Dversfcaon n Bankng Is Nonneres Income he Answer? Kevn J. Sroh Frs Draf: March 5, 2002 Ths Draf: Sepember 23, 2002 Absrac The U.S. bankng ndusry s seadly ncreasng s relance on nonradonal busness acves ha generae fee ncome, radng revenue, and oher ypes of nonneres ncome. Ths paper assesses poenal dversfcaon benefs from hs shf. A he aggregae level, declnng volaly of ne operang revenue reflecs reduced volaly of ne neres ncome, raher han dversfcaon benefs from nonneres ncome, whch s que volale and has become more correlaed wh ne neres ncome. A he bank level, growh raes of ne neres ncome and nonneres ncome have also become more correlaed n recen years. Fnally, greaer relance on nonneres ncome, parcularly radng revenue, s assocaed wh hgher rsk and lower rsk-adjused profs. These resuls sugges lle obvous dversfcaon benef from he ongong shf oward nonneres ncome. Research Offcer, Federal Reserve Bank of New York, New York NY, 10045. Phone: (212) 720-6633; emal: kevn.sroh@ny.frb.org. The auhor hanks Adam Ashcraf, Allen Berger, Bob DeYoung, Mark Flannery, Tm Hannan, Joe Hughes, Cara Lown, Don Morgan, Larry Radeck, Phl Srahan, Mark Vaughn, wo anonymous referees, and semnar parcpans a he Federal Reserve Bank of New York and he Federal Reserve Sysem Commee Conference for helpful commens, Ken Lamar for help wh accounng ssues, and Me Kok, Adrenne Rumble, and Shana Wang for excellen research asssance. The vews expressed n hs paper are hose of he auhor only and do no necessarly represen hose of he Federal Reserve Sysem or he Federal Reserve Bank of New York.
I. Inroducon The U.S. bankng ndusry s seadly shfng away from radonal sources of revenue lke loan-makng and oward nonradonal acves ha generae fee ncome, servce charges, radng revenue, and oher ypes of nonneres ncome. Whle nonneres ncome has always played an mporan role n bankng revenue, Fgure 1 clearly shows s growng mporance. 1 By 2001, nonneres ncome accouned for 43% of ne operang revenue (ne neres ncome plus nonneres ncome), up from only 25% n 1984. Ths shf oward nonneres ncome has conrbued o hgher levels of bank revenue n recen years, bu here s also a sense ha can lower he volaly of bank prof and revenue, and reduce rsk. 2 One poenal channel s ha nonneres ncome may be less dependen on overall busness condons han radonal neres ncome, so ha an ncreased relance on nonneres ncome reduces he cyclcal varaon n bank profs and revenue. Alernavely, expanded produc lnes and crosssellng opporunes assocaed wh growng nonneres ncome may offer radonal dversfcaon benefs for a bank s revenue porfolo. If nonneres ncome and ne neres ncome are negavely or only weakly correlaed, for example, nonneres ncome may dversfy bank revenue and mprove he rsk/reurn rade-off. The ably o reduce rsk s obvously a opc of consderable mporance for ndvdual banks, as well as her regulaors and supervsors. If nonneres ncome lowers he volaly of bank profs and reduces rsk, for example, mgh be reasonable o reduce capal requremens for banks wh a dversfed revenue porfolo and for supervsors o reallocae her scarce resources. Smlarly, he coss of bank supervson are ed o he perceved rskness of he nsuon, so banks have addonal ncenves o reduce rsk. Managers wh large equy neress n banks wh franchse values have furher ncenves o reduce rsk and manan ha value. Fnally, here s evdence ha large banks ac as f hey are rsk-adverse. All of hese facors conrbue o he keen neres n rsk-reducon among bank managers. Ths paper uses aggregae and ndvdual bank daa from he lae 1970s o 2001 o examne how nonneres ncome affecs he mean and varaon of bank profs and revenues, and o deermne emprcally wheher concenraon n nonradonal acves s correlaed wh rsk ndcaors. The academc leraure s mxed on he rsk and reurn effecs of nonradonal acvy expanson by 1 See Radeck (1999) for deals on he sources of nonneres ncome, wh a focus on paymen servces, for large bank holdng companes n 1996. 2 See, for example, dscussons n he Amercan Banker, March 12, 2001 and June 27, 2001, as well as annual repors from several large bank holdng companes n he lae 1990s. Ths shf could also lower coss f 1
banks, so hs ssue deserves furher sudy. By lookng a he hsorcal lnk beween revenue dversfcaon and performance of U.S. banks, hs paper helps o fll he gap. A he aggregae level, volaly of bank revenue growh has ndeed declned n he 1990s, bu hs reflecs lower volaly whn ne neres ncome growh raher han dversfcaon benefs from ncreased nonneres ncome. Nonneres ncome growh s much more volale han ne neres ncome growh, largely due o very volale radng revenue, and he covarance beween he wo has rsen as he lne beween neres and nonneres acves becomes ncreasngly blurred. More crosssellng and a greaer relance on loan subsues lke commmens are wo examples as boh sraeges expose mulple busness segmens o he same economc or fnancal shocks and naurally reduce he poenal for dversfcaon benefs. Moreover, neher aggregae bank profs (ne ncome) nor he prmary componens of bank revenue (ne neres ncome and nonneres ncome) are very hghly correlaed wh GDP growh; n fac, nonneres ncome appears somewha more cyclcal han ne neres ncome. Ths suggess ha he bankng ndusry should no necessarly be counng on nonneres ncome o smooh revenue flows or reduce aggregae cyclcaly. Bank-level daa pan he same pcure. The cross-seconal correlaon beween ne neres ncome growh and nonneres ncome growh across banks n each year has seadly ncreased from 0.32 n 1979 o 0.66 n 2000. As banks become more heavly nvolved n nonradonal acves, poenal dversfcaon benefs seem o be recedng. Lookng a ndvdual banks over me, he medan bank-specfc correlaon beween ne neres ncome growh and nonneres ncome growh for over 14,000 banks s 0.16 and only one-hrd of he banks show he negave correlaon needed for srong dversfcaon benefs. Moreover, margnal ncreases n nonneres ncome shares for he ypcal bank s assocaed wh a hgher correlaon, agan lmng dversfcaon benefs from ncreased nonneres ncome. Fnally, a novel se of resuls focuses on overall profably and rsk, and shows ha rskadjused reurns are srongly, negavely assocaed wh he share of ncome derved from nonneres sources. Tradng ncome, n parcular, s assocaed wh a declne n prof per un of rsk, whle ncreased fducary ncome s assocaed wh a gan. Nonneres ncome shares are also posvely lnked wh nsolvency rsk, measured by he Z-score. 3 Taken ogeher, hese resuls mply ha he move oward nonneres ncome s acually worsenng he rsk/reurn rade-off for he ypcal bank as volaly ncreases whle average reurns do no. economes of scope allow elmnaon of redundan operaons or leveragng of a fxed cos nvesmen lke compuer nfrasrucure. 2
These resuls sugges cauon for hose belevng ha he shf oward nonneres ncome offers large dversfcaon benefs, guaranees more sable bank earnngs, and lowers he rsk of he U.S. bankng ndusry. Nonneres ncome, parcularly radng, s que volale and he correlaon beween ne neres ncome and nonneres ncome s rsng as produc lnes blur and banks ncreasngly subsue nonradonal sources of ncome for neres ncome. Ths means ha he bankng ndusry may no realze he reducon n volaly and rsk ha some are expecng. II. Exsng Leraure Earler work on bank dversfcaon benefs has aken several dsnc approaches: counerfacual exercses of bank combnaons wh non-banks, examnaon of acual operaons of banks nvolved n many acves, and analyss of marke reacons o bank dversfcaon. These approaches do no gve a unform pcure of he dversfcaon possbles for banks, so hs remans an open research queson. Saunders and Walers (1994), for example, revew 18 sudes ha examne wheher nonbank acves reduce bank holdng company rsk, and conclude ha 9 answer yes, 6 answer no, and 3 provde mxed resuls. Ths secon quckly summarzes he exsng leraure and conrass he approach used n he curren sudy. 4 Begnnng wh he counerfacual exercses, Boyd and Graham (1988) and Boyd, Graham, and Hew (1993) smulae mergers beween bank holdng companes and nonbank fnancal frms and conclude ha mergers beween bank holdng companes and lfe nsurance frms would lkely reduce he rsk of bankrupcy. Rose (1989) compares fnancal and nonfnancal frms from 1966 o 1985 and fnds he observed cash-flow correlaon beween bankng and fnancal-servce lnes were small and posve, mplyng some dversfcaon benefs. Saunders and Waler (1994) perform a smulaon exercse and conclude ha here are poenal gans n he reducon of rsk from bank expanson no new acves. They fnd ha propery and casualy nsurance s a parcularly aracve area for money cener bank expanson. More recenly, Lown e al. (2000) conclude ha lfe nsurance companes are he merger canddaes wh he bgges poenal o reduce rsk. The second approach examnes acual reurn and volaly daa relaed o a wde-range of bankng acves. Rosen e al. (1989) focus on 319 banks nvolved n real esae acves from 1980-1985 and conclude ha shfs oward hgh-levels of real esae nvesmen wll lkely ncrease rsk. Templeon and Severens (1992) examne marke daa for 54 bank holdng companes from 1979 o 1986 and conclude ha dversfcaon (measured as he share of marke value no arbued o bank 3 The Z-score s he number of sandard devaons ha profs mus fall o drve a frm no bankrupcy. See Lown e al. (2000) for deals. 4 Saunders and Waler (1994) and DeYoung and Roland (2001) provde dealed leraure revews. 3
