Banks Non-Interest Income and Systemic Risk. October Abstract

Transcription

1 Banks Non-Interest Income and Systemc Rsk Markus K. Brunnermeer, a Gang Dong, b and Darus Pala c October 2016 Abstract Ths paper documents that banks wth hgher non-nterest ncome (noncore actvtes lke tradng and nvestment bankng and venture captal) have a hgher contrbuton to systemc rsk than tradtonal bankng (depost takng and lendng). After decomposng total non-nterest ncome nto two components (namely, tradng ncome, and nvestment bankng/venture captal ncome, respectvely), we fnd that only tradng ncome ncreases a bank s systemc rsk, tal rsk and nterconnectedness rsk. No sgnfcant relatonshp s found for nvestment bankng/venture captal ncome. We also fnd that banks wth hgher tradng ncome one-year pror to the crss earned lower returns durng the crss. No such sgnfcant effect was found for nvestment bankng/venture captal ncome. a Prnceton Unversty, NBER, CEPR, CESfo, b Columba Unversty, and c Rutgers Busness School, respectvely. We thank Vral Acharya, Lnda Allen, Turan Bal, Ivan Brck, Steve Brown, Doug Damond, Robert Engle, Cam Harvey, Kose John, Andreas Lehnert (dscussant), Thomas Ntschka (dscussant), Lasse Pedersen, George Pennacch (dscussant), Thomas Phlppon (dscussant), Matt Rchardson, Anthony Saunders, Anjan Thakor, Lorana Pelzzon (dscussant), Andre Shlefer, Rene Stulz, Rob Vshny, and semnar partcpants at the Amercan Fnance Assocaton Meetngs, CEPR/EBC/HEC/RofF/NYSE/ Euronext Conference on Fnancal Intermedaton and the Real Economy, CREDIT Conference n Stablty and Rsk Control n Bankng, Insurance and Fnancal Markets, European Economc Assocaton Meetngs, NBER Corporate Fnance and Rsks of Fnancal Insttutons Meetngs, FDIC Bank Research Conference, and semnars at Baruch, Fordham, Fed Board of Governors, Oxford, NYU, Rutgers, SMU, and Washngton Unversty n St. Lous for helpful comments and dscussons. All errors reman our responsblty. Correspondng author: Markus Brunnermeer, Prnceton Unversty, 26 Prospect Avenue, Prnceton, NJ 08540, markus@prnceton.edu.

2 These banks have become tradng operatons. It s the centre of ther busness. Phllp Angeldes, Charman, Fnancal Crss Inqury Commsson 1. Introducton The recent fnancal crss of was a showcase of large rsk spllovers from one bank to another heghtenng systemc rsk. But all bankng actvtes are not necessarly the same. One group of bankng actvtes, namely, depost takng and lendng make banks specal to nformaton-ntensve borrowers and crucal for captal allocaton n the economy. 1 However, pror to the crss, banks have ncreasngly earned a hgher proporton of ther profts from non-nterest ncome compared to nterest ncome. 2 Non-nterest ncome ncludes actvtes such as ncome from tradng and securtzaton, nvestment bankng and advsory fees, brokerage commssons, venture captal, and fducary ncome, and gans on non-hedgng dervatves. These actvtes are dfferent from the tradtonal depost takng and lendng functons of banks. In these actvtes banks are competng wth other captal market ntermedares such as hedge funds, mutual funds, nvestment banks, nsurance companes and prvate equty funds, all of whom do not have federal depost nsurance. For the 10-largest banks (by market captalzaton n 2000), Table 1 shows the average rato of non-nterest ncome to nterest ncome has ncreased from a value of 0.18 n 1989, to 0.57 n 2007, and to an even hgher 1.22% n Smlarly, Fgure 1 shows bg ncreases n the average non-nterest ncome to nterest ncome rato startng around 2000 and lastng to 2014 for all banks. Ths effect s more pronounced when we use a value-weghted portfolo than an equally-weghted portfolo. The latter panel shows that the ncrease n non-nterest ncome remans when we remove nvestment banks n the pre-crss perod. 3 *** Table 1 and Fgure 1 *** 1 Bernanke (1983), Fama (1985), Damond (1984), James (1987), Gorton and Pennach (1990), Calomrs and Kahn (1991), and Kashyap, Rajan, and Sten (2002) as well as the bank lendng channel for the transmsson of monetary polcy studed n Bernanke and Blnder (1988), Sten (1988) and Kashyap, Sten and Wlcox (1993) focus on ths role of bankng. 2 When we refer to nterest ncome we are usng net nterest ncome, whch s defned as total nterest ncome less total nterest expense (both of whch are dsclosed on a bank s Income Statement). 3 These are AIG, Amercan Express, Amerprse, Frst Amercan Corp., Frst Marblehead, Frankln Templeton, Goldman Sachs, Morgan Stanley, Raymond James Fnancal, Se Investment, Stfel Fnancal, and T. Rowe Prce. 1

