ARE CAPITAL BUFFERS PRO-CYCLICAL?

Transcription

1 ARE CAPITAL BUFFERS PRO-CYCLICAL? Juan Ayuso (*) Danel Pérez (*) Jesús Saurna (*) Banco de España Ths verson: Aprl 2002 (*) We hank O. Bover, F. Resoy and F. Vargas for useful commens. The vews conaned n hs paper are hose of s auhors and do no necessarly reflec he poson of he Banco de España.

2 1. Inroducon Effors whn he Basel Commee on Bankng Supervson o updae he 1988 Basel Accord have almos fnalsed a new accord on banks capal adequacy, whch s already known as Basel II. There reman, however, some ousandng ssues and deals, among whch ha of pro-cyclcaly has, perhaps, produced mos debae n he leraure 1. The concep of pro-cyclcaly, when appled o he new capal requremens, may n prncple be a lle confusng. As s well known, one of he prmary ams of he new accord s o lnk capal requremens more closely o rsks. Accordngly, n a downurn, for nsance, when rsks are more lkely o maeralse, capal requremens mgh ncrease. Thus, capal requremens and oupu growh wll move n oppose drecons. Bu f capal requremens ncrease, banks would have o reduce her loans and he subsequen cred squeeze would add o he downurn. Capal requremens are herefore sad o be (lkely o be) pro-cyclcal because hey mgh amplfy he flucuaons of he busness cycle. I has been also argued, however, ha f nernal rsk managemen models ake properly no accoun he way defaul probables change hroughou he busness cycle, he effecs on cred and herefore, on oupu should no be oversaed. Whle argumens hghlghng or mnmsng he acual relevance of he pro-cyclcaly problem have prolferaed, he relaed emprcal evdence s scan. Admedly, he emprcal leraure on he mpac of capal requremens on bank behavour s exensve, hough manly confned o he US case 2. Papers have deal wh ssues such as wheher he nroducon of mnmum capal requremens leads banks o hold hgher capal; he mpac of capal requremens on rsk-akng, compeveness and a level playng feld; or wheher capal requremens creae cred crunches affecng he real economy. Neverheless, as far as we know, hese papers have no analysed he cyclcal behavour of capal requremens, perhaps because he curren Basel capal accord es capal requremens less closely o banks' capal rsk. Agans hs background, he am of hs paper s o provde some fresh emprcal evdence whch may prove useful n he debae abou he pro-cyclcaly of he new capal accord. In parcular, we have noced ha mos argumens n hs debae relae o he cyclcaly of he capal requremens, hus gnorng he fac ha only a few banks hold jus he requred capal, whle mos keep capal buffers whch, n some cases, are 1 See, among ohers, BCBS (2001), Boro e al (2001), Danelsson e al (2001), DNB (2001) and ECB (2001). 2 See, for example, Berger (1995), Furfne (2000) and Kwan and Esenbes (1997). Jackson e al (1999) offers a dealed revew. 1