asses) s assocaed wh lower varance of shareholder reurns. Ths suggess some dversfcaon benefs, alhough her measure of dversfcaon s a rough proxy a bes. Kwas (1989) fnds lmed dversfcaon benefs from expanded bank secures powers from 1976 o 1985. Smlarly, Kwan (1998) repors ha bank Secon 20 subsdares ypcally posed more volale accounng reurns, alhough no necessarly hgher reurns. DeYoung and Roland (2001) examne he lnk beween bank profably, volaly, and dfferen revenue shares for 472 large commercal banks from 1988 o 1995. They conclude ha ncreased fee-based acves (revenue from all sources excep loans, nvesmen, depos, and radng acves) ncreases he volaly of bank revenue and bank earnngs. Taken ogeher, here s lle evdence of large dversfcaon benefs from hese papers. Fnally, Acharya e al. (2002) use bank-level daa for Ialan banks from 1993-1999 and conclude ha dversfcaon of bank asses (whn he loan porfolo) does no ypcally mprove performance or reduce rsk. The fnal se of papers uses marke daa o evaluae poenal dversfcaon benefs; some examne acual reurns and oher use smulaon mehods o esmae he mpled volaly of poenal bank expanson. Sanomero and Chung (1992) use opon prcng echnques o smulae he volaly of asse reurns from combnaons of 123 bank holdng companes and 62 non-bank fnancal frms and conclude ha bank expanson no nonbankng busnesses reduces rsk n general. In parcular, bank holdng company mergers wh secures and/or lfe nsurance frms generally reduces he volaly of bank reurns, whle mergers wh propery/casualy nsurance ncreases rsk bu ncreases reurns even more so ha he rsk of falure s no ncreased sgnfcanly. Smlarly, Saunders and Waler (1994) compare he marke reurns of banks and oher fnancal frms and buld porfolo reurns from varous combnaons. They conclude ha lfe and propery nsurance combnaons offer he bgges poenal o reduce sysemac rsk for money cener banks. Houson and Ryngaer (1994) examne he marke reurns for a se of 153 bank mergers from 1985 o 1991 and fnd lle evdence of excess reurns as negave gans o bdders cancel ou posve gans o arges. Whle hs s no a es of dversfcaon drecly, does provde some ndrec evdence as he nsuons are unlkely o operae n he same produc or geographc markes. In fac, hey fnd ha n-marke mergers are beer receved by he marke as hs offers he hghes cos-savng poenal. Fnally, DeLong (2001) uses a smlar approach o examne he dversfcaon queson more drecly. Bank mergers are decomposed no hose ha eher dversfy or focus along eher geographc or acvy dmensons and he resuls show he larges gans for hose mergers ha ncrease focus boh n erms of geographc locaon and acvy. In parcular, he prmary concluson s ha dversfyng mergers do no creae value (pg. 222). Agan, hs s no a drec es of he 4
marke s reacon o ncreases n nonradonal acves, bu does sugges ha dversfcaon gans are no expeced for ypcal bank expansons va mergers. Ths paper bulds upon he second approach and examnes he sources of volaly and average reurns over he las wo decades, alhough several mporan feaures dfferenae from he earler work. Frs, hs sudy uses acual daa for dversfed bankng nsuons raher han smulaons of he mpled volaly of accounng or marke reurns beween possble combnaons of bank and non-bank fnancal nsuons. In parcular, hs s he frs sudy o examne he hsorcal lnk beween a dversfed revenue porfolo and rsk-adjused profably for all banks, whch provdes prma face evdence on he acual effecs of a dversfed earnngs sream. 5 Second, hs sudy examnes he dversfcaon benefs across ypes of bank revenue n wo complemenary ways a cross-seconal and a bank-specfc correlaon whch provdes a fuller undersandng of he rsk/reurn effecs of growng nonneres ncome. For example, he cross-seconal correlaon shows how he correlaon beween nonneres ncome and ne neres ncome s changng over me, whle bank-specfc correlaon shows how he correlaon dffers across banks. Thrd, hs s he frs comprehensve sudy ha examnes he facors ha are assocaed wh he correlaon beween dfferen ypes of bank revenue. Fourh, prevous work only looked a bank-level daa, ofen wh a relavely small sample. In conras, hs paper presens boh aggregae and bank-level resuls based on complee daa for he U.S. bankng ndusry wh over 15,000 observaons. By examnng he enre ndusry, one can gan a beer dea of he pervasveness of hs acvy and s macroeconomc mporance. Fnally, daa are hrough 2001, whch provdes a more curren perspecve on he ssue ha s mos relevan for bank supervsors and regulaors. III. Daa, Summary Sascs, and Trends Two ypes of daa are used n hs paper. Aggregae U.S. bankng ndusry daa were provded by he Federal Depos Insurance Corporaon (FDIC). These daa are on a quarerly frequency from 1984:Q1 o 2001:Q3, and were deflaed wh he GDP deflaor and seasonally adjused. Bank-level daa are from he Consoldaed Repor of Condon and Income ( Call Repors ). These daa are on an annual bass from 1978 o 2000 and deflaed wh he GDP deflaor. In cases of mergers and acqusons, he acqurng bank s code s mananed and he arge drops from he sample. The effec of mergers on he bank-level resuls s addressed below. See he Daa Appendx for deals. Table 1 shows he breakdown of major sources of bank revenue for 1980, 1990, and 2000 for hree ses of banks all banks, banks wh asses greaer han $10 bllon (n 1996 dollars), and banks 5 Saunders and Waler (1994) use a porfolo approach o examne poenal reurns hrough varous smulaon exercses usng marke daa. 5
wh asses below $10 bllon as aggregaed from he Call Repors. Ne ncome, ne operang revenue (defned as ne neres ncome plus nonneres ncome), oal asses, and he number of banks for each sample are repored. Nonneres ncome s a heerogeneous caegory ha comprses many dfferen acves, so s broken down no four prmary componens fducary ncome, servce charges, radng revenue, and fees and oher ncome. Fducary ncome s revenue relaed o he bank s fducary operaons, e.g., admnserng nvesmens for ohers. Servce charges nclude revenue drecly relaed o depos accouns lke ATM or check usage fees. Tradng revenue s prmarly ncome from radng cash nsrumens, off-balance conracs, and mark-o-marke changes n he carryng value of asses and lables. Fees and oher ncome nclude all oher fees, e.g., loan commmen fees, safe depos boxes, commssons, and land renal fees. Of course, here s also consderable heerogeney whn hese broad caegores, bu hs s he bes one can do wh he avalable daa. Deals on hese componens are provded n he Daa Appendx. Kwas (1989) provdes accounng deals on radng revenue n he Call Repors, alhough one aspec deserves specal noe here. Tradng revenue ncludes he effec of markng o marke ceran on-balance shee asses and lables and off-balance shee ems. Ths s he only componen of nonneres ncome ha s based on mark o marke accounng, and accounng dfferences may affec measured volaly as he ncome ems responds more quckly o changng marke condons. I s no possble, however, o precsely quanfy how much of radng revenue s due o hese adjusmens. As shown n Table 1, nonneres ncome for all U.S. banks ncreased as a share of ne operang revenue from 20.4% n 1980 o 32.5% n 1990 o 43.4% n 2000. The bgges ncrease was n fees and oher ncome, bu all four componens of nonneres ncome showed szable ncreases as a share of ne operang revenue. The same paern s rue for he sub-samples of very large and smaller banks. Large banks, for example, saw nonneres ncome rse from 26.5% of ne operang ncome n 1980 o 47.8% n 2000, whle small banks saw an even faser rse from 17.4% o 33.1%. Tradng ncome s hghly concenraed n he larges banks wh subsanal nonradonal operaons. A key pon, however, s ha he shf oward nonradonal sources of ncome seems o be pervasve across banks of all szes, and no jus lmed o a few mega-banks ha perform many dverse acves. IV. Aggregae Flucuaons of Bank Revenue Ths secon examnes he aggregae flucuaons of bank profs and revenue over he las wo decades. I explore wheher ne ncome and ne operang revenue are becomng more cyclcal as banks relance on nonneres ncome grows, wheher he prmary revenue sreams are becomng more 6
volale, and whch componens are drvng changes n aggregae cyclcaly and volaly. All daa n hs secon are he quarerly ncome fgures for he U.S. bankng ndusry provded by FDIC. All shown n Table 1, boh major revenue sources are growng over he sample, so a naural frs sep n examnng wheher ne neres and nonneres ncome move ogeher s smply o plo he quarerly growh raes of hese seres over me. Fgure 2 shows he growh of each ype of revenue from 1984:Q1 o 2001:Q3 wh NBER recesson perods as shaded areas. Casual examnaon does no sugges a srong correlaon wh he aggregae busness cycle for eher seres. Nonneres ncome appears much more volale han ne neres ncome, parcularly n he 1990s. 6 Nonneres ncome also has a hgher mean growh rae, however, and he coeffcen of varaon (he rao of he sandard devaon o he mean) for nonneres ncome growh s only slghly hgher han for ne neres ncome growh (2.0 vs. 1.9). The volaly of ne neres ncome growh s largely drven by he dramac swngs n he lae 1980s when he bankng ndusry experenced subsanal problems relaed o real esae and lendng o lesser-developed counres. If hs perod s excluded and one examnes he daa afer 1989, hen nonneres ncome growh s much more volale han ne neres ncome growh: he coeffcen of varaon for nonneres ncome growh s 2.3 compared o 1.1 for ne neres ncome growh. 7 Thus, nonneres ncome appears o be becomng more volale, whle ne neres ncome s becomng less so. Fgure 3 shows he growh raes of he four componens of nonneres ncome fducary ncome, servce charges, radng revenue, and fees and oher nonneres ncome. Due o he exreme volaly of hese seres, Fgure 3 plos four-quarer growh raes and even hese smoohed seres show enormous volaly for radng revenue. For he perod 1984:Q1-2001:Q3, he sandard devaon of he four-quarer growh rae of radng ncome was 39.8, compared o 7.2 for fees and oher nonneres ncome, 6.6 for fducary ncome, and 2.9 for servce charges. 8 Clearly, radng ncome s he mos volale componen of bank ncome. 6 F-ess are used o compare he volaly of ne neres ncome o nonneres ncome and rejec he null hypohess of equal sandard devaons for he full perod (p-value=0.00), he perod 1990:Q1 o 2001:Q3 (pvalue=0.00), and he perod 1984:Q1-1989:Q3 (p-value=0.06). 7 For he perod 1990:Q1 o 2001:Q3, ne neres ncome growh averaged 3.3% wh a sandard devaon of 3.7, whle nonneres ncome growh averaged 7.0% wh a sandard devaon of 16.0. The null of equal sandard devaons s srongly rejeced (p-value=0.00). 8 F-ess decsvely rejec he null hypohess of equal sandard devaons for all parwse combnaons of he quarerly growh raes for he four componens of nonneres ncome excep for fducary and oher nonneres ncome (p=0.12). 7