3 Ths paper examnes the contrbuton of such non-nterest ncome to systemc bank rsk. In order to capture systemc rsk n the bankng sector we use two promnent measures of systemc rsk. The frst s the CoVaR measure of Adran and Brunnermeer (2016) (AB). AB defnes CoVaR as the value at rsk of the bankng system condtonal on an ndvdual bank beng n dstress. More formally, CoVaR s the dfference between the CoVaR condtonal on a bank beng n dstress and the CoVaR condtonal on a bank operatng n ts medan state. The second measure of systemc rsk s MES or the Margnal Expected Shortfall measure of Acharya, Pedersen, Phlppon, and Rchardson (2015) (APPR). APPR defne MES as a bank s stock returns when the market has the worst stock returns at the fve-percent level n the year. AAPR show that one can nfer what happens to a bank s captal n a real crss (what they call the systemc expected shortfall) when the market s n moderately bad days or MES. Note that CoVaR measures the externalty a bank causes on the system, whle MES focuses how much a bank s exposed to a potental systemc crss. Ths paper address four ponts: (1) We document the relatonshp between systemc rsk and a bank s non-nterest ncome? (2) We then decompose systemc rsk nto three dfferent components. Specfcally, we estmate the relatonshp of non-nterest ncome wth a bank s: tal rsk (alpha), exposure to fundamental macroeconomc and fnance factors (beta), and nterconnectedness (gamma), respectvely. (3) We categorze such tems nto two sub-groups, namely, tradng ncome, and nvestment bankng/venture captal ncome n order to examne f these two sub-groups have a sgnfcant effect on systemc rsk and ts three components. Analyzng these subcategores can only be done from 2001 onwards, snce only after then banks were requred to report detaled breakdowns of ther non-nterest ncome. (4) We examne f there s a relatonshp n the levels of pre-crss tradng ncome, and nvestment bankng/venture captal ncome, and the bank s stock returns earned durng the crss. (5) Fnally, we mplement a number of tests to ensure that the above results are robust. We fnd the followng results. 1. Systemc rsk s hgher for banks wth a hgher non-nterest ncome to nterest ncome rato. Specfcally, a one standard devaton ncrease to a bank s non-nterest ncome to nterest ncome rato ncreases ts systemc rsk contrbuton by 6.02% n CoVaR and 4.73% n MES. Ths suggests that actvtes that are not tradtonally lnked wth banks (such as depost takng and lendng) are assocated wth a larger contrbuton to systemc rsk. 2

4 2. Examnng the bank-specfc control varables we fnd that larger banks, more glamor banks (.e., wth hgher market-to-book ratos), banks wth a hgher proporton of C&I loans to total loans, and a lower proporton of agrcultural loans to total loans are assocated wth hgher systemc rsk. Banks wth hgher lqudty, leverage, proporton of real estate and consumer loans have an ambguous relatonshp wth systemc rsk based on whether systemc rsk s defned as CoVaR or MES. 3. We fnd that non-nterest ncome sgnfcantly ncreases tal rsk (alpha). A one standard devaton ncrease n non-nterest ncome results n a 15.9% ncrease n a bank s tal rsk. These results are consstent wth those of Stroh (2004, 2006) who fnds a postve relatonshp between non-nterest ncome and total bank return volatlty. We next examne beta, the relatonshp of between non-nterest ncome and a bank s exposure to fundamental macroeconomc and fnance factors. We fnd that non-nterest ncome s statstcally nsgnfcantly postvely related to beta, suggestng that non-nterest ncome does not lead to more severe co-movements wth macroeconomc and fnance factors. Ths result s nconsstent wth those of by DeJonghe (2010) who fnds a postve relatonshp. Fnally, we fnd that nonnterest ncome s postvely related to gamma, suggestng that non-nterest ncome leads to more systemc rsk due to nterconnectedness. A one standard devaton ncrease n non-nterest ncome results n a 60.4% ncrease n a bank s systemc rsk of beng nterconnected to other banks. 4. Based on data avalablty after 2001, we are able to decompose non-nterest ncome nto two components, namely, tradng ncome and nvestment bankng/venture captal ncome, respectvely. We fnd that only tradng ncome s postvely related to systemc rsk. A one standard devaton shock to a bank s tradng ncome ncreases CoVaR by 2.43%, tal rsk alpha by 8.20%, and nterconnectedness gamma by 74.1%, respectvely. No statstcally sgnfcant relatonshp s found between nvestment bankng/venture captal ncome and systemc rsk. 5. When we examne realzed ex post rsk, we fnd that banks wth hgher tradng ncome oneyear before the recesson earned lower returns durng the recesson perod. No such sgnfcant effect was found for nvestment bankng/venture captal ncome of commercal banks. We also fnd that larger banks earned lower stock returns durng the recesson as dd banks that had a 3