3 que sgnfcan 3. The behavour of hese buffers and, n parcular, her relaonshp o he busness cycle may hus be an mporan pece of nformaon. A posve relaonshp would mean ha banks rebuld her capal durng upurns n order o face a lkely ncrease n requremens durng he nex downurn. Ths mgh offse, a leas parally, he pro-cyclcaly of he requremens. A negave relaonshp, however, would have less encouragng mplcaons n hs respec. In hs paper, we analyse he relaonshp beween he Spansh busness cycle and he capal buffers (curren capal over he mnmum capal requremens) held by Spansh commercal and savngs banks. Usng sandard economerc panel daa echnques, we buld an ncomplee panel of Spansh nsuons from 1986 o 2000 hus coverng a complee cycle and esmae an equaon for he behavour of capal buffers ha ncludes an ndcaor of he busness cycle. Admedly, focusng on a sngle counry mgh reduce he scope of our analyss. Whle daa avalably (n parcular regardng ndvdual capal requremens) prevens us from a more general approach, s worh menonng ha he busness cycle seems a leas durng he perod consdered o have been relavely pronounced n Span, whch mgh render he Spansh case an neresng case sudy. Moreover, Spansh banks are hghly compeve and effcen, whch reduces he probably of dosyncrac facors basng he resuls. Our fndngs are farly robus and que unequvocal. Afer conrollng for oher poenal deermnans of he surplus capal (cos of capal, rsk profle of he nsuon, sze ) we fnd a (sgnfcan) negave relaonshp beween he busness cycle and he capal buffers ha Spansh nsuons held hroughou he perod analysed. Ths relaonshp s, moreover, asymmerc, beng closer durng downurns han durng upurns. From a quanave sandpon, an ncrease of 1 percenage pon n GDP growh mgh reduce capal buffers by 17%. I s mporan o noe, however, ha our resuls relae o he behavour of banks under Basel I. As canno be oally dscarded ha Basel II may cause a srucural change, he polcy mplcaons of our resuls have o be aken wh due cauon. The res of he paper s srucured as follows. The man equaon o be esmaed and he daa used n he paper are revewed n secon 2. The hrd secon shows he man resuls of he basc economerc analyss, whle some exensons are consdered n secon 4, provdng more nformaon on he pro-cyclcaly of capal buffers. Fnally, he conclusons of he paper are summarsed n he las secon. 3 Papers have usually focused on he deermnans of he capal rao or s rae of growh. See he references n Jackson e al (1999). 2

4 2. The emprcal equaon and he daa Among ohers, Berger e al (1995) argue ha banks may wsh o hold a capal buffer n order o be able o explo unexpeced nvesmen opporunes or o cushon he mpac of unexpeced shocks. In addon, hs buffer may be larger f regulaons enal a capal requremen and he penales for no complyng are hgh (nondsrbuon of dvdends, resrcons on busness expanson, ec.) and/or he coss of rasng fresh capal are szeable. A he same me, however, keepng a capal buffer nvolves a drec cos, as has o be remuneraed. The level of surplus capal over he regulaory mnmum ha each nsuon wshes o manan should depend on he oucome of he foregong rade-off 4. In hs ven, Esrella (2001) analyses a dynamc model where he opmum capal arses from he neracon beween he coss of holdng capal, on one hand, and he coss of falure and of adjusng he level of capal, on he oher. Oher papers (Keeley (1990) and Salas and Saurna (2002a)) underlne he relaonshp beween he level of capal an nsuon holds and s rsk profle n a conex of fallng franchse value arbuable o heghenng compeon. In hs respec, an nsuon wh a hgh-rsk profle may have an ncenve o manan a level of capal closer o he regulaory mnmum han an nsuon pursung a more conservave polcy. All n all, seems reasonable o model he deermnans of he capal buffer of nsuon n perod (BUF ) as 5 : BUF = β 0BUF + η + ε, 1 = 1, + β ROE 1 2,..., N + β NPL 2 + β BIG ( number of banks), 3 + β SMA 4 = 1, + β GDPG 2,..., T 5 (1) Lookng a he rgh-hand sde of equaon (1), s worh nong, frs, ha capal s cosly o rase (or o amorse). The correspondng adjusmen coss n aanng he desred level of capal jusfy ncluson of he endogenous varable wh a one-perod lag, her expeced sgn beng posve. Bu holdng an excess of capal also enals a drec cos n ha has o be remuneraed. Ths cos s approxmaed here usng each nsuon s ROE (reurn on equy). The expeced sgn for hs varable s hus negave. 4 Lke non-fnancal corporaons, he desred level of capal n banks s also nfluenced by oher elemens such as he ax benef mplc n deb -as opposed o equy-, or he poenal use of decsons on capal o convey sgnals o he marke n a framework of asymmerc nformaon. See Berger e al. (1995) and references heren. 5 In he annex, a very smple model s oulned o provde some addonal movaon and raonale for he varables whch, deally a leas, should appear on he rgh-hand sde of an explanaory equaon for capal buffers. 3