(1). 10 These esmaes show ha bank revenue has become slghly less volale, e.g., he varance of a) Aggregae Volaly To gauge how ne neres ncome and nonneres ncome conrbue o he volaly of bank revenue, s useful o hnk of ne operang revenue as a smple porfolo of wo ypes of asses: hose ha generae ne neres ncome and hose ha generae nonneres ncome. 9 I s dffcul o denfy he specfc asses assocaed wh each ncome class o calculae a rae of reurn, however, so I modfy he sandard decomposon of porfolo reurn volaly no a decomposon of porfolo growh volaly. Tha s, f ne operang revenue (OPREV) s defned as he sum of ne neres ncome (NET) and nonneres ncome (NON), hen he volaly of ne operang revenue growh s: 2 2 2 2 2 (1) σ = α σ + ( 1 α ) σ + 2 α ( 1 α ) Cov( d ln NON, d ln NET ) d ln OPREV d ln NON d ln NET where α = NON /( NET + NON) s he nonneres share of bank operang revenue growh, d ln X s he growh rae of X, and he conrbuon of each componen o overall revenue volaly 2 2 s he share-weghed varance, e.g., α s he conrbuon of nonneres ncome. σ d ln NON Sandard porfolo heory mples ha he overall varance of ne operang revenue rses as he nonneres ncome share grows f nonneres ncome s more volale han ne neres ncome. A negave covarance beween nonneres ncome growh and ne neres ncome growh wll drecly lower he overall varance. Even f he covarance erm s posve, however, he rade-off beween growh of ne operang revenue and volaly can mprove. Tha s, average growh s he weghed average of he growh raes of he componens, bu he sandard devaons wll be less han he weghed average as long as he covarance s no exacly one. Table 2 shows esmaes for he componens of Equaon (1) for wo me perods 1984:Q1 o 1989:Q4 and 1990:Q1 o 2001:Q3. The frs column gves he average shares of nonneres ncome (α ) and ne neres ncome ( 1 α ), he second column gves he sample varance or ( ) 2 2 2 covarance of he varables σ, σ, σ, and Cov( d ln NON, d ln NET ) d ln OPREV d ln NON d ln NET, and he hrd column gves he conrbuons (share-weghed varances) as on he rgh-hand sde of Equaon ne operang revenue growh fell from 50.4 for 1984:Q1-1989:Q4 o 46.2 for 1990:Q1-2001:Q3, alhough he dfference n volaly s no sascally sgnfcan. More mporan, however, s he 9 Ths approach s smlar o he bank-level work of Kwas (1989), Rose (1989), and Saunders and Waler (1994), alhough hey focus on reurns across dfferen operaons. 10 The sum of he conrbuons wll no exacly mach he varance of ne operang revenue because he share s changng hroughou he sample. 8
observaon ha he declne n volaly does no seem o reflec any dversfcaon benefs beween ne neres ncome and nonneres ncome growh because he covarance beween he wo became larger across he wo perods ( 29.0 n he frs perod vs. 5.6 n he second). In addon, nonneres ncome s becomng more volale (varance ncreased from 228.9 o 259.1, no sgnfcanly dfferen) and growng n relave sze (share ncreased from 28% o 37%). Ths led o an ncrease n he conrbuon o he volaly of ne operang revenue growh from 18.2 n he frs perod o 35.8 n he second. The enre declne n overall revenue volaly, herefore, s due o ne neres ncome, whch became much less volale (varance fell from 100.2 o 14.2, sgnfcanly dfferen) and relavely less mporan (share fell from 72% o 63%). 11 The ncreased correlaon beween ne neres ncome growh and nonneres ncome growh deserves some dscusson. One plausble explanaon s ha he ncreased focus on cross-sellng dfferen producs o a core cusomer base may expose dfferen lnes of a bank s busness o he same shock, e.g., frm, ndusry, or lendng marke roubles. Tha s, smply sellng more producs o he same cusomers does no mply dversfcaon benefs f a gven cusomer s demands for all producs are hghly correlaed. Smlarly, he ncreased use of loan commmens exposes boh ne neres ncome and fee and oher nonneres ncome (where loan commmen fees are booked) o he same flucuaons n loan demand and marke condons. 12 In boh cases, f a bank provdes boh radonal lendng and nonradonal fee-based acves o a parcular frm or ndusry, hen several lnes of busness may suffer smulaneously f ha frm or ndusry faces fnancal roubles. Ths would ncrease he correlaon beween he wo revenue sreams and reduce dversfcaon benefs. Ths decomposon shows ha bank operang revenue has become less volale smply because ne neres ncome has become less volale. Ths could reflec dversfcaon benefs whn radonal bank acves as banks become less geographcally concenraed. The lae 1980s, for example, were a perod of regonal shocks and regonal bankng problems. An alernave explanaon s ha he early sample was parcularly volale due o he severe bankng problems of he lae 1980s as can be seen n he spkes of ne neres ncome growh. In addon o a more sable bankng envronmen, he U.S. economy n general has become more sable n he 1990s, whch lkely helped o smooh bank revenue. 13 Dversfcaon benefs assocaed wh he shf oward nonneres ncome do no seem o have played an mporan role. 11 The p-values assocaed wh he null hypohess of equal sandard devaons across he wo perods are: 0.81 for ne operang revenue, 0.00 for ne neres ncome, and 0.74 for nonneres ncome. The same paern holds f radng ncome s excluded from he analyss. 12 Aggregae unused loan commmens ncreased from 33% of oal loans n 1984 o over 120% n 2001. 13 Perez-Quros and McConnell (2000) documen a declne n GDP volaly ha began n he md-1980s. 9
b) Aggregae Cyclcaly Ths secon descrbes he cyclcal properes of he dfferen ypes of bank revenue. Raher han smply lookng a recesson perods as n Fgures 2 and 3, I calculae correlaons beween ncome growh and lags of GDP growh by esmang he followng regresson: 4 (2) d ln X = α + β τ d ln X τ + δ τ d ln GDP τ + ε τ = 1 where X s some measure of bank ncome and GDP s real GDP. 14 Table 3 repors esmaes of Equaon (1) for sx measures of bank ncome ne ncome, ne ncome plus provsons, ne neres ncome, ne neres ncome less provsons, nonneres ncome, and nonneres ncome less radng revenue. The frs wo are measures of bank profs; I begn wh he boom lne measure of ne ncome and hen add back loan loss provsons. Ne ncome plus provsons s ncluded because here s some evdence ha banks use loan loss provsons o smooh earnngs over he cycle and addng back provsons may help o beer undersand he cyclcaly of bank earnngs. 15 The second wo columns are ndcaors of radonal bankng acves, measured as ne neres ncome. Agan, I begn wh he sandard measure and hen adjus for provsons, here subracng provsons, because may provde a more accurae measure of he reurn o radonal lendng acves because accouns for expeced defauls. The fnal wo columns measure nonradonal acves n he form of nonneres ncome. Nonneres ncome less radng revenue s ncluded o solae he poron of nonneres ncome no assocaed wh he very volale radng book. Begnnng wh he regressons wh bank profs as he dependen varable n column 1, here appears o be only a weak correlaon beween ne ncome growh and GDP growh, e.g., he conemporaneous and lagged GDP coeffcens are jonly sgnfcan a he 10% level (p-value of 0.08). Addng back provsons n column 2 weakens he sgnfcance of he GDP varables slghly (pvalue=0.19), so ncome smoohng does no appear o he sory. An alernave explanaon s ha GDP growh s oo broad of a measure,.e., banks were geographcally consraned for much of he sample. Meyer and Yeager (2001), however, fnd only a weak correlaon beween local economc acvy and he performance of small rural banks so hs s unlkely o be he whole answer. Movng o he wo broad componens of ne operang revenue, boh ne neres ncome n column 3 and nonneres ncome growh n column 5 show lle evdence of a lnk wh curren and 4 τ = 0 14 Rose (1989) ran a smpler regresson for bankng cash flows usng only conemporaneous GNP growh and oher conrols, bu no lagged dependen varables. He fnds ha bankng cash flow s negavely correlaed wh real GNP growh, bu uncorrelaed wh nomnal GNP growh. 15 See Greenwal and Snkey (1988), alhough Ahmed e al. (1999) conclude ha banks do no use provsons o smooh earnngs. 10
