclose the total amount of each tem. Thus, banks n the small and medum sze categores mght have ncentves to stay n the same categores n order to mantan the same level of nformaton exposure. Thus, from ths perspectve, fx cutoffs of bank sze through tme mght be approprate. To prevent a possble upward bas of the cutoffs of FFIEC, quartles are also used to dfferentate the bank sze. Table 4 shows the cutoffs usng the quartle crtera. Although usng quartle crtera can solve the upward trend of the bank sze, Fgure 4 shows that the dstrbuton of the frst and second quartle may stll be based. 9 For example, n the frst quartle, some banks are extremely large wth sze more than 10 bllon dollars n total assets. 7 Stroh(2000) classfed bank sze n dfferent categores as follows: 200 mllon, 300 mllon, 500 mllon, 1 bllon, and 5 bllon. 8 Schedule RI-B n the call report s the form for banks to dsclose nformaton related to charge-offs recoveres on loans and leases and change n allowance for credt losses. In ths schedule, medum- and small-sze banks have the same nformaton exposure. However, one of the major dfferences to large banks s the requrement to categorze each tem nto the U.S. and non-u.s. operatons. Large banks also have to release nformaton on the amount of credt cards and related plans for loans to ndvduals for household, famly, and other personal expendtures. 9 The banks categorzed n the frst quartle are the those wth total assets more than Q3 n Table 4. The second quartle s between Q3 and Q3.

13 12 In order to capture the effect of bank sze n more detal, the crtera that we use to dfferentate banks by sze s to use the followng categores: 10 mllon, 20 mllon, 30 mllon, 40 mllon, 50 mllon, 60 mllon, 70 mllon, 80 mllon, 90 mllon, 100 mllon, 150 mllon, 200 mllon, 250 mllon, 300 mllon 400 mllon, 500 mllon, 600 mllon, 700 mllon, 800 mllon, 900 mllon, 1 bllon, 1.5 bllon, 2 bllon, 3 bllon, 5 bllon, and 10 bllon dollars and above. Ths categorzaton allows bank effcences be addressed more accurately and the testng results wll not be dstorted because of the nonnormalty n the dstrbuton of bank sze.. Specfcatons of Banks Loan Specalzaton and Charter Locaton To determne loan specalzaton, the crtera used n Ellnger (1994) and Glbert and Klesen (1995) s appled. Banks wth a rato of agrcultural loans to total loans of more than 25% are classfed as agrcultural banks and less than 25% as nonagrcultural banks. Banks are placed nto two categores based on ther charter locaton, MSAs (metropoltan statstcal areas) and non-msas. We assume that commercal banks focus more on the urban market when ther charter locaton s n a MSAs and that commercal banks n non-msas target ther customers n rural areas. As to the specfcatons of bank sze, loan specalzaton, and charter locaton, three major statstcal tests are conducted. Frst, tests of economes of scale are conducted n two dfferent ways, parwse comparson tests and regresson analyss. Secondly, the relatonshp between bank effcency and agrcultural output at the county level s examned. Thrd, the relatonshp between the prce rsks of agrcultural products and bank effcency s examned.

14 13 v. Economes of Scale Tests a. Effcency Comparson Tests Commercal banks wll have effcency dfferences across sze, loan specaltes, and charter locaton. We hypothesze that the commercal banks wth agrcultural specaltes mght have superor performance relatve to the ones wthout agrcultural specaltes, that commercal banks n MSAs may operate more effcently than the ones n non-msas, and large commercal banks may outperform smaller banks because of scale economes. Therefore, parwse comparson tests are proposed to test the bank effcency based on dfferent bank sze, charter locaton and loan specalzaton crtera. Banks are categorzed accordng to ther sze, charter locaton, and specalty. Permutatons of the three factors result n a total of 12 categores. Table 5 shows the number of observatons n each category. The study perod covers 1988 through 1997 for all regons of the U.S. As dscussed n secton, the crtera used to categorze banks by sze are controversal. Although categorzng bank sze n more detal mght provde more accurate nformaton, a lack of observatons mght be a problem for comparsons n certan categores. In order to prevent the bas of unequal varance n the dfferent categores, the Cochran and Cox approxmaton of the probablty level of the approxmate t statstc s employed. We examne both the FFIEC and our modfed verson of Stroh s (2000) sze cutoffs. Statstcs of banks effcency categorzed by the dfferent crtera are shown n Table 6. Most of the observatons are small banks, whch have $100 mllon or less than $100 mllon n total assets. On average, only around one ffth of the banks are regarded as agrcultural banks. However, the number of banks chartered n MSAs s smlar to that n non-msas. Consderng the bank sze, loan specalzaton, and charter locaton jontly, most banks are urban, non-agrcultural banks n the categores of large- and medum-sze banks. However, non-agrcultural banks chartered n non-msas domnate n the small-sze category. The comparson results of bank X-effcency for the 12