5 lower proporton of real estate loans n ther portfolo pror to the crss. Banks that dd well pror to the crss contnued to do well n the crss. 6. These results are robust to alternatve proxes of non-nterest ncome, excludng the largest banks and to defnng systemc rsks wth respect to the market portfolo. Our fndng that procyclcal non-tradtonal actvtes (such as tradng and prvate equty ncome) can ncrease systemc rsk s consstent wth a number of papers. In the model of Shlefer and Vshny (2010), actvtes where bankers have less skn n the game are overfunded when asset values are hgh whch leads to hgher systemc rsk. 4 Smlarly, Song and Thakor (2007) suggest that these transacton based actvtes can lead to hgher rsk. Whle not explctly focusng on tradtonal and nvestment bankng actvtes, Wagner (2010) theoretcally argues that dversfcaton can led to hgher systemc rsk because undertakng smlar actvtes ncrease the lkelhood of falng at the same tme. Our results are also consstent wth Fang, Ivashna and Lerner (2013) who fnd prvate equty nvestments by banks to be hghly procyclcal, and to perform worse than those of nonbank-afflated prvate equty nvestments. In secton 2 of ths paper we descrbe the related lterature, and Secton 3 explans our data and methodology. Secton 4 presents or emprcal results and n Secton 5 we conclude. 2. Related Lterature Recent papers have proposed complementary measures of systemc rsk other than CoVaR and MES. Bsas, Flood, Lo and Valavans (2012) provde an overvew of the growng numbers of systemc rsk measures. 5 Some papers such as Lehar (2005) and Gray and Jobst (2009) have employed a structural approach usng contngent clam analyss. Gven the strong assumptons that have to be made about a bank s lablty structure, other papers have used market data to back out reduced-form measures of market rsk. Allen, Bal and Tang (2012) propose the CATFIN measure whch s the prncpal components of the 1% VaR and expected shortfall, usng estmates of the generalzed Pareto dstrbuton, skewed generalzed error dstrbuton, and a non-parametrc dstrbuton. Tarashev, Boro, and Tsatsarons (2010) suggest Shapley values based on a bank s of default probabltes, sze, and exposure to common rsks could be used to assess regulatory taxes 4 Our non-tradtonal bankng actvtes are smlar to bankng actvtes such as loan securtzaton or syndcaton wheren the banker does not own the entre loan (d < 1 n ther model). 5 Gglo, Kelly and Prutt (2016) fnd that systemc rsk measures have a strong assocaton of the downsde rsk of future macroeconomc shocks. 4