lagged GDP growh raes (p-value=0.27 and 0.12, respecvely). Subracng provsons from ne neres ncome n column 4 weakens he relaonshp dramacally. The large and sgnfcan negave coeffcens on he lagged dependen varable ndcae he negave auocorrelaon of he seres,.e., rapd growh n provsons n one quarer s ypcally followed by negave growh nex quarer as provsons reurn o normal levels. Removng radng revenue from nonneres ncome n column 6 subsanally srenghens he lnk wh lagged GDP, whch mples ha aggregae fducary ncome, servce charges, and fees are more hghly correlaed wh GDP han radng ncome. Whle hese regressons are que smple, hey appear o refue he belef ha he shf oward nonradonal bankng acves can smooh he effec of busness cycle flucuaons. A second way o address hs ssue s wh a smple vecor auoregresson (VAR) framework,.e., a sacked se of regressons of GDP growh, ne neres ncome, and nonneres ncome on lagged values of he same. A pror, one would expec GDP o affec bank revenue, bu s less obvous ha bank revenue should affec GDP; hs suggess a naural orderng of he VAR wh GDP frs. Fgures 4a and 4b show he mpulse response funcon from a VAR wh GDP, ne neres ncome, and nonneres ncome from 1984:Q1 o 2001:Q3. The VAR s esmaed n log-levels for real GDP, real ne neres ncome, and real nonneres ncome wh a rend. The mpulse response funcon shows how a shock o GDP propagaes hrough he sysem and affecs ne neres ncome and nonneres ncome. The resuls ndcae a weak, posve response of ne neres ncome o shocks o GDP for a few quarers, and a sgnfcan response for nonneres ncome n only he frs quarer. The mng s also que dfferen wh ne neres ncome showng a shor-run mpac ha fades, whle nonneres ncome shows a modes, alhough sascally nsgnfcan, ncrease over he longer run. 16 These resuls ndcae ha ne neres ncome and nonneres ncome respond only weakly o changes n real oupu. When radng ncome s removed from nonneres ncome, however, here s sronger lnk wh GDP growh, whch provdes lle suppor for he belef ha nonradonal ncome sources wll remove he cyclcaly of bank revenue. Moreover, any gans ha do exs mus be weghed agans he larger uncondonal volaly of nonneres ncome growh. V. Bank-Level Varably Ths secon moves beneah he aggregae daa o examne he role of nonneres ncome n deermnng he profably and rskness of ndvdual banks. Ths s done n wo ways. Frs, a cross-seconal correlaon measures he correlaon beween ne neres ncome growh and 16 Augmened Dckey-Fuller ess for GDP, ne neres ncome, and nonneres ncome wh four lagged dfferences fal o rejec he null hypohess of a un roo n levels, bu rejec he same null n frs dfferences. 11
nonneres ncome growh across banks a a pon n me. 17 Second, a bank-specfc correlaon measures he correlaon beween ne neres ncome growh and nonneres ncome growh across me for each bank. 18 The nal focus s on bank revenue, raher han bank profs, because he percepon seems o be ha dversfcaon leads o smooher revenue flows and because expenses canno be allocaed across busness lnes o generae reurns on each segmen. Bank profably, safey, and soundness, however, may be he areas of ulmae neres, so he las sub-secon focuses on he lnk beween nonneres ncome and he level and varably of bank profs. In parcular, I look a he hsorcal relaonshp beween nonneres ncome shares and measures of bank profably (ne ncome growh and mean ROE) and rsk (Sharpe rao and Z-scores). a) Cross-Seconal Correlaon defned as: where The cross-seconal correlaon, (3) ρ = Corr( d ln NET = I = 1 I = 1, ρ gves he correlaon across banks n each year and s [( d ln NET d ln NET )( d ln NON d ln NON )] 1/ 2 I 2 ( d ln NET d ln NET ) ( d ln NON d ln NON ),, d ln NON,, ) ρ s esmaed for each year from 1979 o 2000. NET, s ne neres ncome and NON, s nonneres ncome for bank n year, and across all I banks n year. = 1,, d ln NET and d ln NON 2 1/ 2 are he average growh raes The cross-seconal correlaon s esmaed separaely for each year and descrbes how much ne neres ncome and nonneres ncome move ogeher across banks n a parcular year. Tha s, shows wheher a bank wh above average nonneres ncome growh ypcally has above average growh n ne neres ncome. If nonneres ncome plays a dversfyng role for bank revenue, hen one would expec hem o be negavely correlaed,.e., posve shocks o one revenue source are Ths mples ha all seres are negraed of order 1. When he VAR s run n log frs dfferences, here s no sgnfcan response of eher ne neres ncome or nonneres ncome o GDP shocks. 17 Ths analyss s done n growh raes because correlaons of revenue levels would be largely deermned by he scale of he nsuons. 18 The cross-seconal correlaon has one observaon for each year, whle he bank-specfc correlaon has one observaon for each bank. 12
offse by negave shocks o he oher one. A srong posve correlaon, on he oher hand, would sugges lle dversfcaon from he broadenng of bank revenue. One poenally confoundng facor s he mpac of rapdly growng or rapdly shrnkng banks. For example, f a bank acqures anoher bank n a gven year s enre ncome saemen and balance shee would lkely jump, whch would rase he cross-seconal correlaon. Smlarly, a conracng bank ha s shrnkng due o compeve pressures or a declnng marke would lkely see negave growh n boh ne neres ncome and nonneres ncome. Ths s no he ype of dversfcaon effec ha people seem o have n mnd, so I esmae he cross-seconal correlaon boh wh and whou hese ypes of rapdly changng banks. In pracce, I begn wh all banks wh complee daa n wo consecuve years and hen esmae ρ for hree ses of banks: all banks, large banks (asses greaer han $1 bllon n 1996 dollars), and a group of non-jumpng banks (asse growh above he 10 h percenle and below he 90 h percenle n ha year) o remove he effec of rapd growers and shrnkers. 19 In all cases, ncluded banks have daa on ne neres ncome, nonneres ncome, and asses n year and n year - 1 o be ncluded n he esmaon of ρ. Fgure 5 plos he me seres of cross-seconal correlaons, ρ, for he hree ses of banks. When all banks are ncluded, ρ rses seadly from 0.32 n 1979 o 0.66 n 2000; he average across all years s 0.53 wh a mean of 0.45 before 1990 and 0.61 afer 1989. 20 As expeced, when nonjumpng banks are excluded, he esmaed correlaon falls, parcularly for he early years, bu becomes que srong (near 0.50) and sascally dfferen from zero durng he lae 1990s. For large banks, here s less of an upward drf, bu he level s que hgh wh an average of 0.62 across all years. The esmae for large banks s much more volale because of he decreased number of observaons. 21 I s also neresng o noe ha he correlaon for large banks falls durng recessons, bu s dffcul o draw frm conclusons from only wo observaons. Overall, he correlaon has rended up, whch means ha bank-specfc shocks o ne neres ncome and nonneres ncome are 19 The non-jumpng sub-sample excludes banks ha have exraordnary ncreases or decreases n her asses and hus should remove banks nvolved n mergers or large dvesures ha mgh cloud he nerpreaon. Resuls are robus o changng he cu-off. 20 The dfference beween he wo sub-samples s sgnfcan a he 99 percen level and null ha 1979 and 2000 have he same value s rejeced (p-value=0.03). 21 The sample of all banks drops from 14,117 n 1979 o 8,110 n 2000 and he sample of non-jumpng banks falls from 11,319 n 1979 o 6,515 n 2000, whle he sample of bg banks rses from 308 o 358. The dfference beween he wo me perods s sgnfcan a he 95 percen level for he large bank subsample and a he 99 percen level for he non-jumpng banks. 13
becomng more hghly correlaed precsely when banks are ncreasng her focus on nonneres ncome. One can also calculae he cross-seconal correlaon beween he componens of nonneres ncome fducary ncome, servce charges, radng revenue, and fees and oher and ne neres ncome for all banks. Fgure 6 shows ha servce charges are he componen mos hghly correlaed wh ne neres ncome. Ths s no surprsng as servce charges are closely algned wh radonal bankng operaons lnked o depos-akng acves, e.g., manenance, mnmum balance, number of check fees, ec. If all radonal acves respond o he same economc shocks lke regonal economc acvy, hen one would expec o see a hgh correlaon beween hese revenue sreams. The correlaon of ne neres ncome wh fees and oher neres ncome has ncreased by a facor of hree over wo decades, whch s conssen wh he earler dscusson ha ncreased use of loan commmens s blurrng he lne beween lendng and fee ncome. As banks subsue cred commmens for acual loans, he correlaon wll rse f boh respond n a smlar fashon o changes n loan demand. Fnally, radng ncome growh shows he lowes correlaon wh ne neres ncome growh. Ths s reasonable as radng ncome s more dependen on marke flucuaons han radonal bankng acves and herefore responds o dfferen shocks. These resuls show a relavely hgh degree of correlaon beween nonneres ncome and ne neres ncome across banks, whch suggess lle obvous dversfcaon benefs as growh n one ype of ncome s ypcally assocaed wh smlar growh n he oher ype. Moreover, he correlaon has been rendng up, mplyng less dversfcaon benefs as he bankng ndusry seadly shfs s revenue focus o nonneres ncome sources. Tradng ncome and fducary ncome show he weakes correlaon, whch mples he bes dversfcaon opporunes for a bank wh radonal, neresgenerang acves, alhough any poenal gans mus be weghed agans he hgher overall volaly of radng ncome. b) Bank-Specfc Correlaon defned as: The bank-specfc correlaon, (4) ρ = Corr( d ln NET = T = 1 T = 1, ρ gves he correlaon across me for each bank and s [( d ln NET d ln NET )( d ln NON d ln NON )] 1/ 2 T 2 ( d ln NET d ln NET ) ( d ln NON d ln NON ),, d ln NON,, ) = 1,, 2 1/ 2 14