15 14 categores of banks are shown n Table 7. Table 7 ndcates that large banks domnate n effcency, but medum banks are not more effcent than small banks. Metropoltan banks are superor to non-metropoltan banks whle nonagrcultural banks are more effcent than agrcultural banks. Greater competton due to bank densty n metropoltan areas and the uncertanty and volatlty n agrcultural lendng seem to exert major nfluence on bank effcency. As to the comparsons based on the bank s sze shown n Table 8, banks chartered n MSAs always outperform those chartered n non-msas, except for medum szed banks. From the perspectve of the agrcultural loan specalzaton, agrcultural banks operate more cost effcently than non-agrcultural banks n the medum and small sze categores. Comparsons based on loan specalzaton and charter locaton n Table 9, there are no economes of scale n the banks wth agrcultural loan specalzaton. Medum sze appears to be the optmal sze for the banks wth loan specaltes n agrculture to operate most effcently. Large banks operate the least effcently n the specalty of agrcultural lendng. Those banks wthout agrcultural loan specalzaton stll have economes of scale n cost effcency. However, the results based on the charter locaton are ambguous. Large banks stll operate most effcently n both MSAs and non-msas. Banks chartered n MSAs n medum sze are least effcent among all banks. Yet, medum banks chartered n non-msas have the same cost effcency as large banks. In order to nspect the effcency of banks more approprately, banks are broken nto smaller sze categores, wth cutoffs of $10 mllon, $20 mllon, $30 mllon, $40 mllon, $50 mllon, $60 mllon, $70 mllon, $80 mllon, $90 mllon, $100 mllon, $150 mllon, $200 mllon, $250 mllon, $300 mllon $400 mllon, $500 mllon, $600 mllon, $700 mllon, $800 mllon, $900 mllon, $1 bllon, $1.5 bllon, $2 bllon, $3 bllon, $5 bllon, and $10 bllon dollars and above. The averages of X-effcency n every sze categores are obtaned. Fgure 5 shows economes of scale do exst n the bankng ndustry. The larger the sze of banks, the more cost effcent they are. However, ths s not the case f the sample sze s splt nto categores based on loan specalzaton and charter locaton. Interestngly, economes of scale do not apply to the banks wth loan specalzaton n agrculture (Fgure 6 and Fgure 7). Actually, there s

16 15 an optmal sze for banks wth agrcultural specaltes. Fgure 8 and Fgure 9 show that economes of scale exst for banks n MSAs and non-msas. In addton to the comparson based on the cutoff of FFIEC, the comparsons between banks wth loan specalzaton and wthout loan specalzaton n detal sze categores are also conducted. The comparsons n the detaled sze categores are to determne whether agrcultural banks do operate effcently than non-agrcultural banks n certan sze range as suggested by Fgure 6 and Fgure 7. The results on Table 8 and Table 9 show that medum-sze banks wth agrcultural specalzaton outperform banks wth non-agrcultural loan specalzaton. However, the sze crtera used n Table 8 and Table 9 are based on the cutoffs n the call report. It s hard to tell where the exact range n whch agrcultural banks outperform non-agrcultural banks. Thus, tests of the more detaled sze categores are necessary. In Table 10, the results show that there s a specfc range of bank sze n whch agrcultural banks outperformng non-agrcultural banks. Between $20 mllon and $250 mllon n total assets, agrcultural banks operate more effcently than non-agrcultural banks. These results are consstent wth the results found usng the FFIEC cutoffs n Table 8 and Table 9. Thus, non-exstence of economes of scale n banks wth agrcultural loan specalzaton s further proven n ths test. Ths also provdes evdence to suggest that loan specalzaton may beneft smaller szed banks and that there s an optmal sze at whch loan specalzaton s effectve. Ths leads support to the dea that smaller banks do have a valuable survval value and that small banks may survve n envronments of bankng consoldaton. b. Regresson Analyss To study the mpact of agrcultural factors and scale economy on bank effcency, the X-effcency of banks s regressed on bank sze and selected agrcultural factors. Three models, the basc fundamental nformaton, the local economc actvty effects, and