5 Conrollng explcly for he rsk profle of each nsuon s he am of he varable NPL ha measures he non-performng loans rao (rao of non-performng loans o oal loans and creds). Ths s an ex pos measuremen of he rsks assumed by he nsuon and, herefore, s expeced sgn s negave. In addon, here mgh be an dosyncrac me-nvaran componen, whch would be capured, however, by η. BIG and SMA are conrol varables ncluded o deec dfferences n he buffer accordng o he sze of each nsuon. In parcular, BIG (SMA) s a dummy varable akng value 1 for banks n he hghes (lowes) decle. In prncple, bg banks mgh be hough o keep relavely lower buffers. Accordng o he oo-bg-o-fal" hypohess, he larges nsuons mgh be nclned o hold lower capal buffers snce hey know ha n he even of dffcules hey wll receve suppor from he regulaor. Small banks, on he conrary, mgh have o hold relavely larger buffers f hey face more dffcules appng capal markes. Havng conrolled for all hese deermnans of he capal buffer, GDP growh (GDPG) s ncluded n order o deermne wheher he busness cycle has an addonal effec on he capal buffer held by nsuons. The sgnfcance, sgn and magnude of β 5 wll allow us o answer he man quesons ha are he core of hs paper. Fnally, η s an unobservable varable ha capures he dosyncrac feaures of each nsuon ha are consan over me bu vary from nsuon o nsuon -for example, he greaer or lesser rsk averson of bank managers or her more or less conservave polcy wh respec o capal buffers- and ε s a random shock. Our daa are drawn from he fnancal saemens regularly and oblgaorly sen by he nsuons o he Banco de España. Consoldaed fgures have been used (excep, of course, for nsuons ha do no consoldae her daa and do no belong o a consoldable group), as capal requremens are mposed a he consoldaed group level. The scope of he rsks conaned n consoldaed balance shees s, moreover, fuller as nformaon abou Spansh bankng subsdares operang ousde Span s ncluded. The way we measure he capal buffer deserves some addonal commens. In hs respec, s worh brefly revewng he mos sgnfcan regulaory changes o he Spansh capal adequacy rao snce Tha year saw he nroducon n Span of a capal adequacy rao replacng he prevous legally requred rao (equy/lables), whose usefulness was more han quesonable. The 1985 capal adequacy rao was calculaed as he hgher of a gearng rao (equy/oal asses) of 5% and a rsk asse rao under whch a dfferen level of capal was requred (from 0.25% o 35%), dependng on he rsk assocaed wh he dfferen headngs. 4

6 In 1993 he Spansh regulaon was adaped o he Communy drecve whch, broadly speakng, adhered o he 1988 Basel capal accord. The regulaon has no changed fundamenally snce, and s based on he defnon of he componens of he numeraor and denomnaor of he capal adequacy rao, namely capal and capal requremens. Capal should cover a leas 8% of requremens. The new 1993 regulaory framework promped somehng of a declne n capal requremens, wh a subsequen ncrease n he capal buffer gven ha he new regulaon was less demandng han he 1985 regulaon 6. In he lgh of he regulaon n place before 1993, BUF has been defned as he nsuon's capal less he requremens o whch was subjec, dvded by he requremens, hus crcumvenng he drawback whereby, before 1993, here was no mnmum rao appled across he board o all nsuons. From among all cred nsuons a sub-se has been seleced made up of domesc nsuons (ncludng foregn nsuons' subsdares operang n Span) and savngs banks. Foregn bank branches and cred co-operaves whose relave sgnfcance (n erms of busness volume) s scan have been excluded. The daa are annual and span he perod from 1986 o In hs manner a full cycle of he Spansh economy s ncluded, a pon of parcular mporance gven ha he am of hs paper s, as menoned, o analyse wheher here s a relaonshp beween he busness cycle and he capal buffer held by nsuons. Our panel s ncomplee snce new nsuons have sared o operae durng he perod consdered whle ohers have ceased o work. Moreover, he mpac of bank mergers durng he perod has also o be aken no accoun. Mergers pose an obsacle o calculang averages and, parcularly, growh raes. To overcome hs drawback so ha he leas number of observaons possble s los, has been decded o arfcally recreae he merger a perod n advance. Tha s o say, f wo bankng nsuons merge a, for he purposes solely of calculang averages and growh raes, he resulng nsuon s consdered o have already exsed a -1, reconsrucng on he bass of he daa from he ndvdual nsuons nvolved n he merger. A smlar problem arses for nsuons whch, havng belonged a -1 o a consoldaed group, leave such group a. To calculae boh he averages of ceran varables and her growh raes, he fgure a -1 s obaned from her ndvdually repored fnancal saemens. Under hese premses and afer elmnang a seres of nsuons wh exremely aypcal daa (due essenally o he specfc naure of her busness), an ncomplee 6 Noneheless, he Spansh 1993 regulaon connues o be more demandng han ha currenly n force n Basel due manly o he fac ha recognon of unrealsed capal gans s no permed and general and sascal provsons are no consdered as Ter 2 capal. 5

