Does Interbank Borrowing Reduce Bank Risk? Valeriya Dinger* and Jürgen von Hagen ** Abstract:

Transcription

1 Does Interbank orrowng Reuce ank Rsk? Valerya Dnger an Jürgen von Hagen May 2005 a Abstract: In ths paper we nvestgate whether nterbank exposures create ncentves for nterbank montorng as sgnalle by reuce level of rsk unertakng of the nterbank borrowng banks. We present a moel of the cret market base on asymmetrc nformaton an moral hazar. Assumng that banks have montorng costs benefts compare to epostors regarng the lenng actvtes of the other banks, we show that nterbank lenng nuces the borrowng banks to engage n less rsky lenng actvtes than banks that fnance themselves preomnantly n the epost market. We emprcally test the mplcatons of the moel on a large sample of banks from 10 Central an Eastern European countres. The results of the emprcal analyss generally confrm the mplcatons of the moel. Key wors: bank specalzaton, nterbank market, rsk unertakng, transton countres JEL: 21, E53 Center for European Integraton Stues, Unversty of onn. E-mal aress: valerya.nger@un-bonn.e Center for European Integraton Stues, Unversty of onn, Inana Unversty an CEPR. E-mal aress: vonhagen@un-bonn.e a We thank Henrk Hakenes, Iftekhar Hasan an Erk Thessen an partcpants at XII Tor Vergata Conference on Money an ankng -2003, the Far Eastern Meetng of the Econometrc Socety an the Veren für Socalpoltk Tagung 2004 for valuable comments on earler rafts. The remanng errors are ours.

2 1. Introucton Recent lterature on nterbank exposures has emphasze the role of the nterbank market as a source of contagon (Allen an ale, 2000 an Frexas, Parg an Rochet, 2000). However, Rochet an Trole (1996) argue that, by generatng ncentves for lenng banks to montor nterbank-borrowng banks, nterbank exposures may also contrbute to pruent market behavor an reuce the rsk of bank falures an systemc stress. The ea s that banks are partcularly goo at entfyng the rsks of other banks (Calomrs, 1998). Prove wth proper ncentves, they can perform a complementary task to bank regulaton an supervson by the authortes 1. Despte the obvous appeal of ths ea, emprcal research on the ssue s lmte. In a frst step n ths recton, Furfne (2001) examnes the prcng of nterbank lenng agreements as an ncator for the ablty of banks to montor ther nterbank borrowers. Snce nterbank loans on the feeral funs market are large an uncollateralze, they expose lenng nsttutons to sgnfcant cret rsk. Lenng banks, therefore, have an ncentve to montor ther counterpartes an prce these loans as a functon of the cret rsk of the borrowng bank. Furfne s emprcal results support ths hypothess by showng that borrowng banks wth hgher proftablty, hgher captal rato, an fewer problem loans pay lower nterest on feeral funs loans. However, the mpact s farly small; for example, a one stanar evaton rse n the loan-to-captal rato rases the sprea by merely 1.5 bass ponts. A potental reason for the small economc sgnfcance of the results an the low emprcal research nterest n the ssue s that economc analyss of nterbank trae has so far concentrate on hghly evelope bankng markets, where nterbank exposures are mostly 1 The ea of usng banks as montors of other banks s a part of a broaer concept of usng market scplne base on market nformaton as a complement to bank supervson an regulaton (erger, Davs an Flannery, 1998; De Young et al, 1998; Peek, Rosengren an Tootell, 1999). 2

3 generate by short term lquty nees (hattacharya an ale, 1987; Allen an ale, 2000; Hellwg, 1994). In contrast, Rochet an Trole (1996) stuy the ncentves for nterbank montorng generate by long-term lenng commtments. In ths paper, we focus on a sample of countres where nterbank trae s the result of long-term specalzaton of some banks n ssung eposts an of others n lenng to nonbanks. In such an envronment, nterbank lenng s characterze by longer-term maturtes. Ths proves a more sutable framework to test whether nterbank lenng s accompane by nterbank montorng. 2 More specfcally, we stuy a sample of Central an Eastern European (CEE) countres, where nterbank trae s mostly cause by specalzaton. ank specalzaton n CEE has been the result of the fast eralzaton of the bankng nustres whch took place n an envronment of mmature an malfunctonng bankng regulaton. The former monobank systems n these countres were eralze n the early 1990s an a large number of new players entere the market. Nevertheless, n several countres, the successor nsttutons of the monobanks manage to explot the avantages of ther wesprea branch networks an customer relatons to specalze n epost rasng actvtes an preserve large market shares n the market for customer eposts. However, these banks are very nactve n the market for loans for prvate 3. In contrast, most of the new entrant banks specalze n provng cret to the new emergng prvate sector. As a result of ths specalzaton, large ncumbent banks gather persstently more funs as customer eposts than they strbute to the nonfnancal sector. In contrast, new entrant banks, have less customer eposts than the amount of loans 2 Although the role of specalzaton as a cause of nterbank trae has stll not been stue thoroughly, ntal stues ncate that t s a val explanaton for the very hgh ntensty of nterbank trae n countres wth relatvely unerevelope fnancal systems. See e.g. almes an Manzano (1995) for the Spansh bankng system, Sle an Cole (1998) for Inonesa, an onn et al (1998) for several Central an Eastern European (CEE) transton economes. 3 The lterature on transton states the lack of relatons to newly create enterprses, expertse n proper cret allocaton actvtes, an state-of-the-art market orentaton as sources of the ncumbent banks nablty to prove cret to the prvate sector. Another potental reason s the fact that, for a prvate enterprse, to receve a cret from a large bureaucratc bankng nsttuton s assocate wth some extra costs beyon the nterest rate pa. Such costs woul nclue tme spent n a bureaucratc cret approval proceure, but n extreme cases t coul also nclue brbery, etc. 3