17 16 the agrcultural products prce rsk effects, are examned. The fundamental model s examned n ths secton. The purpose of the fundamental model s to examne economes of scale of banks and seasonal effects on bank X-effcency. Due to the unque dstrbuton of all samples, a square term and a cube term for bank sze are added to test whether there are really economes scale n bank effcency. The followng regresson model s estmated for both agrcultural and non-agrcultural banks: XEFF = α + β sze + β β sze 2 Agprce + 3 j= 1 2 γ + β sze j q j, β 4 AAR + β Inter + β bkcapta 5 6 (11) where sze = sze of the bank, measured by the log of total assets. Squared and cubed terms for sze are also appled to capture of the feature of non-lnearty n economes of scale. Based on the results shown n Fgure 5 to Fgure 9, the relatonshp between bank effcency and the bank sze s not lnear. Thus, the square and cube terms of the sze should pck up ths observed non-lnearty. AAR = average agrcultural loan rato, calculated from average agrcultural loans dvded by total assets. Ths varable s used to classfy the bank wth loan specalzatons n agrculture. Inter = nteracton term, testng the relatonshp between agrcultural lendng and the bank s sze. If ths term s postve, banks wth loan specalzaton mght not have economes of scale. Bkcapta = banks per capta of the county level, measured as the rato of the number of commercal banks chartered n the county to the populaton of the county. Ths measure s desgned to capture bank compettveness at the county level. The Census of Populaton s used to develop ths measure. Because of the avalablty of only 1980, 1990, and 1999 populaton data, we assume the growth rate at the county level s steady. Thus, the quarterly populaton at the county level s obtaned n proporton to the

18 17 change of the data n the years for whch census data s avalable. The sgn of the coeffcent of ths varable s expected to be postve because of the ncreasng level of compettveness wthn the bankng ndustry. Agprce = volatlty of agrcultural loans based on the quarterly data of average agrcultural loans over the 10-year examnaton perod. If the proporton of agrcultural lendng fluctuates through tme, the uncertanty n the bank ncreases. Operatonal effcency s expected to declne because of the uncertanty. Therefore, the volatlty of agrcultural loans s expected to have negatve mpact on bank effcency. qj = dummy varables measurng seasonal effect. q1 represents the dummy of the frst quarter, q2 represents the dummy of the second quarter, and q3 represents that of the thrd quarter. The ntercept term measures the effect of the fourth quarter. XEFF = quarterly effcency of each commercal bank over perod. Data for dependent and ndependent varables come from the Call Report, except banks per capta (Bkcapta). The regresson analyss wll also be dvded ndvdually nto 27 ndvdual sze categores. Thus, regardless of the sze effect, detaled nformaton of the mpact on bank effcency caused by bank fundamental nformaton can be observed. Because of the sze effect has been consdered by dong regresson analyss n the smaller sze categores, the model s revsed as follows: XEFF = α + β 3 1AAR + β2bkcapta + β3agprce + j= 1 γ j q j (12) Results for equaton (11), estmated usng all banks, are shown n Table 10. The sgn of the sze and sze 3 varables are sgnfcant and negatve, whch the square term s postvely sgnfcant. The results are consstent wth those n Fgure 5 and wth prevous studes.