where ρ s esmaed for each bank wh a leas sx years of daa for boh ne neres ncome growh and nonneres ncome growh. across all T years of s years of operaons. d ln NET and d ln NON are he average growh raes for bank The bank-specfc correlaon s esmaed for each bank and descrbes how a bank s wo prmary sources of ncome move ogeher over me. Ths s a more radonal measure of correlaon and has drec mplcaons for he dversfcaon queson because measures wheher a gven bank s shocks o one ype of ncome are ypcally accompaned by smlar shocks o he second. A negave correlaon would sugges srong poenal dversfcaon benefs. Dfferences n bank behavor can also play a confoundng role n nerpreng hese esmaes. Consder a bank ha s rapdly growng hroughou he sample perod due o a sream of acqusons. Ths bank would lkely have rapd growh of boh ne neres ncome and nonneres ncome, on average, bu hs would no be a problem here because ρ calculaes wheher perods of above average ne neres ncome growh are ypcally accompaned by above average nonneres ncome growh. Very rapd or very slow growh raes for only a few years, however, wll affec ρ, so he emprcal work conrols for average growh raes. ρ s esmaed wh annual daa for all banks wh growh raes of boh ne neres ncome and nonneres ncome for more han fve years durng he sample perod. 22 Ths lef a sample of 14,503 observaons for dsnc banks wh anywhere from sx years (he mnmum cu-off for ncluson) o 22 years (he complee sample). The mean ρ was 0.20, wh a medan of 0.16 and a sandard devaon of 0.42. For large banks (average asses greaer han $1 bllon n 1996 dollars), he mean ρ was 0.33, supporng he earler resul ha large banks ypcally show a sronger correlaon beween ypes of ncome growh. For non-jumpng banks (average asse growh beween he 10 h percenle and 90 h percenle), he mean was 0.15, whch s close o he full sample and suppors he conenon ha rapdly growng or shrnkng banks are no drvng hese resuls. Taken ogeher, hese resuls show a modes, posve correlaon beween ne neres ncome growh and nonneres ncome growh for he ypcal bank. Ths procedure yelds over 14,500 esmaes of ρ, so s useful o look a he dsrbuon. Fgure 7 shows a wde range of ρ for all banks ha s cenered a 0.2, bu ranges from -0.96 o nearly 15
1.0. An asse-weghed dsrbuon, whch provdes a beer pcure of he dversfcaon poenal for he ndusry as a whole, shows a smlar paern as he unweghed dsrbuon n Fgure 7. The mass n he rgh-hand al wh very hgh correlaons prmarly reflecs banks wh a relavely small number of years of daa; f he graph s lmed o banks wh he full 22 years of daa, hs mass dsappears and he dsrbuon appears more normal. The als of he dsrbuon are also of parcular neres because large negave correlaons mply he bgges poenal dversfcaon benefs and large posve correlaons he leas. For he complee sample, abou one-hrd of he banks show a negave correlaon beween ne neres ncome growh and nonneres ncome growh. Banks wh a negave correlaon end o be smaller on average han banks wh a posve correlaon (mean average asses of $230 mllon vs $370 mllon) and show slower growh over her lfe (2.8% mean average growh vs. 6.4%). One can also focus on he exreme ends of he dsrbuon and compare banks n he 10 h percenle (mean ρ of 0.48) o hose n he 90 h percenle (mean ρ of 0.92). Here, here s lle dfference n average sze, alhough he banks wh a hgh correlaon dd grow subsanally faser (13.9% on average for banks above he 90 h percenle vs. 2.8% for hose below he 10 h percenle). One can look a he facors assocaed wh he correlaon beween growh n he wo ypes of bank revenue by usng he esmae of ρ as he dependen varable n he followng regresson: (5) ρ = α + β A E + β 1 ln( ) 2 3 ln( ) A β FIDSHR 4 + β d + β TRDSHR 5 A + β NONSH + β FEESHR 6 4 + β NONSH + β BANYRS 7 5 2 + + β MULTI 8 + ε where A s asses, E s he leverage rao, d ln(a ) s annual asse growh, NONSH s nonneres A ncome s share of ne operang revenue, FIDSHR s fducary ncome s share of nonneres ncome, TRDSHR s radng ncome s share of nonneres ncome, FEESHR s fees and oher nonneres ncome s share of nonneres ncome, BANKYRS s he number of observaons used n he bankspecfc correlaon, and MULTI s a dummy varable ndcang f he bank belongs o a mul-bank 22 The fve-year cu-off was chosen o drop banks wh only a few years worh of daa for whch he bank-specfc correlaons are less meanngful. Some banks were also dropped due o mssng daa on he breakdown of ne operang revenue. 16
holdng company. 23 Bars over varables ndcae averages for he number of perods bank s observed. 24 Nonneres ncome share s one measure of he focus on nonradonal bankng acves. Whle hs s surely no an exogenous varable and reflecs banks sraegc choces and busness opporunes, one can use o examne he hsorcal relaonshp beween revenue correlaons and nonradonal acvy. The nonneres ncome share s ncluded drecly and as a quadrac erm o accoun for he naural non-lnear relaonshp. Tha s, a bank wh all of s ne operang revenue as eher ne neres ncome (NONSH=0) or as nonneres ncome (NONSH=1) would show lle correlaon beween he growh raes of he wo varables. Average asses are ncluded o conrol for any sysemac dfferences n he correlaon ha are relaed o sze, e.g., large banks may have more developed rsk managemen echnques or may be nvolved n fundamenally dfferen ypes of acves wh dfferen dsrbuons. The leverage rao and asse growh are ncluded o proxy for bank rsk as hgh-rsk and low-rsk banks may have dfferen operang sraeges ha lead hem o focus on revenue generaon n dfferen ways. The number of observaons of each bank, whch ranges from sx o 22, s ncluded because banks ha are observed for many perods may have a lower correlaon, on average, smply because he ncome sreams are longer and subjec o more nose. Fnally, an ndcaor of he bank s afflaon s ncluded because banks owned by a mul-bank holdng company may have more ably o shf ncome and affec observed volaly. Table 4 repors resuls for he hree ses of banks all banks wh an esmae of ρ, nonjumpng banks, and large banks. For each se of banks, I repor esmaes of Equaon (5) boh wh and whou he dealed shares of nonneres ncome. Of course, hese are all reduced form regressons and one canno draw causal nferences; raher, hey are jus condonal correlaons. For all banks and non-jumpng banks (columns 1 and 3), he lnear and quadrac nonneres ncome shares are jonly sgnfcan and show he expeced nvered-u shape,.e., a posve coeffcen on he lnear nonneres ncome share erm and a negave coeffcen on he squared erm. Ths means ha he correlaon s larges for banks n he mddle range. For banks wh low nonneres shares, here appears o be lle dversfcaon benef from margnal ncreases n nonneres ncome as hs ncreases he correlaon. For large banks (column 5), however, he 23 A bank s denfed as belongng o a mul-bank holdng company (MULTI=1) f anoher commercal bank shares he same drec holdng company (RSSD9379 n he Call Repors) or hgh holdng company (RSSD9348 n he Call Repors). 24 One mus arbrarly drop one of he four componens of nonneres ncome o avod perfec collneary. Servce charges s chosen because has he larges mean share of nonneres ncome. 17
relaonshp s negave, whch lkely reflecs he relavely hgh shares of nonneres ncome for large banks,.e., hey are on he downward par of he nvered-u. The oher conrols show ha larger banks, banks wh hgh equy raos, and rapdly growng banks end o show hgher correlaons beween ne neres ncome growh and nonneres ncome growh, alhough he sgnfcance vares across samples. In addon, banks ha belong o mul-bank holdng companes always show a hgher correlaon. Movng o he lnk wh dfferen shares of nonneres ncome (columns 2, 4, and 6), he esmaed coeffcens on he nonneres ncome shares do no change much, whle he coeffcens on he dealed componen shares are unformly negave and ofen sascally dfferen from zero. As banks shf her nonneres ncome away from servce charges (he omed componen of nonneres ncome) and oward fducary, radng, or fees and oher nonneres ncome, he correlaon drops, whch suggess ha here may be some dversfcaon benefs from hese acves. Hgher concenraons of servce charges, on he oher hand, are assocaed wh hgher correlaons beween ne neres ncome growh and nonneres ncome growh. Ths suppors he cross-seconal esmaes n Fgure 6 ha showed servce charges were he componen of nonneres ncome mos hghly correlaed wh ne neres ncome n each year. Agan, hs s reasonable as servce charges are he componen of nonneres ncome mos closely algned wh radonal bankng acves. 25 c) Nonneres Incomes Shares, Bank Rsk, and Reurn The fnal sep s o examne how nonneres shares are correlaed wh oher varables n whch we are ulmaely neresed. For example, do banks wh a dversfed revenue sream show less volale profs? Hgher average profs? Lower nsolvency rsk? Examnng he hsorcal lnk beween nonneres ncome shares and fundamenal measures of bank profably and rsk provdes a beer undersandng of he mpac of he shf oward nonneres ncome on he srengh and sably of he bankng ndusry. Bankers and regulaors, for example, may be mos neresed n how nonneres ncome affecs banks profably and nsolvency rsk. As a frs pass, I graph he hsorcal relaonshp beween nonneres ncome shares and wo sandard measures of performance and rsk he Sharpe Rao and he Z-score. The Sharpe Rao (average reurn on equy (ROE) dvded by he sandard devaon of ROE) s a measure of rskadjused profs. 26 The Z-score (average reurn on asses (ROA) plus average equy o asses, dvded by he sandard devaon of ROA) measures how many sandard devaons profs mus fall below s 25 These resuls are robus o he excluson of radng ncome from he nonneres ncome varable and he shares. 18