4 they coul fnance (onn et al, 1998). Where ths s the case, the nterbank market tens to clear the screpances between eposts an loans for banks wth a strong specalzaton n the one or the other recton. Thus, a bankng system wth a two-ter market structure has emerge. The frst ter conssts of ncumbent banks that gather eposts from the nonbank publc an transfer them through the nterbank market to the secon ter banks, manly new entrants, whch prove cret to the nonfnancal sector. Wthn ths structure, the new banks that borrow from the ncumbent banks an len to the nonbank publc are typcally much smaller than the ncumbent banks. As a result, there s not much pressure on governments to exten epost guarantees or bal out fnancally trouble nterbank-borrowng banks, snce the perceve rsk of systemc crss s small. In contrast, ncumbent banks typcally enjoy explct bal-out guarantees or are covere by epost nsurance offere by the governments. Ths renforces ther compettve avantage n the market for eposts. At the same tme, the ncentves for nterbank montorng are not storte by the expectaton that the borrowng bank wll be bale out by the government (e.g., Rochet an Trole, 1996). In the next secton, we present a moel of the cret market base on asymmetrc nformaton an moral hazar. We assume that large banks have montorng cost avantages over epostors regarng the lenng actvtes of other banks. In ths settng, we show that banks that fnance themselves through the nterbank market fnance less rsky projects n equrum than banks that fully fnance themselves through customer eposts 4. The reason s that the former are montore by ther cretors n the nterbank market, whle the latter are not. In secton 3, we present our ata set an n secton 4 we prove emprcal evence for the two-ter structures of the bankng nustry n CEE countres base on the flow of funs from 4 Ths result s consstent wth llet, arfnkel an O Neal (1998) who postulate that banks have the ablty to thwart market-base montorng by shftng ther labltes from unnsure epostors wth a great ncentve to montor to nsure epostors wth lttle ncentve to montor. 4

5 the large banks to the rest of the bankng system. 5 In secton 5 we test the man hypotheses erve from the theoretcal moel an some relate extensons. Secton 6 conclues. 2 ank specalzaton, Interbank orrowng, an Montorng We ntrouce a moel where montorng of bank nvestment behavor s necessary to prevent a cret market collapse but s too costly to be performe by the nvual epostors. In ths stuaton, the montorng functon can be performe by large bankng nsttutons whch len funs to smaller banks an have, ue to economes of scale, cost avantages n montorng the cret actvtes of the banks borrowng funs from them. In our moel banks are able to refnance n two segmente markets: the market for customer eposts an the nterbank market 6. If a bank chooses to refnance n the nterbank market ts nvestment behavor wll be montore, whereas f t refnances n the customer eposts market t wll not be montore. Conser an ntermeate cret market wth two types of entrepreneurs: those wth a goo project an those wth a ba project. oth projects requre an nvestment of sze 1. The return of a goo project s wth probablty return of a ba project s wth probablty n case of success or 0 otherwse; the an 0 otherwse. oo projects have a postve net present value, whereas ba projects have a negatve net present value: 1 > 0 > 1, <, > (1) oth types of entrepreneurs have no equty an can only fnance the project f they receve a bank cret of sze 1. 5 The sample conssts of ten CEE countres: ulgara, the Czech Republc, Estona, Hungary, Latva, Lthuana, Polan, Romana, Slovaka, an Slovena. 6 The stncton of two segmente markets for sources of funs s smlar to the one n llet an arfnkel (2004). 5

6 The bankng sector conssts of two types of banks: one large bank ( L ) an n small banks ( S, = 1,2,..., n ). The large bank has an mplct or explct epost guarantee by the government. The epost rate wth the large bank ( ) s, therefore, equal to the rskless nterest rate, consere for smplcty to be zero. That s, the nomnal repayment to the epostors, D, for a eposte amount of 1 s equal to 1. We assume that the large bank has zero margnal costs of epost gatherng snce t has a wesprea epost gatherng network. However, t has a compettve savantage compare to the small banks n lenng to the non-fnancal prvate sector. For smplcty, we assume that the large bank oes not len to the non-bank publc at all. Thus t nvests only n two types of assets: lqu bons whose net return s normalze to zero an nterbank loans whch bear an nterest rate b. The nomnal repayment rate that the large bank requres on an nterbank loan of volume 1 s enote by ( =1 + ). Sellng bons, the large bank can supply any eman for nterbank loans. b Small banks are moele as bankng nsttutons wth unversfe portfolos an no government protecton. They ffer among each other n the volume of equty, enote wth E. They can refnance ther nvestments n two segmente markets, the market for customer eposts an the nterbank market. Small banks have no access to epost nsurance 7. Denote by sb the small banks nterest rate on customer eposts. Ths epost rate s postve (hgher than the one of the large bank, > ) because epostors are aware that eposts wth small sb banks are not guarantee an are thus rsker than the ones wth the large bank 8. The amount of nomnal repayment of a small bank to ts epostors for a epost n volume of 1 s enote by D, D =1 + sb. Denote wth α the volume of customer eposts that S accumulates. In 7 We assume that no unversal epost nsurance scheme s n force n the market. 8 We o not explctly moel competton n the epost market n ths moel. At the en of ths secton we prove a proof of the exstence of equrum n the market for customer epost n small banks. 6

7 aton to nterest payments each C( α ). 9 S ncurs ncreasng margnal costs of epost gatherng Alternatvely, the small banks can refnance ther nvestments n the nterbank market. The only prover of funs n the nterbank market s the large bank, snce t s the only one that can gather eposts at a preferental rate. If S eces to accumulate customer eposts, t wll ask for an nterbank epost n the volume of 1 E α. α n the market for Each small bank can only fnance the project of one entrepreneur. Small banks can observe the type of the entrepreneur at no cost an set the respectve repayment rates R an R. The expecte return on a loan for a ba project s negatve, whereas the one for a goo project s postve: R 1 > 0 > R 1, R > R (2) Entrepreneurs an banks have lmte lablty an are rsk neutral. ecause of the lmte lablty a moral hazar problem emerges: a small bank coul have an ncentve to fnance an nvestment n a ba project, espte ts knowlege of the project s negatve net present value. Ths result s formalze n the followng lemma: Lemma 1: Denote by R the repayment S owes to ts cretors ( R = 1 E ) D + C(1 E ) ( f S s fully fnance by customer eposts an R = 1 E α ) + α D + C( α ) f t s ( partly fnance by customer eposts an partly by an nterbank epost). The small bank wll choose to nvest n the goo project f, an only f: 9 Small banks n our moel are assume to have no epost gatherng network. They operate wth few branches locate near to the potental cret users. Thus epost accumulaton beyon the area of operaton of these branches s assocate wth ncreasng margnal costs snce more stant epostors have to be compensate for transportaton costs, nformaton asymmetres, etc. 7