7 panel has been obaned comprsng up o 142 nsuons over a perod of up o 15 years, oallng 1309 observaons. Table 1 shows he descrpve sascs of dfferen varables, whle Char 1 plos he aggregae course of he aggregaed capal buffer and s relaonshp o he busness cycle, he correlaon coeffcen beween boh seres beng [Table 1] [Char 1] 3. Economerc resuls Frs of all, s worh nong ha varables n he emprcal equaon (1) are defned n levels, whle some (such as NPL) are lkely o be correlaed wh η. As usual n panel daa analyss, we proceed o ransform (1) no frs dfferences, o enable unbased esmaes o be obaned. On he oher hand, as he lagged endogenous varable s ncluded among he regressors and oher explanaory varables are lkely o be endogenous, an esmaon procedure based on he generalsed mehod of momens (GMM) seems he mos approprae 7. In parcular, he nsrumens chosen for he lagged endogenous, NPL and ROE, are wo-o-four-perod lags of he same varables. These lags have been chosen o avod correlaon wh he error erm ε (whch now appears n frs dfferences) whle mnmsng, a he same me, he number of observaons los. The varables of sze and busness cycle are consdered o be exogenous and herefore used as her own nsrumens. Table 2, column 1, shows he man resuls of he esmaon of equaon (1). Regardng he sgnfcance and sgn of he coeffcen of oupu growh, we fnd ha here s a clearly sgnfcan (a 1%) negave relaonshp beween he capal buffer and he phase of he cycle. Accordngly, n he case of Span, capal requremens barely sensve o he cycle (.e. he 1988 Basle Accord) have ranslaed no relavely procyclcal buffers. [Table 2] The long-erm sem-elascy of he buffer relave o GDP growh, calculaed a he sample buffer average, s.17. Tha s, an ncrease of a percenage pon n GDP growh reduces he long-erm relave buffer by around 17%. Gven ha he average sample of he relave buffer has, n he sample consdered, been around 40% (n recen years, however, s around 25%), and he usual magnude of he changes n he rae of growh of GDP, he mpac of he cycle, despe beng very sgnfcan, seems o be 7 We have used he DPD package (Arellano and Bond (1991 and 1988)), he GMM esmaor of whch s specally desgned o oban unbased and effcen esmaes n dynamc models wh lagged endogenous varables as regressors. 6