6 on each bank. Bllo, et al. (2012) use prncpal components analyss and lnear and nonlnear Granger causalty tests and fnd nterconnectedness between the returns of hedge funds, brokers, banks, and nsurance companes. Chan-Lau (2010) proposes the CoRsk measure whch captures the extent to whch the rsk of one nsttuton changes n response to changes n the rsk of another nsttuton whle controllng for common rsk factors. Huang, Zhou, and Zhu (2009) derves the prce of nsurance aganst dstress as the bank s expected loss condtonal on the fnancal system beng n dstress exceedng a threshold level. Brownlees and Engle (2016) defne (SRISK) as the captal shortfall of a frm condtonal on a severe market declne, and s a functon of sze, leverage and rsk. Pror emprcal papers that have examned whether dversfcaton has been benefcal or detrmental to the rsk of an ndvdual bank (Saunders and Walter 1994, and DeYoung and Roland 2001 provde detaled lterature revews). Whle our study focuses on the effect of such dversfyng actvtes on a bank s contrbuton to systemc rsk, the lterature on ndvdual bank rsk shows mxed evdence. On the one hand, Stroh (2004, 2006) and Fraser, Madura, and Wegand (2002) fnd that non-nterest ncome s assocated wth more volatle bank returns. DeYoung and Roland (2001) fnd fee-based actvtes are assocated wth ncreased revenue and earnngs varablty. Acharya, Hassan and Saunders (2006) fnd dseconomes of scope when a rsky Italan bank expands nto addtonal sectors. On the other hand, Whte (1986) fnds that banks wth a securty afflate n the pre-glass Steagall perod had a lower probablty of default. In samples of nternatonal banks, Demurgc-Kunt and Huznga (2010) fnd that bank rsk decreases up to the 25 th percentle of non-nterest ncome and then ncreases, whereas De Jonghe (2010) fnds non-nterest ncome to monotoncally ncrease systemc tal rsk. All these studes focus on the rsk of a partcular bank, but not necessarly on the externalty a bank mposes on the fnancal system. A number of papers have used the CoVaR measure n other contexts. Among them are Wong and Fong (2011), who examne CoVaR for credt default swaps of Asa-Pacfc banks, whereas Gauther, Lehar and Souss (2012) use t for Canadan nsttutons. Zhou (2010) uses extreme value theory rather than quantle regressons to get a measure of CoVaR. 5

7 3. Data, Methodology, and Varables Used 3.1 Data We focus on all publcly traded bank holdng companes n the U.S., namely, wth SIC codes 60 to 67 (fnancal nsttutons) and flng Federal Reserve FR Y-9C report n each quarter. Ths report collects basc fnancal data from a domestc bank holdng company (BHC) on a consoldated bass n the form of a balance sheet, an ncome statement, and detaled supportng schedules, ncludng a schedule of off balance-sheet tems. By focusng on commercal banks we do not nclude nsurance companes, nvestment banks, nvestment management companes, and brokers. Our sample s from 1986 to 2014, and conssts of an unbalanced panel of 1,035 unque banks. 55 of these banks have zero non-nterest ncome n at least one year. We obtan a bank s daly equty returns from CRSP whch we use to convert nto weekly returns. Fnancal statement data s from Compustat and from Federal Reserve form FR Y-9C fled by a bank wth the Federal Reserve. T-bll and LIBOR rates are from the Federal Reserve Bank of New York and real estate market returns are from the Federal Housng Fnance Agency. The dates of recessons are obtaned from the NBER ( Detaled sources for each specfc varable used n our estmaton are gven n Table 2. *** Table 2 *** 3.2 Systemc rsk defnton usng CoVaR We descrbe below how we calculate the CoVaR measure of Adran and Brunnermeer (2016) (AB). Such a measure s calculated one perod forward and captures the margnal contrbuton of a bank to overall systemc rsk. AB stress that rather than usng a bank s rsk n solaton whch s typcally measured by ts VaR regulaton should also nclude the bank s contrbuton to systemc rsk measured by ther CoVaRs. Importantly, n order to avod procyclcalty and the volatlty paradox, regulaton should be based on relably observed varables that predct future CoVaRs (n our regressons by one-quarter ahead). 6

8 Value-at-Rsk (VaR) 6 measures the worst expected loss over a specfc tme nterval at a gven confdence level. In the context of ths paper, VaR q s defned as the percentage R of asset value that bank mght lose wth q% probablty over a pre-set horzon T : Probablty ( R VaR ) q (1) q Thus by defnton the value of VaR s negatve n general. 7 Another way of expressng ths s that VaR q s the q% quantle of the potental asset return n percentage term ( R ) that can occur to bank durng a specfed tme perod T. Consstent wth the prevous lterature and AB, we reverse the sgn for easy nterpretaton. The confdence level (quantle) q and the tme perod T are the two major parameters n a tradtonal rsk measure usng VaR. We consder 1% quantle and weekly asset return/loss R n ths paper, and the VaR of bank s Probablty ( R VaR ) 1%. 1% Let CoVaR denote the Value-at-Rsk of the entre fnancal system (portfolo) system q condtonal upon bank beng n dstress (n other words, the loss of bank s at ts level of VaR q ). That s, CoVaR whch essentally s a measure of systemc rsk s the q% quantle of system q ths condtonal probablty dstrbuton: Probablty R CoVaR R VaR q (2) system system ( q q), Smlarly, let CoVaR system medan denote the fnancal system s VaR condtonal on bank operatng q n ts medan state (n other words, the return of bank s at ts medan level). That s, CoVaR measures the systemc rsk when busness s normal for bank : system, medan q Probablty R CoVaR R medan q (3), ( system system medan q ) Bank s contrbuton to systemc rsk can be defned as the dfference between the fnancal system s VaR condtonal on bank n dstress ( CoVaR system q, VaR condtonal on bank functonng n ts medan state ( CoVaR system medan ): system system, medan q q q q ), and the fnancal system s CoVaR CoVaR CoVaR (4) 6 See Joron (2006) for a detaled defnton, dscusson and applcaton of VaR. 7 Emprcally the value of VaR can also be postve. For example, VaR s used to measure the nvestment rsk n an AAA coupon bond. Assume that the bond was sold at dscount and the market nterest rate s contnuously fallng, but never below the coupon rate durng the lfe the nvestment. Then the q% quantle of the potental bond return s postve, because the bond prce ncreases when the market nterest rate s fallng. 7