8 R R RC = R + ( ) E (3) Proof: See Appenx If R < 1, S has an ncentve to nvest n a ba projects for every feasble repayment rate C an the cret market collapses, therefore we assume that R 1, for now on. C Now, assume a stuaton of strong moral hazar where R C < 1 for each = 1,2,..., n. Snce otherwse epostors woul expect a loss even f all small banks nvest n goo projects, t must be true that D 1. Ths mples that D > R C for all S. Therefore, each nvest n the ba project, f t s fully fnance by customer eposts. That s, f all S wll S are fully fnance by customer eposts, the cret market wll collapse as epostors realze losses n expectatons. A technology for montorng the small banks exsts. At a fxe cost M > 0 the cretors of a small bank can screen the type of the project an ntrouce fferent nterest rates accorng to the project type. 10 However, n orer for montorng to be conucte the fxe cost of montorng shoul be justfe by the benefts of t. Invual epostors have an enowment of sze ε << 1, whch they choose to epost n ether the large bank or n one of the small banks. We assume ε < M, from whch t follows that nvual epostors cannot nvest n montorng. Therefore, they cannot stngush whether S woul fnance the goo or the ba project. That s why they requre a unform epost rate for eposts wth small banks. 10 We aopt a broa concept of montorng ntrouce n Hellwg (1991), accorng to whch montorng nclues: screenng of projects (a pror) preventng opportunstc behavour of the borrower urng the realsaton of the project (moral hazar) punshng or autng a borrower who fals to meet contractual oblgatons. 8

9 There are two tme peros. In pero 0, each small bank eces on the structure of ts labltes an fnances a goo or a ba project epenng on whch one bears a hgher net expecte return. In pero 1 the project returns are realze, the small banks get repayment from the entrepreneurs an repay ther cretors. To smplfy computatons we assume that the costs of epost gatherng C( α ) are ncurre by S at tme pero 1. The problem of the large bank If the large bank eces to montor the nvestment behavor of the small bank t coul set fferent nterbank rates epenng on whether the goo or the ba project woul be fnance. Moreover, uner the assumpton of a low nterest rate for eposts at the large bank we can prove that the large bank s only ntereste n nvestng n goo projects. Ths result s formalze n Lemma 2. Lemma 2: The large bank wll maxmze ts net expecte return, f t proves nterbank loans only to small banks, whch nvest n goo projects. Proof: See Appenx However, the large bank as a cretor of the small bank wll montor f, an only f, the beneft of montorng, that s the fference of the expecte repayments, s no lower than the montorng costs. That s, the volume of the nterbank epost has to be hgh enough to justfy the cost of montorng. From whch the followng proposton follows: Proposton 1: The large bank wll supply an nterbank epost f an only f ts volume 1 E α satsfes: M 1 E α. (4) ( ) 9

10 In ths case t wll nvest n the montorng technology to ensure that the goo project wll be fnance. Proof: See Appenx The problem of the small bank Small banks ecsons on the structure of ther labltes an on whch project to fnance conser the fact that f they ask for nterbank fnancng they woul be montore an can thus only fnance the goo project. Therefore, they wll ask for nterbank fnancng f an only f the goo project can brng hgher net expecte return than the ba project. The net expecte return of the ba project s equal to the expecte return n case of success net of refnancng costs mnus the loss n case of falure (n ths case S loses ts equty): [ R ( 1 E ) D C(1 E ) E ] (1 ) E (5) On the other han, the net expecte return of the goo project s a functon of α an s equal to the expecte return n case of success net of refnancng costs mnus the expecte loss n case of falure (the loss of equty): [ R ( 1 E ) Dα C( α ) E ] (1 ) E α (6) Lemma 3: Let us enote wth α the volume of customer eposts at whch the net expecte return of S has a maxmum. Then α s the soluton of: [ R (1 E α ) Dα C( α ) E ] (1 E maxα (7) ) an s equal to the volume of customer eposts, at whch the margnal cost of epost gatherng equals the fference between the nterest rate on nterbank eposts an the nterest 10

11 rate on customer eposts. Ths volume oes not epen on the amount of equty, an α = 1 = α 2 =... = α n α. Proof: See Appenx Then project at S wll ask for an nterbank loan f an only f the net expecte return of the goo α s hgher than the net expecte return of the ba project. That s, for an nterbank loan f an only f the followng nequaton (8) hols: S wll ask [ R (1 E α ) Dα C( α ) E ] [ R (1 E ) D C(1 E ) E ] (1 ) E (1 ) E (8) From whch the followng proposton follows: Proposton 2: S wll accumulate α as customer eposts an eman 1 α E n the nterbank market, f an only f: U ( E, + )) I(, E ) I[ 0, 1] E S = ((, E) m, (9) where E m M = 1 α s the crtcal value below whch the large bank wll exten ( ) an nterbank loan, an E (, E) U ( E, + ) enotes the set of s for whch the hghest feasble net expecte return of the goo project s hgher than the one of the ba project. E If E S, then ether S oes not eman or the large bank reject to supply an nterbank loan. In ths case S refnances only n the market for customer eposts. Proof: See Appenx Proposton 2 formalzes the ntuton that nterbank borrowng takes places f both the small bank has an ncentve to ask for an nterbank epost an the large bank has an ncentve to 11