19 18 The average agrcultural loan rato has a sgnfcant negatve mpact on banks cost effcency. Ths means that the hgher the level of agrcultural loans, the lower cost effcency. However, the nteracton (Inter) between the agrcultural loans and bank sze (AAR * sze) has sgnfcant postve nfluence on bank effcency. The dynamc relatonshp mples there mght be an optmal sze for banks wth an agrcultural specalty. Fgure 6 shows evdence of the regresson results. Banks wth agrcultural specalzaton operate most effcently n the optmal sze range between 90 mllon and 400 mllon dollars n total assets. Banks per capta (Bkcapta) s regarded as a proxy of the compettveness of banks n the local fnancal market. Bkcapta has postve nfluence on bank effcency. Increasng competton n the local fnancal market ncreases the cost effcency of the bank. However, the volatlty of agrcultural loans (Agprce) nfluences bank effcency negatvely. In the other words, stablty n the agrcultural lendng envronment mght mprove the cost effcency of the bank. Other than those fundamental factors, seasonal effects also are tested. The results shows seasonal effects do exst. The fluctuatons of loan demand mght depend on the seasons. The seasonal effects mght be also caused by macro economy. The correlaton coeffcent matrxes are ncluded n the results as a check for possble multcollnearty based on the observatons that we use n the model. The results are shown n Table 12, Table 13, and Table 14. The results show that the correlaton among the dependent varables, except for sze, sze 2 and sze 3, are low. c. Local Economc Actvty Effects Snce ths study focus on the cost effcency of banks specalzng n agrculture, agrcultural producton at the county level s regarded as a proxy for the local economc factors. Thus, agrcultural producton may mpact the effcency of agrcultural banks. Agrcultural producton at county level are ncluded nto the fundamental nformaton model to test local economc effects on bank effcency. Accordng to the crtera n Table 15, agrcultural products are categorzed nto ten sectors. They are food grans, feed grans and hay, cotton, tobacco, ol-bearng crops, all frut and nuts, commercal

20 19 vegetables, meat anmals, dary products, and poultry and eggs. In order to calculate the proporton of agrcultural loans to specfc agrcultural products, we assume that the product of the average agrcultural loan rato (AAR) and the specfc agrcultural product rato among all agrcultural products can be regarded as the level agrculture lendng to specfc agrcultural products of the ndvdual bank. Agrcultural products ratos at county level are avalable from the 1992 Census of Agrculture. 10 Thus, the model showng the relatonshp between bank effcency and varous agrcultural factors s examned on equaton (13). The dependent and ndependent varables are the quarterly average of data between 1988 and 1992 because of the avalablty of agrcultural data. Furthermore, we examne the relatonshp between agrcultural prce change from 1987 to 1992 and X-effcency change from 1988 to 1992 s examned. The future change of bank effcency from 1992 to 1997 s examned to see f t s nfluenced by the change of agrcultural factors from 1988 to The above two models, whch examne the mpacts of the change of agrcultural factors on the smultaneous and future changes of bank effcency, are shown on equaton (14). XEFF = α + β sze + β sze + β Agprce + 10 j= β sze j 3 ρ ( AgR j 3 + β 4 * AAR ). AAR + β Inter + β bkcapta 5 5 (13) XEFF = α + β sze + β + 10 j= 1 1 ρ ( AgR j j 2 AAR + β Inter+ β * AAR ) 3 bkcapta+ β Agprce 4 5 (14) 10 In the Census of Agrculture, agrcultural products data are avalable based on unt, prce, land sze, and farm numbers. Because of the varety of unts of agrcultural products, we calculate the agrcultural product ratos based on prce. Farm sze s not an approprate proxy. The producton of one acre of tobacco s dfferent from that of one acre of vegetables. Because the census s formed by the voluntary submsson of nformaton, the number of farms s not representatve of all producton. Thus, the value of sale of agrcultural products n dollars wll be more approprate to measure the rato of agrcultural products n a gven county.