9 In the above equaton, the frst term on the rght hand sde measures the systemc rsk when bank s return s n ts q% quantle (dstress state), and the second term measures the systemc rsk when bank s return s at ts medan level (normal state). To estmate ths measure of ndvdual bank s systemc rsk contrbuton CoVaR q, we need to calculate two condtonal VaRs for each bank, namely CoVaR and system q CoVaR. system, medan q For the systemc rsk condtonal on bank n dstress ( CoVaR system q ), run a 1% quantle regresson 8 usng the weekly data to estmate the coeffcents R t t 1,, system, system and system Z (5) R Z R (6) system system system system system t t 1 t 1 : and run a 50% quantle (medan) regresson to estmate the coeffcents, medan and, medan : R Z (7), medan, medan, medan t t 1 where R t s the weekly growth rate of the market-value equty of bank at tme t : R t MV 1 (8) MV t t 1 and R s the weekly growth rate of the market-value equty of all N banks ( j 1,2,3..., N ) system t n the fnancal system at tme t : R MV R N system t 1 t t N 1 j MVt 1 j 1 (9) In equaton (8) and (9), MV t s the market value of bank s equty at tme t, and when we calculate the equty return of the entre fnancal system n equaton (9), the ndvdual bank s equty return s value-weghted by ts equty market value (MV). Z n equaton (7) s the vector of macroeconomc and fnance factors n the prevous t 1 week, ncludng market return, equty volatlty, lqudty rsk, nterest rate rsk, term structure, default rsk and real-estate return. We obtan the value-weghted market returns from the database of S&P 500 Index CRSP Indces Daly. We use the weekly value-weghted equty returns 8 See Koenker and Hallock (2001) and Koenker (2005) for a detaled explanaton about quantle regresson estmaton methodology. 8

10 (excludng ADRs) wth all dstrbutons to proxy for the market return. Volatlty s the standard devaton of log market returns. Lqudty rsk s the dfference between the three-month LIBOR rate and the three-month T-bll rate. For the next three nterest rate varables we calculate the changes from ths week t to t-1. Interest rate rsk s the change n the three-month T-bll rate. Term structure s the change n the slope of the yeld curve (yeld spread between the 10-year T-bond rate and the three-month T-bll rate. Default rsk s the change n the credt spread between the 10- year BAA corporate bonds and the 10-year T-bond rate. All nterest rate data s obtaned from the U.S. Federal Reserve webste and Compustat Daly Treasury database. Real estate return s proxed by the Federal Housng Fnance Agency s FHFA House Prce Index for all 50 U.S. states. ˆ, ˆ Hence we predct an ndvdual bank s VaR and medan equty return usng the coeffcents,, ˆ medan and ˆ, medan q, t t t 1 estmated from the quantle regressons of equaton (5) and (7): VaR Rˆ ˆ ˆ Z (10) R Rˆ ˆ ˆ Z (11), medan, medan, medan t t t 1 The vector of state (macroeconomc and fnance) varables Zt 1 s the same as n equaton (5) and (7). After obtanng the uncondtonal VaRs of an ndvdual bank VaR ( qt, ) and that bank s asset return n ts medan state ( R, medan t ) from equaton (10) and (11), we predct the systemc rsk condtonal on bank n dstress ( CoVaR system q ) usng the coeffcents ˆ system, ˆ system, ˆsystem estmated from the quantle regresson of equaton (6). Specfcally, CoVaR Rˆ ˆ ˆ Z ˆ VaR (12) system system system system system q, t t t 1 q, t Smlarly, we can calculate the systemc rsk condtonal on bank ( CoVaR system, medan q ) as: functonng n ts medan state Bank CoVaR ˆ ˆ Z ˆ R (13) system, medan system system system, medan q, t t 1 t s contrbuton to systemc rsk s the dfference between the fnancal system s VaR f bank s at rsk and the fnancal system s VaR f bank s n ts medan state: CoVaR CoVaR CoVaR (14) system system, medan q, t q, t q, t Note that ths s same as equaton (4) wth an addtonal subscrpt t to denote the tmevaryng nature of the systemc rsk n the bankng system. As shown n the quantle regressons of 9