12 supply t. In ths case the small bank nvests n a goo project. If U ( E, + )) I(, E ) I[ 0, 1] E S = ((, E) m, then S oes not borrow from the large bank. Ths s ether because S oes not ask for an nterbank epost because the ba project s always more proftable for t than the goo one. Or because the large bank rejects to prove the nterbank epost snce ts volume s too small to justfy montorng costs. In ths case project. S s fully fnance by customer eposts an snce D > R C t nvests n the ba Exstence of equrum n the market for customer eposts In the next step we erve the exstence of equrum of the market for customer eposts n small banks uner contons (a) strong moral hazar ( < 1, 1,2,... n) an (b) a RC = proporton of small banks are montore. Thus, we have to erve the exstence of an equrum repayment rate on customer eposts n small banks D, such that for the populaton of small banks as a whole D D = 1, where D enotes the probablty of repayment to the epostors. We pool together montore an unmontore small banks, as nvual epostors are not able to stngush among them, an charge a unform repayment rate. We assume that wthout nvestng n montorng the nvual epostors are not able to gather any nformaton about small banks balance sheet 11. Assume the strbuton of E s such that for a proporton β of the n small banks E S a result βn small banks receve nterbank loans an fnance the goo project, whereas ( 1 β )n small banks are fully fnance by customer eposts an fnance the ba project. Therefore, the amount eposte by customer epostors n the montore banks equals α βn.. As 11 Ths assumpton s crucal for the exstence of an equrum rate on eposts n small banks. If epostors are able to observe nterbank borrowng an assocate t wth lower rsk, then they wll requre hgher nterest rate from small banks whch are not nterbank borrowers, at ths rate all small banks wll fnance the ba project an epostors wll realze losses n expectaton. 12

13 The amount eposte by customer epostors n banks, whch are not montore equals k (1 β ) n + ( 1 β ) nγ, where γ = (1 E ) /[(1 β ) n] = k enotes the average amount of customer epost of small banks whch o not borrow n the nterbank market. The expecte repayments by both groups of banks are expresse by α βn D an ( 1 β ) nγ D, respectvely. Exstence of compettve equrum for eposts n small banks requres that the sum of expecte repayments equals the amounts eposte, that s: α βn D + (1 β ) γn D = α βn + (1 β ) γn, (10) from whch follows: α β + (1 β ) γ D = α β + (1 β ) γ (11) β belongs to the nterval (0, 1] an γ belongs to the nterval (0, 1), D s lmte n the nterval [ /,1 / ] 1. Therefore, for all feasble β an γ there exsts a epost rate D for whch a compettve equrum n the market for customer eposts n small banks exsts. If β =1(all small banks receve nterbank fnancng an are therefore montore), then D = 1 / (all fnance projects are goo). β = 0 s not a feasble opton as t means none of the small banks s montore, an therefore they all fnance a ba project an the cret market collapses. To recaptulate, n the case of strong moral hazar equrum n the market for customer eposts s only possble f a montorng technology s ntrouce an some of the small banks are montore, prove that nvual epostors cannot stngush between montore an unmontore small banks. In ths case, whether a bank wll nvest n a goo or a ba project epens on whether t s fnance by nterbank eposts. The small bank wll have a 13

14 lower rsk level f t receves an nterbank epost from a large bank. The volume of the nterbank epost shoul be suffcently large to justfy montorng costs. 3 Data sources Our sample covers bankng nsttutons from ten CEE countres, ulgara, the Czech Republc, Estona, Hungary, Latva, Lthuana, Polan, Romana, Slovaka, an Slovena. For the macro level varables we use ata prove by IMF n the Internatonal Fnancal Statstcs CD-ROM ssues. Mcro level ata stem from banks fnancal statements prove by ankscope 12. Our sample nclues 296 banks, of whch 28 are ulgaran, 35 Czech, 12 Estonan, 36 Hungaran, 28 Latvan, 14 Lthuanan, 56 Polsh, 34 Romanan, 24 Slovakan, an 29 Slovenan. In each of the sample countres, ankscope covers 70-90% of the banks calculate as percentage of bankng assets. We have restrcte the analyss to the pero of Data for some of the banks are avalable for only some of the years, whch results n an unbalance panel ataset. 4 The two-ter bankng system To measure the magntue of nterbank transfers of funs on the nterbank market, we use two varable. The frst one s the rato of large banks net nterbank assets to customer eposts. (NIA /CD, where NIA enotes net nterbank assets, calculate as the fference between eposts wth banks an eposts from banks, CD enotes customer eposts, an the subscrpt stans for large banks). Ths rato, whch we call large banks lenng rato, s closest to the phenomenon we stuy an represents the share of eposts of the large banks that are passe on to other banks. If ths rato s postve, large banks are net leners n the nterbank market. 12 ankscope s a atabase create by ICA an ureau van Djk 14

15 Snce the large banks borrowers mght be nsttutons locate abroa rather than omestc nsttutons, we also look at the net nterbank poston of small banks relatve to ther loans to nonbanks. We call ths the small banks borrowng rato, NIA sb /L sb, where L enotes loans.it s an ncator of the extent to whch small banks rely on nterbank funs for the fnancng of ther loans. A large negatve rato results from the small banks relatve strong relance on nterbank funs. The constructon of these varables s base on a stncton between large an small banks. We choose to treat as large those nsttutons, whch nvually gather at least 20% of the total amount of customer eposts n ther bankng systems 13. All other banks are consere to be small. The values of the transfer of funs varables by country an year are presente n Table 1an Table 2. As shown n Table 1 ulgara, the Czech Republc, Hungary, Polan an Slovaka have persstently hgh postve values of large banks lenng rato, whereas Lthuana, Romana an Slovena have very low (mostly negatve) values of ths varable. The values for Estona an Latva fluctuate over the years. The countres wth hgh values of the large banks net nterbank assets varable are countres wth strong ncumbent banks, whereas the countres wth low values of ths varable are countres where no ncumbent banks exst. 14 As shown n Table 2, the small banks borrowng rato s close to zero n ulgara (after 1995), Estona (for most of the observe peros), Latva, Lthuana, Romana, an Slovena, ncatng that small banks o not epen n ther cret actvty on funs gathere on the nterbank market. In contrast, the Czech Republc, Hungary, Polan an Slovaka show very large negatve values of ths varable (even below -30% n the early years), suggestng that n 13 The banks that have at least 20% share n the respectve epost market an year are lste n Table A.2 n the Appenx 14 Estona, Latva, Lthuana an Slovena are newly nepenent countres whch o not have nherte nsttutons from pre-transton tme; Romanan ncumbent banks lost customer confence n the early transton pero when the frst peros of stress were observe; later on one of the major ncumbent banks went nsolvent. 15