8 moderae n quanave erms. Agan, s far o noe ha he resuls of he foregong esmaon canno be exrapolaed mechancally no he fuure, as he mplemenaon of Basel II may mply a srucural change. The remanng parameer esmaes provde oher neresng resuls. Thus, some relevan perssence n he capal buffer s deeced, whch, as expeced, reveals he exsence of non-neglgble shor-erm adjusmen coss. The capal cos has, as mgh be expeced, a sgnfcan negave mpac on he surplus capal ha he nsuons wsh o manan 8. Moreover, banks whch, accordng o he proxy chosen, have a more conservave profle, end o hold hgher buffers o mee poenal adverse shocks 9. Kwan and Esenbes (1997) also fnd ha he defaul rao and he capal rao are negavely relaed. And fnally, he sgns of he dummy varables BIG and SMA are, respecvely, conssen wh he oo-bg-o-fal hypohess and he relavely greaer dffcules for small banks o draw on capal markes. Neverheless, boh varables are only margnally sgnfcan (he p-values of he correspondng ess beng.15 and.12). The equaon, on he oher hand, passes whou any major problem he sandard goodness-of-f ess. Thus, all varables have he expeced sgn and mos of hem are sgnfcan even a 1%, and here s sgnfcan negave frs-order auocorrelaon n he resduals (m 1 sasc) and nl second-order correlaon (m 2 ), as should be he case f he error erm (n levels) s whe nose. The Sargan es for valdy of he nsrumens used s also fully sasfacory, showng a p-value of The remanng columns n Table 2 provde some furher esng on he robusness of our resuls. Thus, mgh be argued ha capal buffers are mananed no only o face conemporary unexpeced shocks bu also o cover problem loans n fuure perods 10. A smple way of conrollng for hese effecs s o use fuure values of such varable as nsrumens. The second column of Table 2 shows ha he man resuls n column 1 are no alered (eher as regards her sgn or sgnfcance) when fuure values of nonperformng loans raos are used as nsrumens. We have also esed wheher our resuls mgh be nfluenced by he fac ha he nonperformng loans rao s a markedly cyclcal varable. I mgh be argued ha hs varable could nfluence he sgn and sgnfcance of GDP growh. The las column of Table 2 shows ha f non-performng loans are excluded, GDP growh s sll 8 Berger (1995) fnds a sgnfcan posve relaonshp beween ROE and he rao of capal o oal asses (whou weghng asses by her rsk). Our resuls show a sgnfcan negave relaonshp beween ROE and surplus capal above he regulaory mnmum. The regulaory mnmum capal s calculaed by weghng he asses (and off-balance-shee ransacons) accordng o her capal requremen. Boh elemens may explan he dfferences n he resuls. 9 Somemes (Wall and Peerson (1995)) unexpeced shocks are approxmaed by means of he sandard devaon of he rae of reurn n prevous perods (4 years for nsance). The problem wh hs approach s ha a sandard devaon calculaed on he bass of only four observaons may no be very sgnfcan. For us has he addonal dsadvanage ha nvolves losng he frs four years of nformaon for each nsuon, subsanally reducng he sample sze. 7

9 sgnfcanly negave. The change (.e. bas) n he pon esmae of he parameer of GDPG reveals, moreover, ha a relevan varable has now been omed from he equaon. 11 All n all, we can conclude from Table 2 ha he capal buffer held by Spansh nsuons has behaved pro-cyclcally over he las 15 years. Also, he capal buffer s found o depend, farly robusly, on he rsk profle of he nsuon, he cos of holdng such a surplus and, o a lesser exen, he nsuon s sze. 4. Capal buffer cyclcaly: some addonal resuls In hs secon, we nclude some exensons of model [1] ha provde some furher evdence relaed o he capal buffer pro-cyclcaly found n he prevous secon. Table 3 summarses our man fndngs. [Table 3] Frs, we have red a dfferen measure of he busness cycle, whch akes no accoun he possbly of non-consan poenal oupu growh. The frs column n Table 3 shows ha he pro-cyclcaly remans f GDP growh s replaced wh he oupu gap (OUTGAP), obaned afer applyng a sandard Hodrck-Presco fler. The long-erm sem-elascy falls somewha (12%) whle he oher properes of he model reman unchanged. Nex, we also nvesgaed wheher he pro-cyclcaly of capal buffers could depend on specfc feaures of he banks. In hs respec, we frs red banks sze, neracng GDPG and he dummy varables BIG and SMALL. As can be seen n he second column n Table 3, we dd no fnd any meanngful dfference n he behavour of bg or small banks. Anoher neresng feaure ha mgh affec he cyclcaly of capal buffers s he ownershp srucure of he dfferen nsuons. In parcular, he sample analysed n hs paper s made up of commercal and savngs banks. The former are all n he hands of prvae shareholders (concenraed o a greaer or lesser degree dependng on each bank) whle he he governance of savngs banks s shared among represenaves of several sakeholder groups, publc auhores (from local and regonal 10 Insofar as fuure problem loans also arse from losses due o unexpeced shocks, hs deermnan of he buffer would also be conrolled for when ncludng NPL. 11 We also performed oher robusness ess ha are no repored here for he sake of concseness. I mgh be worh menonng, however, ha we nroduced dummy varables o conrol for he regulaory change ha occurred n 1993, whou fndng any sgnfcan effec on he conclusons drawn from he frs column of Table 2. 8