mean o bankrup he frm and s relaed o he probably of falure. In boh cases, lower values ndcae ncreased rsk. Fgure 8 plos he Sharpe Rao and Z-score agans nonneres ncome shares and shows a srong negave slope: banks wh hgh nonneres ncome shares earn lower rsk-adjused profs and are relavely rsky. Whle here are no conrols and nonneres shares are obvously an endogenous varables so one canno draw causal nferences from hs, banks wh hgh nonneres ncome shares seem o have hgher rsk. 27 One can be more formal abou hs by ulzng regressons smlar o Equaon (5) wh dfferen dependen varables. Columns 1 and 2 of Table 5 use he mean and sandard devaon of ne ncome growh, respecvely, as he dependen varables. These regressons show how he mean and volaly of bank prof growh vary wh he composon of he revenue sream. Columns 3 hrough 5 of Table 5 shf o he mean and volaly of reurn on equy (ROE). Column 3 uses mean ROE as he dependen varable, column 4 uses he sandard devaon of ROE, and column 5 uses he average Sharpe Rao (defned above). Column 6 uses he Z-score (defned above) o examne he lnk beween nonneres ncome shares and nsolvency rsk. Column 1 shows he same nvered-u when he mean growh rae of ne ncome s he dependen varable as banks wh a balanced revenue porfolo show hgher average ne ncome growh, even afer conrollng for average growh n oal asses. The coeffcens on he componens of nonneres ncome sugges ha radng ncome shares are mos hghly assocaed wh hgher growh, alhough s no sascally dfferen from zero. Fducary ncome shows he mos negave correlaon, whch s conssen wh he dea ha s a slow growh busness. Wh he volaly of ne ncome growh as he dependen varable (column 2), he coeffcens agan show an nvered-u. The peak of he U, however, s far beyond he ypcal nonneres ncome share so an ncrease n he nonneres ncome share s assocaed wh ncreased volaly of ne ncome for he ypcal bank. 28 Average sze s srongly negavely correlaed wh ncome volaly, perhaps because large banks are nherenly more dversfed across and whn busness lnes and geographc markes. The negave coeffcen on he average capal rao lkely sgnals he rskpreference of banks; relavely rsky banks have boh low capal raos and hgh volaly of ncome. 26 More accuraely, a Sharpe Rao subracs ou he rsk-free rae of reurn. If hs s consan across all banks, would no affec he resuls. 27 To creae hs fgure, I sored he nearly 15,000 banks by her average nonneres ncome shares and creaed 50 bns wh an equal number of banks (abou 300) n each. I hen ploed he average Sharpe Rao and Z-score for all banks n each bn o effecvely smooh he relaonshp. A smple regresson of he average Sharpe Rao or Z-score on he average nonneres ncome share shows a srongly negave and sascally sgnfcan relaonshp. 19
As expeced, radng ncome s he nonneres ncome componen mos hghly correlaed wh ncome growh volaly; hs s smlar o DeYoung and Roland (2001). Columns 1 and 2 sugges ha radng ncome rases he volaly of ne ncome growh, bu no he mean growh rae, whle fducary ncome s a low rsk/low reurn ha lowers boh he mean and he volaly of ne ncome growh. 29 The ROE regressons show a smlar sory. 30 Begnnng wh mean ROE as he dependen varable n column 3, he nonneres ncome shares are no sascally dfferen from zero (eher ndependenly or jonly) and he regresson has essenally no explanaory power. In fac, he only varable ha eners sgnfcanly s he coun varable (Bank Years), whch s a survvor effec as more profable banks end o operae for longer perods of me. Column 4 provdes weak evdence ha he nonneres ncome share s posvely correlaed wh he volaly of ROE, alhough here s agan essenally no explanaory power. The quadrac erm s negave bu quanavely small and sascally nsgnfcan, whch mples a posve relaonshp beween nonneres ncome shares and volaly of bank ROE. The only oher fnancal varable ha s sgnfcan s he negave coeffcen on he average leverage rao, whch agan sgnals rsk preferences. Wh he Sharpe Rao as he dependen varable (column 5), he negave and sgnfcan coeffcen on he nonneres ncome share effecvely summarzes he man fndng banks wh hgher nonneres ncome shares have lower profably per un of rsk. The coeffcens on he componens of nonneres ncome agan show ha fducary ncome s assocaed wh he mos prof per rsk, whle radng ncome offers he leas. Ths suppors some earler work ha poned ou he downsde of radng acves. Kwas (1989), Kwan (1998), and DeYoung and Roland (2001), for example, fnd ha secures or radng acves were assocaed wh hgher volaly of reurns, bu no necessarly hgher average reurns, whle Morgan and Sroh (2001) repor ndrec evdence as banks wh large radng posons pay hgher spreads n he deb marke. Fnally, he resuls wh he Z-score as he dependen varable (column 6) are que smlar. The Z-score falls as he nonneres ncome shares rses, whch ndcaes ha banks wh a large relance on nonneres ncome have hgher relave nsolvency rsk. Increased fducary ncome agan lowers rsk, whle ncreased fees and oher ncome rases. Tradng ncome eners wh a negave sgn, bu s no sascally sgnfcan. Agan, bank years s srongly posve, ndcang an mporan survvor effec. 28 The mean of he average nonneres ncome share for all banks s abou 15%. Maxmzng he fed value, y=72.6+129.3x-96.4x 2, gves a peak a abou 67%. 29 Resuls are smlar usng ne ncome before provsons n columns 1 and 2. 30 Resuls are smlar wh reurn on average asses (ROA), wh radng ncome excluded, or wh he sub-sample of non-jumpng banks. 20
The resuls n Table 5 and Fgure 8 show ha greaer relance on nonneres ncome has been assocaed wh hgher volaly of bank ncome and hgher rsk, bu no wh hgher reurns. Ths suggess ha he move oward nonneres ncome may acually be worsenng he rsk/reurn rade-off for he ypcal bank and no generang large dversfcaon benefs. Two caveas deserve menon, however. Frs, as menoned above, here s a poenal smulaney ssue here because a bank s focus on nonradonal acves s chosen by bank managers and s no exogenous. Second, hese resuls do no mean ha no banks are able o successfully manage he nonradonal ncome and mprove profs or lower rsk, bu raher ha hs s no he case for he ypcal bank. VI. Conclusons Ths paper explores he lnk beween he growng relance on nonneres ncome and he volaly of bank revenue and profs. Resuls from boh aggregae and bank daa provde lle evdence ha hs shf offers large dversfcaon benefs n he form of more sable profs or revenue. A he aggregae level, nonneres ncome s much more volale han more radonal ne neres ncome. Alhough ne operang revenue has n fac become less volale n he 1990s as nonneres ncome grew n mporance, hs can be drecly raced o he declnng volaly of ne neres ncome ha more han offse he ncreased conrbuon from he growng share of he relavely volale nonneres ncome. Tradng ncome, n parcular, shows enormous volaly. Moreover, ne neres ncome and nonneres ncome growh raes have become more hghly correlaed n he 1990s. A he bank level, nonneres ncome growh also shows an ncreased correlaon wh ne neres ncome over he las decade. Servce charges and fees n parcular are hghly correlaed wh ne neres ncome, whle radng and fducary ncome are less so. In erms of bank rsk and reurn, here s a clear negave assocaon beween nonneres ncome shares and profs per un of rsk. Tradng acves appear o be he bgges drag on prof per un of rsk and sugges ha connued expanson may ulmaely lower rsk-adjused reurns, whle fducary ncome s assocaed wh hgher prof per rsk and more sable ne ncome growh. These resuls rase fundamenal doubs abou he belef ha nonneres ncome wll sablze revenue and profably and hereby reduce rsk. Ne neres ncome and nonneres ncome growh are posvely correlaed for he ypcal bank, and he correlaon seems o be rsng for boh ndvdual banks and n he aggregae. Poenal explanaons nclude he ncreased focus on cross-sellng sraeges and ncreased loan commmens, boh of whch could expose dfferen busness lnes o he same shock and rase he correlaon of revenue sreams. 21
As a fnal pon, s worh menonng ha hese resuls do no mean ha no nsuon has benefed from a more dversfed revenue sream. Recall ha abou a hrd of he banks examned had a negave correlaon beween ne neres ncome and nonneres ncome growh. Raher, he hsorcal daa smply show hs has no been he case for he ypcal bank and ha appears ha poenal dversfcaon benefs are recedng as bank revenue sreams become more closely lnked. Moreover, he daa analyzed here cover a ranson perod, and s possble ha banks are sll n he early sages of learnng how o reap he benefs from broader and more dverse acves. Undersandng how banks maxmze he gans from a dversfed revenue porfolo s a crcal pece of nformaon for successful rsk managemen decsons and fuure emprcal work wll ry o sor ou hese alernave explanaons and dvergen oucomes. 22
Daa Appendx Aggregae Daa The Federal Depos Insurance Corporaon (FDIC) publshes annual daa on he U.S. bankng ndusry n s Hsorcal Sascs on Bankng daa, www.fdc.gov. These daa nclude balance shee and ncome saemen ems, and are aggregaes of all commercal banks. FDIC also provdes quarerly daa for mos varables from 1983:Q1 o 2001:Q3. Due o reporng ssues, quarerly daa pror o 1983 s no approprae for hs ype of analyss. For example, some small banks only fled regulaory repors n he 2 nd and 4 h quarer, whch makes quarerly analyss mpossble. The dealed componens of nonneres ncome, however, are only avalable on a conssen bass begnnng n 1984:Q1. All seres were deflaed usng he GDP prce deflaor and ransformed no consan 1996 dollars. All varables were seasonally adjused usng he Census Bureau s X11 procedure. Bank-level Daa Annual bank-level daa are aken from he Consoldaed Repor of Condon and Income ( Call Repors ) for all commercal banks n operaon n any year from 1978 o 2000. These annual daa were deflaed usng he GDP prce deflaor and ransformed no consan 1996 dollars. For each bank, he Call Repors have a unque code ha denfes he bank over me and allows calculaon of bank-specfc growh raes. In cases of mergers and acqusons, he acqurng bank s code s mananed and he arge drops from he sample. The avalably of daa for some of he ncome saemen ems changes hroughou he hsory of he Call Repors. In parcular, he four componens of nonneres ncome fducary ncome, servce charges, radng revenue, and fees and oher were no always conssenly defned over hs perod. These varables were defned and consruced from he Call Repor varable codes as follows: Fducary ncome ncludes gross ncome from servces rendered by he bank s rus deparmen or by any of s consoldaed subsdares acng n any fducary capacy,.e., admnserng nvesmens for ohers. Fducary ncome was conssenly defned for 1978-2000 as Call Repor em RIAD4070. Servce charges on depos accouns nclude charges for manenance of depos accouns, falure o mee mnmum balances, excess check wrng, whdrawals from nonransacon accouns, early whdraw or closure fees, dorman accouns, exensve acvy, ATM usage, bounced check charges, and oher fees. Servce charges was conssenly defned for 1978-2000 as Call Repor em RIAD4080. 23