16 these countres a large proporton of the small banks heavly epen on nterbank funs for the fnancng of ther cret actvtes. Table 1: Large banks lenng rato ulgara n.a. 0,17-0,50 0,29 0,47 0,44 0,66 0,49 Czech Republc 0,20 0,13 0,06 0,16 0,20 0,34 0,53 0,51 Estona 0,29 0,20 0,00-0,14-0,12-0,13-0,05 0,06 Hungary 0,00 0,02 0,22 0,10 0,14 0,15 0,11 0,17 Latva 0,29 0,12 0,13 0,09-0,11 0,00 0,07-0,09 Lthuana 0,04 0,04 0,00 0,01-0,04-0,03 0,09 0,09 Polan 0,15 0,10 0,10 0,09 0,07 0,08 0,10 0,12 Romana 0,60 0,35-0,27-0,61 0,08-0,07 0,07 0,06 Slovaka 0,16 0,15 0,09 0,07 0,06 0,11 0,12 0,17 Slovena -0,21 0,26 0,24 0,10-0,02-0,04-0,07-0,11 Source: Own calculatons base on ankscope an IFS Table 2: Small banks borrowng rato ulgara -0,66-0,24 1,60 1,03 0,76 0,73 0,76 0,66 Czech Republc -0,38-0,33-0,31-0,22-0,09-0,04 0,05-0,12 Estona 0,43 0,10 0,03-0,16-0,25-0,09 0,22 0,07 Hungary -0,42-0,26-0,17-0,25-0,17-0,11-0,12-0,07 Latva 0,50 1,10 1,56 1,02 0,16 0,38 1,08 0,83 Lthuana -0,10-0,17 0,04 0,27-0,05-0,03-0,06 0,00 Polan 0,05-0,11-0,19-0,08-0,19-0,11-0,04-0,06 Romana 0,37 0,18 0,11 0,52 0,33 0,32 0,30 0,37 Slovaka -0,41-0,34-0,29-0,36-0,30-0,32 0,00-0,01 Slovena 0,05 0,10 0,17 0,07 0,07 0,01 0,05 0,05 Source: Own calculatons base on ankscope an IFS To sum up, n all the countres where ncumbent banks stll have omnant poston n the market for customer eposts, except ulgara, we observe sgnfcant flows of funs through the nterbank market from the ncumbent banks to the rest of the bankng sector. In the countres where, ue to hstorcal reasons, ncumbent banks no longer exst or have lost customers confence n the early transton, banks omnatng the epost market o not channel sgnfcant amounts of funs to the rest of the bankng system. In some of the sample countres we even observe the contrary case where large banks are net recevers of funs but the magntue of ths transfer s much lower an coul be assocate wth the classcal motvaton of nterbank borrowng, namely coverng of short term lquty. 16

17 5 Comparson of banks rsk characterstcs: emprcal evence The theoretcal moel n Secton 2 erves the hypothess that nterbank borrowng can reuce the level of rsk unertakng of a bank. We test ths hypothess by emprcally estmatng the mpact of nterbank borrowng on bank rsk characterstcs. In aton we test whether banks operatng n two-ter bankng systems have lower rsk levels on average Econometrc moels To test the hypothess of the rsk effects of nterbank borrowng we use the followng Moel 1: R jt = β β β β β β + ε NIPjt + 3 X jt + 4 Y jt + 5 C j + 6 Tt jt, (12) where: R jt enotes a measure of the rsk ncurre by bank n country j n tme t; NIP jt enotes the net nterbank poston of bank n country j n tme t; X jt s a vector of control varables at the level of the nvual bank; Y jt s a vector of control varables at the level of country of operatons, an ε jt s the error term. To test the mpact of a two-ter structure of the bankng system on the level of rsk, we use the followng Moel 2: R jt = η η η η η η + ν TS jt + 3 X jt + 4 Y jt + 5 C j + 6 Tt jt, (13) where, n aton to the notatons n equaton (12), we ntrouce: TS jt as the vector of varables escrbng the type of the bankng system; ν jt as the error term. 17

18 We perform the regressons on a sample consstng only of the banks that are regare as small by the constructon of the transfer of funs varables, snce for large banks the relaton between nterbank borrowng an nterbank montorng coul be hampere by moral hazar an too-bog-to-fal concerns an thus be fferent from what our moel prects. Depenent varable We use four varables that have been wely use n the lterature to measure the rskness of a bank s busness 15 : loan loss reserves to gross loans (LLR), loan loss provsons to gross loans (LLP), net-charge offs to gross loans (NCO), an non-performng loans to gross loans (NPL) 16. The rato of loan loss reserves to gross loans expresses what proporton of the total loan portfolo has been prove for but not charge off. Loan loss reserves are an entry on the lablty se of the balance sheet an represent accumulate provsons for expecte loan losses. Assumng smlar accountng polcy an regulatons of provsonng, hgher LLR rato woul mply that banks expect losses on hgher proporton of ther loans an s thus an ncator for rsker loan portfolo of the bank 17. The rato of loan loss provsons to gross loans expresses the proporton of total loans that have been prove for urng the current pero. Loan loss provsons are expenses aganst current earnngs n the ncome statement. They represent allocatons n the current pero to the loan loss reserves an shoul reflect estmate losses for specfcally entfe loans as well as estmate probable cret losses nherent n the remaner of the portfolo at the 15 See Martn (1977) an onzales-hermosllo, et al (1996) 16 The use of more sophstcate market base measures of bank rsk, e.g. nterest rate on certfcates of eposts, bank bon prces, etc., s not possble for the sample of CEE banks snce such nstruments were not use n most of the CEE countres urng the pero we stuy. 17 Cavallo an Majnon (2001) 18