10 governmen), he foundng eny, deposors and workers. There s exensve emprcal leraure on he mpac ha dfferen ownershp srucures and dsnc corporae governance arrangemens may have on he rsk profle of nsuons 12. We have, herefore, neraced GDP growh and a dummy varable (COM) akng a value equal o one when he nsuon s a commercal bank and o zero oherwse (.e. when s a savngs bank). As shown n he hrd column of Table 3, hs varable s only margnally sgnfcan, alhough has a posve sgn, meanng ha capal buffers n commercal banks are, f anyhng, less pro-cyclcal han n savngs banks. The remanng properes of he model are no affeced n any meanngful way. Fnally, we have esed wheher he pro-cyclcaly we have found s symmerc,.e. wheher operaes n he same way durng upurns and downurns. In parcular, we have added a new varable o he rgh-hand sde of model (1): he absolue value of he dfference beween he oupu growh and he average oupu growh n he sample. As shown n he las column n Table 3, hs varable s negave and sascally sgnfcan (even a 1%). Accordngly, when oupu growh s above s average, buffers dmnsh proporonally more han hey ncrease when growh s below s average. The long-erm sem-elascy durng upurns ncreases o 29%, sandng a 12% durng downurns. 5. Conclusons The desgn of a new capal accord (Basel II) has promped an neresng debae among regulaors, supervsors, academcs and praconers. The ssue of he poenal pro-cyclcaly of he new capal requremens s currenly playng an mporan role n he debae. Whle mos argumens abou he cyclcaly of he new agreemen are of a purely heorecal naure and are cenred on he capal requremens hemselves, hs paper ams o provde some emprcal evdence and focus on he behavour of he capal buffers ha mos banks hold above he mnmum requred by domesc regulaons. The cyclcaly of hese buffers mgh offse or add o he poenal cyclcaly of he requremens. Usng annual daa on Spansh banks from 1986 o 2000, we have bul an ncomplee panel of 1309 observaons and esmaed an emprcal equaon o explan how capal buffers have behaved n ha perod. Afer conrollng for oher poenal deermnans of he excess of capal cos of capal, rsk profle of he bank, adjusmen coss, sze and unobservable dosyncrac feaures we have found a farly robus and sgnfcan negave relaonshp beween he capal buffers and he busness cycle. Neverheless, 12 See, for example, Esy (1997), Goron and Rosen (1995), Saunders e al (1990) or, regardng he Spansh case, Salas and Saurna (2002b). 9

11 n quanave erms, he pro-cyclcaly effec s moderae: an ncrease of 1 percenage pon n GDP growh mgh reduce he buffer by 17%. These resuls are, of course, obaned under he capal accord sll n place (Basel I) and, herefore, he possbly of a srucural change followng he mplemenaon of Basel II canno be compleely ruled ou. Wh due cauon, our fndngs advse, frs, a close monorng of he behavour of banks own resources durng he expansonary sages of he busness cycle o preven poenal negave effecs on solvency should a sudden cyclcal correcon occur. Bu hey also suppor a more encouragng vew by llusrang ha, despe hs markedly cyclcal behavour of capal buffers, even a he dephs of recessons banks have managed o keep farly safe levels of capal. 10

12 Annex In hs annex we buld a very smple and sylsed model o provde some (addonal) raonale for he emprcal equaon proposed n Secon 2 and esmaed n Secon 3. We sar by assumng ha he followng equaon descrbes he dynamcs for he capal sock of a represenave sngle bank: K + I = K 1 where K represens he capal a he end of perod, and I represens sock ssues or repurchases (wh a negave sgn), plus reaned profs. Thus, capal s no known unl he end of perod. Holdng capal has a cos for banks, as has o be remuneraed 13. Ye banks also face oher sor of coss ha are negavely relaed o capal. In parcular, banks face coss of falure, whch are lower he hgher he ne worh of he bank. Moreover, mos banks are also subjec o compulsory capal requremens. Includng hem as an addonal cos of (no havng enough) capal nsead of addng a consran o he opmsaon problem s a useful opon o capure he way hey operae 14. Fnally, seems sensble o assume ha adjusng he capal level enals adjusmen coss. 15 In he smples possble approach, all hese coss mgh be gahered n he followng equaon: C = 2 ( α γ ) K + 1/ 2) δ I ( whereα represens he cos of remunerang he capal, γ represens he coss of falure and/or he penales for no complyng wh he regulaory mnmum, δ reflecs he exsence of adjusmen coss. Boh, lneary (regardng he frs wo groups of coss) and symmery (as o adjusmen coss) are assumed. In hs seng, he represenave sngle bank mnmses s neremporal coss by solvng he followng problem: Mn E + 0 { I } 0 = β C + 13 Myers and Majluf (1984), for nsance, offer a heorecal model supporng he vew ha capal s more cosly han alernave lables. 14 Before lms are reached, supervsory auhores usually place some resrcons on he acvy of he bank. 15 See, for nsance, Wner (1994). 11