Tradng revenue ncludes he ne gan or loss from radng cash nsrumens, off-balance shee dervave conracs, and sales of asses and oher fnancal nsrumens. Also ncluded are revaluaon o carryng value of asses and lables due o markng o marke, revaluaon of neres rae, foregn exchange, equy dervave, commody and oher conracs due o markng o marke, and ncdenal ncome and expense relaed o he purchase and sale of asses and lables. Tradng ncome was defned as RIADA220 for 1996-2000, and as RIAD4075 plus RIAD4077 for 1984-1995. Fees and oher ncome ncludes all oher nonneres ncome ems, such as servce charges, commssons, and fees no repored elsewhere. Ths ncludes fees for safe depos boxes, nsurance sales, bank drafs, money orders, ec., bll collecon, savngs bond redempon, execuon of accepances and leers of cred, morgage servcng fees, and noary, consulng or advsory servces), perodc cred card fees, merchan cred card charges, renal fees, and loan commmen fees. Also ncluded here are ne gans on sales of real esae, loans, or premses, daa processng servces, and sales of oher asses, as well as nonneres ncome on oher foregn ransacons. Fees and oher ncome was defned as RIAD5407 plus RIAD5408 for 1997-2000, RIAD5407 plus RIAD5408 plus RIAD4076 for 1991-1996, RIAD4076 plus RIAD4078 for 1984-1990. For 1978-1983, only radng ncome s ncluded n nonneres ncome for many banks, so he dealed breakdown s no possble. 24
References Acharya, Vral V., Ifekhar Hasan, and Anhony Saunders. The Effecs of Focus and Dversfcaon on Bank Rsk and Reurn: Evdence from Indvdual Bank Loan Porfolos. Mmeo, New Jersey Insue of Technology, March 15, 2002. Ahmed, Anwer S., Carolyn Takeda, and Shawn Thomas. Bank Loan Loss Provsonng: A Reexamnaon of Capal Managemen, Earnngs Managemen and Sgnalng Effecs. Journal of Accounng and Economcs, 28, 1999, 1-25. Boyd, John H. and S. L. Graham. The Profably and Rsk Effecs of Allowng Bank Holdng Companes o Merge wh Oher Fnancal Frms: A Smulaon Sudy. Quarerly Revew, Federal Reserve Bank of Mnneapols, 1988, Vol. 12, No. 2, Sprng, 3-20. Boyd, John H., S.L, Graham, and R. S. Hew. Bank Holdng Company Mergers wh Nonbank Fnancal Frms: Effecs on he Rsk of Falure. Journal of Bankng and Fnance, 1993, Vol. 17, 43-63. DeLong, Gayle L. Sockholder Gans from Focusng versus Dversfyng Bank Mergers. Journal of Fnancal Economcs, 59, 2001, 221-252. DeYoung, Rober and Karn P. Roland. Produc Mx and Earnngs Volaly a Commercal Banks: Evdence from a Degree of Toal Leverage Model. Journal of Fnancal Inermedaon, 2001, Vol. 10, 54-84. Greenwal, M. and J. Snkey Jr. Bank Loan Loss Provsons and he Income Smoohng Hypohess: An Emprcal Analyss, 1976-1984. Journal of Fnancal Servces Research, 1, 301-318. Houson, Joel F. and Mchael D. Ryngaer. The Overall Gans from Large Bank Mergers. Journal of Bankng and Fnance, 18, 1994, 1155-1176. Kwan, Smon. Rsk and Reurn of Banks Secon 20 Secures Afflaes. FRBSF Economc Leer, 98-32, Ocober 23 1998. Kwas, Myron. The Impac of Underwrng and Dealng on Bank Reurns and Rsk. Journal of Bankng and Fnance, 1989, Vol. 13, 101-125. Lown, Cara S., Carol L. Osler, Phlp E. Srahan, and Amr Suf. The Changng Landscape of he Fnancal Servce Indusry: Wha Les Ahead? Economc Polcy Revew, Federal Reserve Bank of New York, Ocober 2000, Vol. 6, No. 4, 39-54. Meer, Andrew P. and Tmohy J. Yeager. Are Small Rural Banks Vulnerable o Local Economc Downurns? Revew, Federal Reserve Bank of S. Lous, March/Aprl 2001, 25-38. Morgan, Donald P. and Kevn J. Sroh. Bond Marke Dscplne of Banks: The Asse Tes. Journal of Fnancal Servces Research, Vol. 20, No. 2/3, Ocober 2001, 195-208. Perez-Quros, Gabrel and Margare M. McConnell. Oupu Flucuaons n he Uned Saes: Wha Has Changed snce he Early 1980s? Amercan Economc Revew, December 2000, 90(5), 1464-76. Radeck, Lawrence J. Banks Paymens-Drven Revenue. Economc Polcy Revew, Federal Reserve Bank of New York, July 1999, 53-70. Rose, Peer S. Dversfcaon of he Bankng Frm. The Fnancal Revew, 24(2), May 1989, 251-280. 25
Rosen, Rchard J., Peer R. Lloyd-Daves, Myron L. Kwas, and Davd B. Humphrey. A Porfolo Analyss of Bank Invesmen n Real Esae. Journal of Bankng and Fnance, 13(3), July 1989, 355-366. Sanomero, Anhony W. and Eek-june Chung. Evdence n Suppor of Broader Bankng Powers. Fnancal Markes, Insuons, and Insrumens, 1(1), January 1992, 1-69. Saunders, Anhony and Ingo Walers. Unversal Bankng n he Uned Saes: Wha Could We Gan? Wha Could We Lose? 1994, New York, NY: Oxford Unversy Press. Templeon, Wllam K. and Jacobus T. Severens. The Effec of Nonbank Dversfcaon on Bank Holdng Companes. Quarerly Journal of Busness and Economcs, 31(4), Auumn 1992, 3-16. 26
Fgure 1: Rsng Share of Nonneres Income n Ne Operang Revenue 50 45 40 Percen 35 30 25 20 1984:Q1 1986:Q1 1988:Q1 1990:Q1 1992:Q1 1994:Q1 1996:Q1 1998:Q1 2000:Q1 Noe: Nonneres ncome as a share of ne operang revenue (nonneres ncome plus ne neres ncome). Source: Aggregae daa from FDIC.
Fgure 2: Nonneres Income s More Volale han Ne Ineres Income 60 50 40 Growh Rae (Annualzed Percen) 30 20 10 0 1984 1986 1988 1990 1992 1994 1996 1998 2000-10 -20-30 -40 Ne Ineres Income Noe: All esmaes are annualzed quarerly growh raes. Shaded areas are NBER recessons. Source: Aggregae daa from FDIC. Nonneres Income
Fgure 3: Tradng s he Mos Volale Type of Nonneres Income 140 120 100 80 Growh Rae (Annualzed Percen) 60 40 20 0 1985-20 1987 1989 1991 1993 1995 1997 1999 2001-40 -60-80 -100-120 -140 Fducary Income Servce Charges Tradng Income Fees and Oher Noe: All esmaes are four-quarer growh raes. Shaded areas are NBER recessons. Source: Aggregae daa from FDIC.
Fgure 4a: Impulse Response Funcon of Ne Ineres Income o an Innovaon n Real GDP 0.02 0.01 0.00 1 2 3 4 5 6 7 8 9 10-0.01 Quarers Noe: Response of ne neres ncome growh o a Cholesky one sandard devaon nnovaon n real GDP, esmaed from a VAR ncludng four lags of real GDP, ne neres ncome, and nonneres ncome. The VAR s esmaed n log-levels wh a rend for 1984:Q1 o 2001:Q3. Fgure 4b: Impulse Response Funcon of Nonneres Income o an Innovaon n Real GDP 0.03 0.02 0.01 0.00 1 2 3 4 5 6 7 8 9 10-0.01-0.02 Quarers Noe: Response of nonneres ncome growh o a Cholesky one sandard devaon nnovaon n real GDP, esmaed from a VAR ncludng four lags of real GDP, ne neres ncome, and nonneres ncome. The VAR s esmaed n log-levels wh a rend for 1984:Q1 o 2001:Q3.
0.9 Fgure 5: Cross-Seconal Correlaon beween Nonneres Income Growh and Ne Ineres Income Growh 0.8 0.7 0.6 Correlaon 0.5 0.4 0.3 0.2 0.1 0.0-0.1 1979 1982 1985 1988 1991 1994 1997 2000 All Banks Asses>$1B Non-jumpng Banks Noes: Plo of he cross-seconal correlaon, ρ, beween ne neres ncome and nonneres ncome for banks operang n each par of consecuve years.
Fgure 6: Cross-Seconal Correlaon beween Ne Income Growh and Growh n Componens of Nonneres Income 0.9 0.8 0.7 0.6 Correlaon 0.5 0.4 0.3 0.2 0.1 0.0-0.1 1979 1982 1985 1988 1991 1994 1997 2000 Fducary Income Servce Charges Tradng Income Fees and Oher Nonneres Income Noes: Plo of he cross-seconal correlaon, ρ, beween ne neres ncome and each componen of nonneres ncome for all banks operang n each par of consecuve years.
Char 7: Bank-Specfc Correlaons, All Banks.041716 Fracon 0-1 -.5 0.5 1 Rho
Fgure 8: Relaon beween Rsk Measures and Nonneres Income Share 3.5 40 3.0 35 Average Sharpe Rao 2.5 2.0 1.5 1.0 30 25 20 15 10 Z-score 0.5 5 0.0 1 6 11 16 21 26 31 36 41 46 Bns of Nonneres Income Shares Sharpe Rao (lef-axs) Z-score (rgh-axs) 0 Noe: Sharpe Rao s average reurn on equy (ROE) dvded by he sandard devaon of ROE. Z-score s he average reurn on asses (ROA) plus average Equy/Asses, dvded by he sandard devaon of ROA. Boh measures are averaged for all banks n each bn, where bns are creaed by sorng banks by her average nonneres ncome share and makng 50 equal szed groups.
Table 1: Commercal Bank Ne Operang Revenue, by Source, Year, and Bank Sze All Banks Asses >$10B Asses<=$10B 1980 1990 2000 1980 1990 2000 1980 1990 2000 Percen of Ne Operang Revenue Ne Operang Revenue 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Ne Ineres Income 79.6 67.5 56.6 73.5 58.6 52.2 82.6 74.3 66.9 Nonneres Income 20.4 32.5 43.4 26.5 41.4 47.8 17.4 25.7 33.1 Fducary Income 3.9 5.0 6.2 4.9 6.7 6.5 3.4 3.7 5.4 Servce charges 4.5 6.7 6.6 2.1 5.7 6.7 5.7 7.4 6.4 Tradng Revenue 1.5 2.8 3.5 3.1 6.0 4.9 0.7 0.4 0.2 Fees and Oher 10.5 18.0 27.1 16.3 23.1 29.7 7.6 14.1 21.1 Bllons of 1996 Dollars Ne Income 24.4 18.4 66.0 7.6 6.3 44.9 16.8 12.2 21.1 Ne Operang Revenue 123.4 196.9 333.7 41.0 86.0 233.1 82.4 110.8 100.6 Toal Asses 3,243.0 3,898.6 5,789.2 1,375.1 1,719.9 4,026.8 1,867.9 2,178.7 1,762.4 Number of Banks 14,523 12,370 8,388 31 66 79 14,492 12,304 8,309 Noes: Ne operang revenue s ne neres ncome plus nonneres ncome. Source: Consoldaed Repors of Condon and Income for a Bank wh Domesc and Foregn Offces - FFIEC 031.