19 balance sheet ate 18. Agan, assumng smlar accountng polcy an regulatons of provsonng, hgher LLP rato mples rsker loan portfolo. The rato of net charge-offs to gross loans (NCO) llustrates the proporton of wrtten-off loan losses 19 n the amount of the gross loan portfolo. The lower the NCO rato, the lower the level of rsk unertakng of a bank, as long as the wrte-off polces are consstent across comparable banks. Snce charge-offs llustrate n current fnancal statements reflect the rsk of loans strbute n prevous balance peros, we use as epenent varables the values of the NCO rato for one year ahea (one-year lea NCO). In other wors, we regress the values of the NCO rato n pero t on the lagge values of the explanatory varables (from pero t-1). Non-performng loans to gross loans represent the proporton of mpare (oubtful) loans n the loan portfolo 20. A hgh value of ths rato ncates that a large proporton of a bank s loans have not been serve accorng to the repayment scheule. Thus, t suggests that efaults on a hgh proporton of a bank s loans are expecte 21. Followng Demsetz (1996) we use the logarthmc form of all epenent varables n the econometrcal estmatons. Explanatory varables To measure the mpact of nterbank borrowng on bank rsk levels, we nclue as an explanatory varable the net nterbank poston of a bank as measure by the rato of net nterbank assets to total assets (NIA/TA). If ths rato has negatve values, the bank borrows 18 Cavallo an Majnon (2001) 19 Net charge-offs are efne as the amount wrtten-off from loan-loss reserves mnus recoveres (see ankscope: Rato efntons ) 20 see ankscope: Rato efntons 21 We regress current values of NPL rather than leas, because most of the loans are short-term an thus potental repayment elays are alreay reflecte n the year of loan extenson. 19

20 on the nterbank market. On the other han, postve values of the rato ncate that the bank s a net prover of nterbank funs. However, an one-stage OLS estmaton of the effect of a bank s net nterbank poston on a bank s rsk may suffer uner smultanety, because as escrbe n the theoretcal moel, the nterbank poston varable wll be an outcome of the same equrum that etermnes a bank s level of rsk unertakng. To eal wth the smultanety problem, we choose to nstrument a bank s net nterbank poston (NIA/TA) wth a bank s current rato of loans to total assets (Loans/TA), an two transfer of funs varables (NIA_L an NIA_S) equal to the ratos of large banks lenng (NIA /CD ) an small bank borrowng (NIA sb /L sb ), respectvely. Each of these nstruments can be consere as exogenous wth respect to current rsk, but s correlate wth a bank s current nterbank poston. The rato of loans to total assets s closely correlate wth a bank s current nterbank poston an ncates a bank s specalzaton n cret supply. The transfer of funs varables ncate the country average nterbank postons of large an small banks an are sgnfcantly correlate wth nvual small banks nterbank poston (see Table A.3 n the Appenx for the reuce form estmatons of NIA/TA). In the estmaton of the type of the fnancal system s mpact on bank rsk (Moel 2) we nclue as regressors the varables measurng the nterbank transfer of funs, efne n Secton 4 as ncators for the exstence of a two-ter bankng system. Thus, the vector conssts of two varables measurng the transfer of funs (NIA_L an NIA_S) equal to the ratos of large banks lenng (NIA /CD ) an small bank borrowng (NIA sb /L sb ), respectvely. A hgh value of NIA_L shows that large banks have hgh net nterbank assets. A hgh value of NIA_S mples that small banks have hgh net nterbank assets. Therefore, n a two-ter bankng system, where large banks len an small banks borrow substantal amounts n the nterbank market, the values of NIA_L are hgh an those of NIA_S are low. TS jt 20

21 Several control varables are nclue n the estmatons of both econometrc moels. On the nvual bank level we ntrouce bank sze, captalzaton level, an foregn ownershp as control varables. We proxy a bank s sze by the rato of ts total assets to the mean bank s total assets n the respectve sample country. The normalzaton ams at a better comparablty across banks wth fferent countres of orgn an neutralzes the effects of exchange rate evatons. In aton, we use the square bank sze term to control for non-lnear form of the epenence between bank s sze an rsk unertakng. Captalzaton s measure by the rato of equty to total assets. Foregn ownershp s measure by a ummy varable equal to one f at least 50% of a bank s equty s owne by an nsttuton base abroa an to 0 otherwse. We nclue ths varable to account for the possblty that foregn-owne banks have better technology for assessment of cret worthness an are thus less lkely to generate nonperformng loans. On the level of country of operatons we nclue the followng macroeconomc varables as controls: nflaton, per capta DP, an the growth rate of DP 22. Inflaton s efne as the percentage change n the DP eflator. Per capta DP s use as a general nex of economc evelopment an s measure n ten thousans of US ollars. DP growth s use to measure cyclcal effects on bank rsk an s measure as the growth rate of real per capta DP. Tme an country fxe effects are ntrouce n the regressons to capture other unobserve varables Estmaton technque As mentone above, n orer to eal wth the enogenety of the net nterbank poston varable we estmate Moel 1 usng nstrumental varable panel ata approach. The estmatons are performe by generalze two-stage least squares ranom effects 23 technque. 22 Usng these varables as controls for bank rsk has been propose by Demsetz et al (1996) 23 Hausman test rejects the hypothess of fxe effects n both moel 1 an moel 2. 21