21 20 In order to understand the mpact of agrcultural factors on banks n dfferent sze categores, the model smlar wth equaton (13) s also conducted to observe the agrcultural effect of the smaller sze categores. Due to the level of dversfcaton of loan portfolos, agrcultural factors are expected to have more explanatory power on small banks than on large banks. The equaton can be obtaned as follows: XEFF = α + β + 10 j= 1 j 1 AAR + β ρ ( AgR j bkcapta+ β 2 * AAR ) 3 Agprce + 3 j= 1 γ j q j (15) The results are splt nto three groups, all observatons, agrcultural banks, and non-agrcultural banks. Ths wll allow us to determne the specfc mpact of agrcultural factors on agrcultural banks. Table 16 shows the results on equaton (13). The correlaton matrxes of the model based on dfferent observatons are shown n Table 17, Table 18, and Table 19. The results n Table 16 show that bank sze has postve effect at 1% sgnfcance level on bank X-effcency n all three dfferent categores based on all banks, agrcultural banks, and non-agrcultural banks observaton. Average agrcultural loans have the same effect as well. Interacton term (Inter), the product of bank sze and agrcultural loans, s 1% postve sgnfcant. However, only agrcultural loan volatlty has negatve mpact on bank effcency at 1% sgnfcance level. Because agrcultural banks more rely on agrcultural loans, the volatlty of agrcultural loans play a more mportant role n agrcultural banks than non-agrcultural banks. As to agrcultural factors, food gran s only postve sgnfcant on all banks observatons. Feed grans and hay, tobacco, ol-bearng crop, meat anmals, and dary products are all postve sgnfcance on bank effcency. Frut and nuts and vegetable, whch are regarded as commercal agrcultural products, have negatve sgnfcant mpact on only agrcultural bank effcency. Usually, those commercal agrcultural products have hgher proft margn than other crops. As to the demand and supply theory, more supply may make prce and proft margn lower. Thus, the hgher producton of those products, the lower profts they wll be. The probablty to pay off agrcultural loans may be lower. Indrectly, t affects cost of bank operaton, whch has hgher porton of agrcultural

22 21 loans. Thus, the hgher producton of frut, nuts, and vegetable only affect agrcultural bank X-effcency negatvely. On the other hand, cotton, poultry and eggs affect nonagrcultural banks effcency postvely. Those products are more relatve to urban lves. The hgher producton of cotton, poultry, and egg may drve the prce down and reduce the costs of ndustral producton, whch are the major target customers of nonagrcultural banks. Thus, the lower producton costs may drectly ncrease the probablty of ndustral borrowers to payoff the loans. Thus, the cost effcency of non-agrcultural banks ncreases. As to the adjusted R Square, the results show that our model has more predctablty on agrcultural banks than on non-agrcultural banks. Table 20 shows the results of relatonshp between percentage change of bank effcency between 1988 and 1992 perods and percentage change of agrcultural factors between 1988 and 1992 perods on equaton (14). The results are splt nto three categores, based on dfferent banks observatons. Table 21, Table 22, and Table 23 show the matrxes of correlaton of those ndependent varables. Percentage change of bank sze has postve effect n all three categores. However, nteracton term (Inter), product of bank sze and agrcultural loans, has negatve sgnfcant mpact on all banks and non agrcultural banks effcences. It mples that the ncrease of agrcultural loan raton decrease all banks and non-agrcultural banks effcency. The compettveness of local fnancal market does not enhance bank effcency. The volatlty of agrcultural loans has sgnfcant mpact on bank effcency. It postvely affects all bank and nonagrcultural bank effcences, but negatvely affect agrcultural effcency, whch s consstent wth the prevous results. As to the percentage changes of agrcultural factors, generally, changes of agrcultural factors have more mpacts on agrcultural banks. The percentages change of food grans, feed grans and hay, cotton, frut and nuts, meat anmals, dary products, and poultry and eggs nfluence agrcultural bank effcency negatvely. However, the percentage change of tobacco has postve effect, whch s the reverse result on non-agrcultural banks.. We also test whether post nformaton of percentage change of bank fundamental nformaton and agrcultural factors affect the future percentage change of bank effcency by usng the same equaton on equaton (14). The percentage change of bank