13 12 s.. ( ) 2 2) 1/ ( I K C δ γ α + = I K K + = 1 where he FOC regardng C can be wren as follows: ( ) = + + = 0 1 E I α γ β δ and herefore: ( ) = = ) ( E K K E α γ β δ The regulaory mnmum capal may now be subraced from boh sdes and he expeced capal replaced by he observed capal plus an expecaon error o oban: ( ) ( ) = + = = E E K K K K ε α β δ γ β δ Accordngly, he followng varables have o be ncluded on he rgh-hand sde of an explanaory equaon for he capal buffer: he frs lag of he dependen varable, whch should have a posve sgn. varables proxyng he coss of remunerang capal, whch should have a negave sgn and varables proxyng he expeced bank falure coss, whch are lkely o be closely lnked boh o he bank s aude owards rsk and o he bank s dmenson.

14 References Arellano, M. and S. Bond (1991): Some es of specfcaon for panel daa: Mone Carlo evdence and applcaon o employmen equaons. Revew of Economc Sudes, 58 (1991), pp Arellano, M. and S. Bond (1988): Dynamc panel daa esmaon usng DPD- A gude for users. Workng Paper, 88/15, The Insue for Fscal Sudes. Basel Commee on Bankng Supervson (2001): Revew of procyclcaly. Research Task Force. Mmeo. Berger, A.N. and G.F.Udell (1994): Dd rsk-based capal allocae bank cred and cause a cred crunch n he Uned Saes?. Journal of Money, Cred, and Bankng, vol. 26, n 3, pp Berger, A.N. (1995): The relaonshp beween capal and earnngs n bankng. Journal of Money, Cred, and Bankng, vol. 27, n 2. Berger, A.N., R.J. Herrng and G.P. Szegö (1995): The role of capal n fnancal nsuons. Journal of Bankng & Fnance, 19, pp Boro, C., C.Furfne and P. Lowe (2001): Procyclcaly of he fnancal sysem and fnancal sably: ssues and polcy opons. BIS papers, 1. pp Crocke, A. (1997): The heory and pracce of fnancal sably. Essays n nernaonal fnance, 203, Aprl. Prnceon Unversy. Daníelsson, J., P. Embrechs, C. Goodhar, C. Keang, F. Muennch, O. Renaul and H. Song Shn. (2001): An Academc Response o Basel II. Specal Paper 130. Fnancal Markes Group, London School of Economcs. DNB (2001): Towards a new Basel Capal Accord. Quarerly Bullen, March, pp ECB. Monhly Bullen, May (2001): The new capal adequacy regme-he ECB perspecve, pp Esrella, A. (2001): The cyclcal behavor of opmal bank capal. Mmeo. Esy, B.C. (1997): Organzaonal form and rsk akng n he savngs and loan ndusry. Journal of Fnancal Economcs, 44, pp