Table 2: Decomposon of Varance of Ne Operang Revenue Varance/ Conrbuon Share Covarance o Varance 1984:Q1 o 1989:Q4 Ne Operang Revenue 50.38 Ne Ineres Income 72.0 100.18 51.9 Nonneres Income 28.0 228.89 18.2 Ne Ineres / Nonneres -28.95-11.7 1990:Q1 o 2001:Q3 Ne Operang Revenue 46.16 Ne Ineres Income 63.1 14.24 5.7 Nonneres Income 36.9 259.14 35.8 Ne Ineres / Nonneres 5.61 2.6 Noes: Share s he percen of ne operang revenue. Conrbuon o varance of ne operang revenue s measured n squared uns and defned n he ex n Equaon (1). Conrbuons do no exacly add up because shares are changng hrough he sub-perods. All daa are aggregaes for he U.S. bankng ndusry from FDIC.
Table 3: Lnk Beween Aggregae Bank Income Growh and GDP Growh Dependen Varables (X ) Bank Profs Tradonal Revenue Non-radonal Revenue Ne Ne Income Ne Ineres Ne In Inc Nonneres Nonneres Income Income plus Provsons Income less Provsons Income less Tradng X -1-0.366*** -0.526*** -0.178-0.813*** -0.309** -0.176 (0.118) (0.130) (0.133) (0.132) (0.131) (0.141) X -2-0.031-0.003-0.011-0.495*** -0.084-0.051 (0.119) (0.142) (0.136) (0.169) (0.136) (0.142) X -3 0.038 0.184-0.174-0.313* 0.076 0.023 (0.110) (0.137) (0.134) (0.169) (0.138) (0.143) X -4 0.137 0.110 0.095-0.114-0.228* -0.096 (0.094) (0.129) (0.128) (0.137) (0.127) (0.133) GDP 0.034-0.079 0.660 1.570 1.863* 1.616* (3.706) (1.276) (0.407) (6.902) (0.933) (0.813) GDP -1-7.014* -1.788-0.173 6.094-0.113-0.409 (3.604) (1.329) (0.430) (7.223) (0.971) (0.844) GDP -2-6.836* -1.204 0.162 2.388-1.751* -1.568* (3.637) (1.370) (0.446) (7.477) (0.983) (0.851) GDP -3-0.536 1.238-0.602-6.516 1.481 2.300*** (3.519) (1.316) (0.431) (7.265) (0.971) (0.827) GDP -4 3.525 2.750** 0.327 3.811 1.013-0.039 (3.227) (1.274) (0.403) (6.778) (0.948) (0.845) Consan 49.547*** 2.899 2.677-14.763 4.462 4.042 (17.633) (6.945) (2.119) (36.254) (5.152) (4.284) J. Sg. of Lagged X 0.005 0.001 0.271 0.000 0.015 0.659 J. Sg. of GDP and Lagged GDP 0.080 0.192 0.272 0.853 0.117 0.026 No. Obs 54 66 66 66 66 66 Adjused-R 2 0.287 0.195 0.039 0.344 0.183 0.111 Noes: All varables are quarerly growh raes. X, he dependen varable, s defned n he column headngs. Independen varables nclude four lags of he dependen varable (X -1 o X -4 ), conemporaneous GDP (GDP ), and four lags of GDP (GDP -1 o GDP -4 ). Daa are from 1984:Q1 o 2001:Q3. J. Sg. repors p- values assocaed wh he null hypohess ha he se of ndependen varables are jonly nsgnfcan. The ne ncome observaon has fewer observaons due o he quarers wh negave ncome ha prevened calculaon of growh raes. ***,**,* ndcae sascal sgnfcance a he 99%, 95%, and 90% level, respecvely.
Table 4: Deermnans of Correlaon beween Ne Ineres Income Growh and Nonneres Income Growh All Banks Non-Jumpng Banks Large Banks Log Asses 0.166 1.186*** -0.351 0.353 1.882 3.128* (0.346) (0.428) (0.326) (0.396) (1.825) (1.876) Equy/Asses 0.451** 0.540** 0.227 0.284 1.250 1.703** (0.230) (0.214) (0.235) (0.218) (0.819) (0.836) Growh n Asses 2.258*** 2.196*** 3.440*** 3.378*** 1.889*** 1.813*** (0.075) (0.076) (0.112) (0.113) (0.296) (0.306) Nonneres Income Share 89.143** 86.335** 59.972* 57.504** -54.839*** -32.709** (36.381) (33.952) (31.103) (28.969) (9.778) (13.000) Nonneres Income Share 2-124.069** -112.092** -68.329-59.506 5.586 11.945 (57.552) (54.053) (54.326) (50.637) (11.444) (11.807) Fducary Income Share -18.524*** -11.161** -27.371 (4.607) (4.749) (16.979) Tradng Income Share -12.577-25.727*** -34.951 (11.514) (7.406) (22.834) Fees & Oher Nonneres Income Share -18.058*** -14.013*** -44.863*** (4.673) (4.363) (15.452) Bank Years -0.487*** -0.513*** -0.274*** -0.283*** 1.823*** 1.650*** (0.064) (0.065) (0.072) (0.072) (0.419) (0.422) Mul 1.753** 1.584* 3.063*** 2.970*** 27.710*** 23.455** (0.873) (0.871) (0.959) (0.956) (10.000) (9.744) Consan 0.033 3.942-4.485-1.249-50.620** -34.621* (5.527) (5.740) (5.020) (5.253) (19.675) (20.692) Adjused-R 2 0.19 0.19 0.11 0.12 0.25 0.26 No. Obs. 14,503 14,503 11,601 11,601 463 463 J. Sg.: Nonneres Income Shares 0.006 0.000 0.000 0.000 0.000 0.000 J. Sg.: Componen Shares 0.000 0.000 0.030 Noes: Dependen varable s correlaon beween ne neres ncome growh and nonneres ncome growh for each bank, mulpled by 100. Robus sandard errors are n parenheses. Sample ncludes all banks wh a growh rae for boh ne neres ncome and nonneres ncome for a leas sx years. Non-jumpng banks have average asse growh raes above he 10h percenle or below he 90h percenle. Large banks have average asses greaer han $1 bllon. All ndependen varables are averaged for he number of perods he bank operaes, excep Mul whch s a dummy varable se equal o 1 f he bank s par of a mul-bank organzaon; equal o 0 oherwse. Nonneres ncome share s nonneres ncome dvded by ne operang revenue (ne neres ncome plus nonneres ncome). Fducary ncome share, radng ncome share, and fees and oher nonneres ncome shares are all as a share of oal nonneres ncome. J. Sg. repors p-values from he F-es of he jon sgnfcance of he followng ses of varables: nonneres ncome share and nonneres ncome share-squared; he hree componens shares of nonneres ncome. ***, **, * ndcaes sascal sgnfcance a he 99%, 95%, and 90% level, respecvely.
Table 5: Nonneres Income Shares as Deermnans of Bank Rsk and Reurn Dependen Varables Ne Income Growh Reurn on Equy (ROE) Mean Sd. Dev. Mean Sd. Dev. Sharpe Rao Z-score Log Asses 0.162-4.255*** 0.217-2.809 0.440*** 2.786*** (0.120) (0.401) (2.252) (6.270) (0.023) (0.168) Equy/Asses 0.126** -0.307* 0.386-3.124*** 0.059*** 0.940*** (0.054) (0.163) (0.279) (0.814) (0.012) (0.083) Growh n Asses 1.053*** 0.316*** -0.104-0.067-0.023*** -0.169*** (0.034) (0.086) (0.104) (0.306) (0.002) (0.022) Nonneres Income Share 19.140*** 129.273*** -21.508 117.236** -8.402*** -74.588*** (3.831) (16.782) (20.109) (57.311) (2.120) (12.244) Nonneres Income Share 2-17.057** -96.423*** 36.743-8.158 3.453 2.276 (6.925) (29.590) (33.937) (90.564) (3.375) (19.677) Fducary Income Share -2.489** -19.602*** 17.598-71.679 1.851*** 17.402*** (1.163) (4.333) (16.297) (45.705) (0.285) (2.069) Tradng Income Share 3.523 19.674* 4.793 11.717-2.084* -6.401 (2.616) (11.539) (14.637) (36.957) (1.089) (5.171) Fees & Oher Nonneres Income Share -0.856 8.974*** -22.079 72.205-0.466*** -3.197*** (0.595) (3.371) (19.176) (52.396) (0.122) (0.865) Bank Years -0.161*** -1.283*** 1.535** -5.243*** 0.050*** 0.456*** (0.023) (0.071) (0.642) (1.708) (0.004) (0.028) Mul -0.003-1.768* 3.293-11.239 0.394*** 0.554 (0.308) (1.009) (3.728) (11.049) (0.043) (0.389) Consan 0.190 72.626*** -18.450* 131.152*** 0.114 7.153*** (1.034) (3.626) (10.727) (32.476) (0.339) (2.066) Adjused-R 2 0.34 0.10 0.00 0.00 0.15 0.18 No. Obs. 11,697 11,697 14,500 14,500 14,500 14,500 J. Sg.: Nonneres Income Shares 0.000 0.000 0.489 0.044 0.000 0.000 J. Sg.: Componen Shares 0.033 0.000 0.686 0.100 0.000 0.000 Noes: Dependen varable are mean ne ncome growh, sandard devaon of ne ncome growh, mean reurn on equy (ROE), sandard devaon ROE, ROE-based Sharpe Rao (mean ROE / sandard devaon of ROE), and Z-score ((mean ROA+mean Equy/Asses) / sandard devaon of ROA). Robus sandard errors are n parenheses. Sample ncludes banks wh a growh raes for a leas sx years. All ndependen varables are averaged for he number of perods he bank operaes, excep Mul whch s a dummy varable se equal o 1 f he bank s par of a mul-bank organzaon; equal o 0 oherwse. Nonneres ncome share s nonneres ncome dvded by ne operang revenue (ne neres ncome plus nonneres ncome). Fducary ncome share, radng ncome share, and fees and oher nonneres ncome shares are all as a share of oal nonneres ncome. J. Sg. repors p-values from he F-es of he jon sgnfcance of he followng ses of varables: nonneres ncome share and nonneres ncome share-squared; he hree componen shares of nonneres ncome. ***, **, * ndcaes sascal sgnfcance a he 99%, 95%, and 90% level, respecvely.