22 It estmates the precte value of NIA/TA usng the nstrumental varables escrbe above an plugs ths precte value of NIA/TA nto the structural moel. The estmatons of the mpact of the type of fnancal system on the level of bank rsk unertakng (Moel 2) are performe by stanar (one-stage) ranom effects panel ata estmaton Estmaton results Table 3 llustrates the results of the regressons of the fferent proxes for bank rsk unertakng on the measure of bank s net nterbank poston. For all measures of bank rsk the net nterbank poston of a bank as measure by the rato of net nterbank assets to total assets (nstrumente by the ratos of loans to total assets (Loans/TA) an the transfer of funs varables (NIA_L an NIA_S)) has a sgnfcant postve coeffcent. Table 3: Two-stage panel regressons of bank rsk on nterbank poston LLR LLP LNCO NPL net nterbank assets/total assets bank sze bank sze square equty/total assets foregn nflaton per capta DP DP growth const country ummes yes yes yes yes tme ummes yes yes yes yes R Observatons roups Note: Coeffcents n bol, stanar errors below coeffcents.,, ncate sgnfcance at the 10%, 5% an 1% level, respectvely. 22

23 These results ncate that a hgher share of net nterbank assets n total assets mples hgher rsk levels of the loan portfolo, as measure by the LLR, LLP, lea NCO, an NPL ratos. Therefore, net nterbank borrowng, whch necessary mples negatve net nterbank assets, s assocate wth a lower level of rsk unertakng: banks borrowng on the nterbank market have on average lower LLR, LLP, lea NCO an NPL ratos than banks fully fnance through customer eposts. The values of the coeffcents ncate that the effect s at strongest when LNCO s use as rsk proxy. The results of the estmatons of the mpact of the type of bankng system are llustrate n Table 4. The level of net nterbank assets of small banks (NIA_S) has a sgnfcant postve coeffcent n all, but the lea NCO, regresson specfcatons. Ths result mples that n systems where small banks are the provers of nterbank funs, the rsk levels of the small banks are generally hgher than n systems where small banks receve nterbank funs (have negatve net nterbank assets). Net nterbank assets of large banks (NIA_L) have negatve coeffcents n all but the lea NCO specfcatons, but these are all statstcally nsgnfcant. The nsgnfcance of the NIA_L coeffcents can be explane by the fact that snce borrowers of nterbank funs are not necessary omestcally locate banks, hgh net nterbank assets of large banks o not necessary mply that omestc small banks are nterbank borrowers (see, for example, the values of the transfer of funs varables for ulgara, an Latva, ). In general, the estmatons of the mpact of the structure of the bankng system present evence on the mportance of the source of nterbank fnancng. If nterbank funs are prove by small banks, the level of rsk unertakng s on average hgher for the whole populaton of small banks. Ths result supports the argument that free rer ssues an too-bgto-fal concerns coul hamper lenng banks ncentves to montor the borrowng banks n a system where numerous small banks prove cret to a few larger banks. 23

24 Table 4: Panel regressons of bank rsk on the type of bankng system LLR LLP LNCO NPL net nterbank assets Ls (NIA_L) net nterbank assets Ss (NIA_S) bank sze bank sze square equty/total assets foregn nflaton per capta DP DP growth const country ummes yes yes yes yes tme ummes yes yes yes yes R Observatons roups Note: Coeffcents n bol, stanar errors below coeffcents.,, ncate sgnfcance at the 10%, 5% an 1% level, respectvely. To summarze, small banks n two-ter systems, where small banks have low net nterbank assets, have lower levels of rsk than small banks n one-ter systems. Control varables ank sze an bank sze square have nsgnfcant coeffcents n all regresson specfcatons, ncatng that wthn the sample of small banks, the relatve sze of a bank s not a etermnant of ts levels of rsk unertakng. Equty to total assets has a sgnfcant negatve mpact on LLR an LLP (only n the estmaton of the mpact of the nterbank poston), whch s complant wth the theoretcal noton that banks wth hgher proporton of own captal nvest n less rsky projects. The coeffcents of equty to total assets are nsgnfcant n the specfcatons usng lea NCO an NPL as proxy for rsk. The foregn ownershp ummy has sgnfcant negatve coeffcents n all regresson specfcatons, presentng evence for lower levels of rsk unertakng by banks owne by foregn enttes. 24

25 Ths result supports smlar fnngs n the lterature on foregn bank entry n transton an evelopng countres (Clarke et al, 2003). Concernng the macroeconomc varables hgher nflaton s assocate wth hgher levels of LLR an LLP, ncatng that banks reckon wth hgher rsk of ther portfolo n hgh nflaton peros. The negatve sgnfcant coeffcents n the NPL regressons ncate that the e facto loan efaults are lower n hgh nflaton tmes, whch s an ntutve result as long as loans are contracte to pay a fxe nterest. Per capta DP has a negatve but nsgnfcant coeffcent n all regresson specfcatons. DP growth sgnfcantly reuces bank rsk as measure by LLR, LLP, an NPL ncatng the cyclcal mpact on bank rsk. To summarze, we fn emprcal evence supportng the hypothess that nterbank borrowng s assocate wth lower levels of rsk unertakng, whch s an ncator that nterbank-borrowng banks are beng montore. Ths result mples that nterbank-lenng banks feel responsble for losses they ncur on nterbank transactons. Furthermore, the estmates of the mpact of the type of the fnancal ntermeaton system ncate that n two-ter systems banks unertake less rsk on average. Such a result s consstent wth our noton that nterbank montorng s mostly feasble when few nsttutons len large amounts. 6 Concluson In ths paper we prove evence on the rsk allevatng role of long-term nterbank borrowng n the absence of too-bg-to-fal concerns. In the theoretcal part we present a moel base on nformaton asymmetres an moral hazar. In the settng of the moel montorng the nvestment strategy of the banks s necessary to prevent the cret market from a collapse, but snce t s costly only the large 25