15 Fernández de Ls, S., J. Marínez, and J. Saurna (2000):"Cred growh, problem loans and cred rsk provsonng n Span". Workng Paper Servco de Esudos. Banco de España. Furfne, C (2000): Evdence on he response of US banks o changes n capal requremens. BIS workng papers, 88. Goron, G. and R. Rosen. (1995): Corporae Conrol, Porfolo Choce, and he Declne of Bankng. The Journal of Fnance. vol. L, n 5, pp Jackson, P. e al (1999): Capal requremens and bank behavour: he mpac of he Basel accord. Basel Commee on Bankng Supervson, Workng paper 1, Aprl. Keeley, M. C., (1990): Depos nsurance, rsk, and marke power n bankng, Amercan Economc Revew, 5, pp Kwan, S. and R.A. Esenbes (1997): Bank rsk, capalzaon, and operang effcency. Journal of Fnancal Servces Research, 12, pp Myers, S. C. and N.S. Majluf (1984): Corporae fnancng and nvesmen decsons when frms have nformaon ha nvesors do no have. Journal of Fnancal Economcs, 13, pp Salas V., and J. Saurna (2002a): Deregulaon, marke power and rsk akng n Spansh banks. Forhcomng n European Economc Revew. Salas V., and J. Saurna (2002b): Cred rsk n wo nsuonal regmes: Spansh commercal and savngs banks. Forhcomng n Journal of Fnancal Servces Research. Saunders, A., E. Srock and N.G. Travlos (1990): Ownershp Srucure, Deregulaon, and Bank Rsk Takng. The Journal of Fnance, vol. XLV, n 2, pp Wall, L.D.and D.R. Peerson. (1995): Bank holdng company capal arges n he early 1990s: The regulaors versus he markes. Journal of Bankng & Fnance, n 19, pp Wner, R.A. (1994): The dynamcs of compeve nsurance markes. Journal of Fnancal Inermedaon, 3, pp

16 Table 1. Summary sascs Sample perod: (annual daa) Sample sze: 1309 observaons Varable Mean Sandard Mnmum Maxmum devaon BUF NPL ROE GDPG Noes: - See he man ex for he defnon of he varables 15

17 Table 2. Esmaon of equaon (1) Dependen varable: BUF, Sample perod: (1029 observaons) Esmaon mehod: GMM, equaon n frs dfferences Explanaory varable Model 1 Model 2 Model 3 BUF,-1.40 (.00).43 (.00).42 (.00) ROE, -.43 (.01) -.39 (.01) -.51 (.00) NPL, (.00) (.00) -- BIG, (.15) (.11) (.13) SMA, (.12) (.17) (.17) GDPG (.00) (.00) (.00) m (.00) (.00) (.00) m2.16 (.87).14 (.89).23 (.82) Sargan es (.26) (.35) (.13) Noes: - See he man ex for he defnon of he varables - p-values n brackes - m1 and m2 sand for frs- and second-order resdual auocorrelaon ess - In all models, BIG, SMA and GDPG are consdered as exogenous - Insrumens for he endogenous varables: lags 2 o 4 n models 1 and 3, and leads 1 o 4 n model 2, as seleced n DPD (Arellano & Bond, 1991). 16

18 Table 3. Exensons of equaon (1) Dependen varable: BUF, Sample perod: (1029 observaons) Esmaon mehod: GMM, equaon n frs dfferences Explanaory varable Model 1 Model 2 Model 3 Model 4 BUF,-1.30 (.00).40 (.00).41 (.00).39 (.00) ROE, -.56 (.00) -.44 (.01) -.43 (.01) -.32 (.04) NPL, -.91 (.04) (.00) (.00) (.00) BIG, (.07) (.40) (.18) (.17) SMA, (.07) (.45) (.11) (.11) GDPG (.00) (.00) (.00) OUTGAP (.00) BIG, *GDPG (.28) SMA, *GDPG (.14) COM, *GDPG (.11) GDPG (.00) m (.00) (.00) (.00) (.00) m2.15 (.88).17 (.87).23 (.82).13 (.90) Sargan es (.20) (.25) (.30) (.26) Noes: - See he man ex for he defnon of he varables - p-values n brackes - m1 and m2 sand for frs- and second-order resdual auocorrelaon ess - BIG, SMA, GDPG, OUTGAP and COM are consdered as exogenous - Insrumens for he endogenous varables: lags 2 o 4 n models as seleced n DPD (Arellano & Bond, 1991). 17

19 Char 1. The lef-hand vercal axs measures he capal buffer n relave erms (curren capal less capal requremens over capal requremens). The rgh-hand vercal axs measures he GDP growh rae of he Spansh economy. Perod: Percenage pons. 60 Capal Buffer GDP Growh