11 equaton (5) and (7), we are nterested n the VaR at the 1% confdent level, therefore the systemc rsk of ndvdual bank at q=1% can be wrtten as: CoVaR CoVaR CoVaR (15) system system, medan 1%, t 1%, t 1%, t We hence also splt CoVaR qt, nto ts three components. system, medan ˆ ˆ, medan CoVaR ˆ ˆ ˆ q, t [( ) ( ) Zt 1] (16) wheren we defne alpha ˆ ˆ and, medan ( ) ˆ ˆ. Then, ( medan beta ) Z t 1 CoVaR, gamma ( alpha beta). (17) qt We can further nterpret alpha, beta, and gamma as follows: alpha captures bank s dosyncratc tal rsk that s ndependent of the (tme-varyng) macroeconomc and fnance factors Z; beta captures the tme-varyng component between tal dependency and central dependency that s drven by the macroeconomc and fnance rsk factors, and gamma measures the bank s nterconnectedness. alpha and beta measures a bank s mcro-prudental rsk, whereas gamma measures a bank s macro-prudental rsk per unt of mcro-prudental rsk. 3.3 Systemc rsk defnton usng MES Acharya, Pedersen, Phlppon and Rchardson (2015; APPR) propose a model-mpled measure of systemc rsk that they call margnal expected shortfall (MES) whch captures a bank s exposure gven that there s a systemc moderate crss n a gven year. AAPR show that the MES measure s able to predct the systemc expected shortfall that a bank faces n a real crss. 9 In general, MES ncreases n the banks expected losses durng a crss. Note that the MES reverses the condtonng. Instead of focusng on the return dstrbuton of the bankng system condtonal on the dstress of a partcular bank, MES focuses on the bank s return dstrbuton gven that the whole system s n dstress. AB s CoVaR framework refers to ths form of condtonng as exposure CoVaR, as t measures whch fnancal nsttuton s most exposed to a systemc crss, and not whch fnancal nsttuton contrbutes most to a systemc crss. Followng the emprcal analyss of APPR (2015), we estmate bank s MES at the 5% rsk level usng daly equty returns. The systemc crss event s the 5% worst days for the aggregate 9 APPR (2015) calculates the annual realzed systemc expected shortfall usng equty return data durng the crss. 10

12 equty return of the entre bankng system 10 n any gven year, and the average equty return of bank durng these worst market days s defned as bank s MES at the 5% level: MES 1 R (18) 5% t # dayst: systems n 5% tal 3.4 Regresson specfcatons and summary statstcs To nvestgate the relatonshp between the bank characterstcs and lagged bank s contrbuton to systemc rsk, we run OLS regressons wth quarterly fxed-effects of the ndvdual bank s systemc rsk contrbuton ( CoVaR or MES) on the bank s rato of non-nterest ncome to nterest ncome. In dong so, we control for the followng bank-specfc varables: the natural logarthm of total assets, fnancal leverage, market-to-book, lqudty, C&I loans to total loans, real estate loans to total loans, agrculture loans to total loans, and consumer loans to total loans. We focus on the mpact of bank s non-nterest ncome to nterest ncome rato on ts systemc rsk contrbuton. From 2001 onwards, we can decompose the non-nterest ncome to nterest ncome rato nto two components, namely, tradng ncome to nterest ncome, and nvestment bankng/venture captal ncome to nterest ncome. 11 Tradng ncome ncludes tradng revenue, net securtzaton ncome, gan (loss) of loan sales and gan (loss) of real estate sales. Investment bankng and venture captal ncome ncludes nvestment bankng and advsory fees, brokerage commssons and venture captal revenue. The detaled defntons and sources of the accountng ratos are lsted n Table 2. *** Table 2 *** Table 3 presents the summary statstcs. When we compare our results to those found n AB, we fnd that the average CoVaR of ndvdual banks to be slghtly hgher. Our average 10 To make an easy comparson wth our regressons usng the CoVaR measure, we defne systemc rsk as stock returns earned by all banks. Smlar results are obtaned for MES when we defne systemc rsk as stock returns earned by the entre market. 11 We also ncluded a component that ncluded all other non-nterest ncome tems such as fducary ncome, depost servce charges, net servcng fees, servce charges for safe depost box and sales of money orders, rental ncome, credt card fees, gans on non-hedgng dervatves. Ths component was not sgnfcant n any of the regressons so we dropped t from all our regressons. 11