23 22 effcency between 1992 and 1997 regresses on the percentage changes of banks fundamental nformaton and agrcultural factors between 1988 and Table 24 shows that the percentage change of banks fundamental nformaton and agrcultural factors do not affect much the future performance of bank operaton. Accordng to R Squares of the model, the changes of agrcultural factors have poor predctablty of bank future effcency change. Table 25 shows the results of the relatonshp between bank effcency change from 1988 to 1997 and changes of bank s fundamental nformaton and agrcultural factors from 1988 to The results are smlar wth those n Table 20. However, nteracton term (nter) s not sgnfcant on agrcultural banks. The percentage change of tobacco s not sgnfcant on agrcultural banks, too. The R Squares are also consstent wth prevous results that the model usng agrcultural banks observatons has hgher adjusted R Square. d. Agrcultural Products Prce Rsk Effects If the bank s loan portfolos put too much weght on a specfc leadng specalzaton, bank-operatng costs wll solely depend on the local economc actvtes related to the bank s lendng specalty. Thus, economc fluctuaton wll affect bank s effcency n some crcumstances. Dversfcaton of loan portfolos mght be an mportant ssue to bank effcency. However, because of the nnovaton of fnancal markets, there are several fnancal nstruments for hedgng the rsk n the commodty market, such as futures contracts. We assume banks take advantage of futures contracts to hedge rsk. Therefore, the dversfcaton of loan portfolos mght not be an ssue. In addton to agrcultural prce as a proxy for agrcultural factors, the prce volatlty of agrcultural products s also consdered n ths model. Prce volatltes of the agrcultural products (AgPV), shown n Table 26, are calculated from the prce ndex of each agrcultural category at the natonal level to proxy for the prce rsk of the agrcultural products. The ndexes for agrcultural products are avalable from the USDA Statstcal Bulletn. A dummy varable (DumFC) s desgned to represent the avalablty hedgng opportuntes by the bank usng agrcultural commodty futures contracts.

24 23 Although agrcultural banks wth hgher agrcultural loans and less dversfed loan portfolos mght have hgher rsk exposure, banks can employ hedges to mnmze prce rsk n order to prevent losses from agrcultural shocks. The ratonale for the hedgng varable s as followng. If there s commodty n producton at the county level and a futures contract on the commodty s also avalable, the bank could requre the borrower to engage n a future contract to hedge the poston. 11 The dummy s defned as zero n ths stuaton as all rsk would be hedged. If the commodty s not avalable at the county level, the bank makes no loans on such commodtes, t s not essental for banks to requre borrowers to hedge. Hence, the dummy n ths stuaton s zero as well. If the commodty s avalable but the future contract s not traded, banks have a rsk exposure wthout any rsk hedgng. The dummy s one n ths case, represented the presence of an unhedged prce rsk. Table 27 summarzes the desgn of the dummy varable. Table 28 shows the futures contracts that are avalable for commodtes. Therefore, f there s a rsk exposure n the partcular agrcultural product, the volatlty of the prce should matter. The nteracton terms of agrcultural prce volatlty and the futures contract dummy varable (AgPV*DumFC) represents the prce rsk of the agrcultural products. The model consderng agrcultural prce rsk s gven n equaton (16). Because of the avalablty of agrcultural data, a test s done by usng quarterly average of bank X-effcency and quarterly average of banks fundamental nformaton and agrcultural factors n 1992 based on three dfferent observaton categores. XEFF + β = α + β 6 1 sze + β j= 1 2 sze 2 + β Agprce + ρ j ( AgRj * AAR ) + ω j 3 sze 3 + β 4 j AAR + β 5 ( AgPV Inter + β j 5 * DumFC bkcapta j ) (16) 11 Farmers would be short n the futures contract.