26 cretors of a bank can perform t. Therefore, banks that are fnance by large nterbank eposts are montore, whereas those that are fully fnance by customer eposts are not. As a result, the moel shows that banks whch borrow n the nterbank market engage n less rsky lenng actvtes than banks that fnance themselves preomnantly n the epost market. The emprcal part conssts of two steps. Frst, we propose a methoology for measurng the flow of funs from the large banks to the rest of the bankng system. We use ths methoology to etermne those countres where the nterbank flow of funs s hgh enough to clam that the bankng system conssts of two ters of banks. The banks from the frst ter (large ncumbent banks) omnate the epost market but are very nactve n the loan market wth prvate borrowers, whle the banks from the secon ter engage n lenng but have only small shares of the epost market. Secon, we prove econometrc evence about the exstence of rsk allevatng effects of nterbank borrowng. We test the mpact of nterbank borrowng an the exstence of a twoter bankng structure on fferent parameters use as proxes for bank s level of rsk unertakng. The results show a sgnfcant effect of both nterbank borrowng an the twoter structure. anks whch borrow on the nterbank market unertake less rsky projects. Furthermore, small banks n countres wth a two-ter structure of the bankng system are characterze by lower rsk levels than small banks n one-ter structure countres. In general, large ncumbent banks aopt montorng functons over those of the new entrant banks, whch fnance themselves through nterbank eposts. In an envronment of neffcent bankng regulaton an unversfe portfolos of the new entrant banks, montorng by the large banks coul play essental role for the establshment of pruent nvestment behavor. 26

27 Lterature: Acharya, V. V., Hasan, I. an A. Sauners The Effects of Focus an Dversfcaton on ank Proftablty an Loan Qualty, mmeo Altman, E. I Fnancal Ratos, Dscrmnant Analyss an the Precton of Corporate ankrupcy, Journal of Fnance 23(4) Anerson, R. an C. Kegels Transton ankng: Fnancal Development of Central an Eastern Europe, Oxfor Unversty Press erger, A. N., Davs, S. M. an M. J. Flannery Comparng Market an Supervsory Assessment of ank Performance: Who Knows What When? Fnance an Economcs Dscusson Seres , Feeral Reserve oar of overnors hattacharya, S., an D. ale Preference Shocks, Lquty an Central ank Polcy. New Approaches to Monetary Economcs. Cambrge: Cambrge Unversty Press llett, M. T. an J. A. arfnkel Fnancal Flexblty an the Cost of External Fnance for U.S. ank Holng Companes. Journal of Money, Cret an ankng, forthcomng llet, M. T., arfnkel, J. A. An E. S. O Neal The Cost of Market versus Regulatory Dscplne n ankng. Journal of Fnancal Economcs 48: onn, J Fnancal Intermeaton n Southeast Europe: ankng on the alkans, Venna Insttute for Internatonal Economc Stues, mmeo onn, J., Mzse, K., Szekely, I., an P. Wachtel ankng n Transton Economes: Developng Market Orente Sectors n Eastern Europe, Insttute for EastWest Stues onn, J. an P. Wachtel Fnancal Sector Development n Transton Economes: Lessons from the Frst Decae, OFIT Dscusson Paper 9/2002 Calveras, A ank Specalzaton: The Role of the Interbank Market. mmeo Cavallo, M. an. Majnon Do anks Provson for a Loans n oo Tmes? Emprcal Evence an Polcy Implcatons, Worl ank Workng Papers, No 2619 Clarke,., Cull, R., Pera, M. S., an S. M. Sanchez Foregn ank Entry: Experence, Implcatons for Developng Economes, an Agena for Further Research. The Worl ank Research Observer, vol. 18 (1) Cole, D., an. Slae ulng a Moern Fnancal System: The Inonesan Experence. Cambrge Unversty Press 27

28 De ant, O. an E. P. Davs A Cross-Country Comparson of Market Structures n European ankng, EC Workng Paper Demsetz, R. S., Seenberg, M. R., an P.E. Strahan anks wth Somethng to Lose: The Dscplnary Role of Franchse Value. FRNY Economc Revew Demesetz, R.S. an P. E. Strahan Dversfcaton, Sze, an Rsk at ank Holng Companes, Journal of Money, Cret an ankng De Young, R., Flannery, M. J., Lang, W. W. An S. M. Sorescu The Informatonal Avantage of Specalze Montors: The Case of ank Examners. Workng Paper, Unversty of Flora Damon, D. W Fnancal Intermeaton an Delegate Montorng. Revew of Economc Stues LI: Damon, D. W Fnancal Intermeaton as Delegate Montorng: A Smple Example. Feeral Reserve ank of Rchmon, Volume 82/3 Summer Dttus, P. an S. Prowse Corporate Control n Central Europe an Russa: Shoul anks Own Shares? Worl ank Frexas, X. an J. C. Rochet Mcroeconomcs of ankng, MIT Furfne, C. H Interbank Exposures: Quantfyng the Rsk of Contagon. ank for Internatonal Settlements Workng Paper 70 Furfne, C. H anks as Montors of Other anks: Evence from the Overnght Feeral Funs Market. The Journal of usness 74(1): almes, S. an M. C. Manzano El Mercao Interbancaro e Depostos y las Entaes e Creto. oletn Economco el anco e Espana Febrero 1995 Hellwg, M Lquty Provson, ankng, an the Allocaton of Interest Rate Rsk. European Economc Revew 38(7): Ho, S. Y., an A. Sauners A Mcro Moel of the Feeral Funs Market. The Journal of Fnance 49(3): Keeley, M. C Depost Insurance, Rsk an Market Power n ankng. The Amercan Economc Revew: Lelan, H. E., an D. H. Pyle Informatonal Asymmetres, Fnancal Structure an Fnancal Intermeaton. The Journal of Fnance 32:

29 Martn, D Early Warnngs of ank Falure. A Logt Regresson Approach. Journal of ankng an Fnance 1: Mller, J. an S. Petranov The Fnancal System n the ulgaran Economy, ulgaran Natonal ank Peek, J., Rosengren, E. S. an. M.. Tootell Is ank Supervson Central to Central ankng. Quarterly Journal of Economcs 114(2): Rochet, J-C., an J. Trole Interbank Lenng an Systemc Rsk. Journal of Money, Cret an ankng 28: Snkey, J. F A Multvarate Analyss of the Characterstcs of Problem anks, Journal of Fnance 30 (1): Transton Report European ank for Recunstructon an Development 29