25 24 Table 29 shows the results on equaton (16). Interestngly, sze effect s not consstent wth the prevous results. Bank sze has negatve sgnfcant effect on all banks, agrcultural banks, and non-agrcultural banks effcences. Average agrcultural loans rato has the same negatve mpact as well. However, nteracton term (nter), the product of bank sze and agrcultural loan rato, has postve nfluence n three categores. The proxy of local fnancal market compettveness has postve mpact on bank effcency. The result of agrcultural loan volatlty (Agprce) s consstent wth prevous result. It (Agprce) negatvely affects agrcultural banks effcency. As to agrcultural factors, feed grans and hay, ol-bearng crop, and dary products have postve mpacts on banks effcency based on dfferent observaton categores. Tobacco, meat anmals, and poultry and eggs have postve sgnfcant effects on nonagrcultural banks effcency nstead of agrcultural banks. Although loans to tobacco producton do not sgnfcantly affect agrcultural banks effcency, fluctuaton of tobacco prce does have mpact on not only on all bank and non-agrcultural banks but agrcultural banks. However, the stuaton for the frut and nuts and vegetable s dfferent. The loans to frut and nuts and vegetables have sgnfcantly negatve mpact on agrcultural banks. However, the prce rsks of frut and nuts and vegetable also have nsgnfcantly negatvely mpact on agrcultural banks. Because agrcultural banks are specalzed n agrcultural lendng, they are senstve to the producton of those commercal products, lke frut and nuts and vegetable. Although there are no futures contract to hedge the rsk, agrcultural banks have more nsder nformaton than nonagrcultural banks. Thus, agrcultural banks may have more ablty to protect from the prce rsk of frut and nuts and vegetable. As to the adjusted R Squares, the results are also consstent wth the prevous results and our expectaton. The model examned by usng agrcultural banks observatons has more explanatory power than by usng all banks and non-agrcultural banks observaton. Accordng to the prevous hypotheses, we also examne prevous models by usng detaled sze categores. The results of bank fundamental nformaton effect model, local economc actvty effect model, and agrcultural prce rsk effect model n detaled sze categores are shown n Table 30. In general, models examned by usng agrcultural banks observatons have hgher adjusted R Squares than by usng all banks and non-

26 25 agrcultural banks observatons. Especally n the local economc actvty effect model and agrcultural prce rsk effect model, agrcultural factors have more explanatory power on agrcultural banks effcency. Agrcultural banks wth 70 to 90 mllon dollars n total assets have the hghest adjusted R Square among those tests. IV. Conclusons In ths study, we nvestgate two effects on bank effcency by employng X- effcency and agrcultural factors at the county level. Frst, we examne f bank s sze, charter locaton, and lendng specaltes explan dfferences n X-effcency across commercal banks. We do fnd support for the hypothess that bank s sze, charter locaton, and loan specaltes play an mportant role n determnng effcency n the bankng ndustry. Between 1988 and 1997, large commercal banks operate more cost effcently than other smaller commercal banks. Non-agrcultural commercal banks and commercal banks chartered n MSAs also outperform agrcultural commercal banks and commercal banks chartered n non-msas. If the bank s sze, charter locaton, and lendng specaltes consdered jontly, large and small non-agrcultural commercal banks chartered n MSAs have the hghest X-effcency n those specfc categores. Only n the medum-szed commercal bank category do non-agrcultural commercal banks chartered n non-msas outperform medum-szed commercal banks n non-msas. Overall, we can conclude that commercal banks chartered n MSAs wthout agrcultural specaltes operate most effcently. There s no dfference n X-effcency between agrcultural commercal banks n MSAs and non-msas. Interestngly, agrcultural commercal banks operate less effcently than do non-agrcultural commercal banks n non-msas. The results show that economes of scale n commercal banks do exst. However, n the perspectve of the bank s loan specalzaton, t s not necessary that the bank s sze should be large to be effcent.

27 26 The results show that there s an optmal sze for banks wth a loan specalzaton. Banks wth loan specalzaton n agrcultural lendng are more effcent wthn the sze range of 80 mllon to 400 mllon dollars n total assets. Ths mples that smaller banks do have survval value n terms of cost effcency. It s not necessary for specalty banks to be large to be effcent. Regulators also do not have to worry about the survval problem and competton of the smaller banks. Banks wthout large asset sze but wth agrcultural loan specalzaton are expected to reman compettve n the local market. As long as those smaller banks stll exst and perform well after the deregulaton n the bankng ndustry, the local fnancal market wll reman compettve and the worry of the lack of adequate credt source mght be redundant. Our second test s that the X-effcency of commercal banks may be nfluenced by agrcultural factors. We nvestgate the relatonshp between the X-effcency of commercal banks and agrcultural factors from the 1992 Census of the Agrculture at the county level. We fnd that agrcultural factors do affect the cost effcency of commercal banks. Interestngly, non-agrcultural commercal banks are also nfluenced by agrcultural factors. Thus, agrcultural factors mght be one of the consderatons n the evaluaton of the commercal banks X-effcency, especally for small-szed commercal banks. The X-effcences of large-szed commercal banks are not shown to be affected by agrcultural factors. Large-szed commercal banks may depend less on local economc actvtes or prce change, thus, elmnatng ther dependence on agrcultural actvty.