13 (medan) CoVaR s 1.37% (1.23%) versus AB s average CoVaR s 1.17% (medan not reported). Comparng our results to APPR, we fnd an average (medan) MES of 3.58% (2.85%) for the years , whereas AAPR fnd an average (medan) SES of 1.63% (1.47%) for the crss perod July 2007 to December The correlaton between the two systemc rsk measures CoVaR and MES s 0.15, suggestng that these two measures capture some smlar patterns n systemc rsk. As n the prevous lterature, we also fnd that banks are hghly levered wth an average debt-to-asset rato of approxmately 89%. The average asset sze of the banks s $16 bllon and the medan asset sze s $1.8 bllon. We fnd that the average rato of non-nterest ncome to nterest ncome across all bank years to be 0.24, and the medan rato s *** Table 3 *** 4. Emprcal Results 4.1 Relatonshp of non-nterest ncome and systemc rsk We begn by regressng our measures of systemc rsk on the rato of non-nterest ncome to nterest ncome whle controllng for a number of bank-specfc varables, the results of whch are gven n Table 4. The dependent varables are the two measures of systemc rsk CoVaR and MES. Columns 1-2 are the CoVaR regressons, and columns 3-4 are the MES regressons. All ndependent varables are estmated wth a one quarter lag, and also nclude quarter fxed-effects whch are not reported. The t-statstcs are calculated usng Newey-West standard errors whch rectfes for heteroskedastcty. *** Table 4 *** We frst examne columns 1 and 3 where we only nclude our man varable of analyss, namely, the rato of non-nterest ncome to nterest ncome. In dong so, we ensure that our results are not due to some spurous correlaton between the varous ndependent varables. We fnd that the rato of non-nterest ncome to nterest ncome s sgnfcantly postvely related to both CoVaR and MES, suggestng that t contrbutes adversely to systemc rsk. In columns 2 and 4 we nclude the control varables to check f our results change. We stll fnd that non-nterest 12

14 ncome to nterest ncome rato s sgnfcantly postvely related to both CoVaR and MES, although ther economc magntude s smaller. Specfcally, a one standard devaton shock to a bank s non-nterest ncome to nterest ncome rato ncreases systemc rsk defned as CoVaR by 6.02%, and by 4.73% when systemc rsk s defned as MES. Examnng the bank-specfc control varables we fnd that larger banks, more glamor banks (.e., wth hgher market-to-book ratos), banks wth a hgher proporton of C&I loans to total loans, and a lower proporton of real estate loans to total loans are assocated wth hgher systemc rsk. Banks wth hgher lqudty, leverage, proporton of real estate and consumer loans have an ambguous relatonshp wth systemc rsk based on whether systemc rsk s defned as CoVaR or MES. That sad, our result of a postve statstcally sgnfcant relatonshp between non-nterest ncome and systemc rsk s unaffected by whch systemc rsk measure we use and whether we control or not control for other bank-specfc varables. We now examne f systemc rsk ncreased wth expansons n non-nterest ncome pror to the fnancal crss that started n Accordngly, we only nclude the sample years 1986 to 2006 and repeat our above regressons. The results are gven n Table 5. We stll fnd that nonnterest ncome to nterest ncome rato s sgnfcantly postvely related to both CoVaR and SES. When we examne the economc magntude the margnal effect s larger than for the entre sample. Specfcally, a one standard devaton shock to a bank s non-nterest ncome to nterest ncome rato ncreases systemc rsk defned as CoVaR by 8.78%, and by 13.76% when systemc rsk s defned as MES. *** Table 5 *** 4.2 Reverse causalty It s possble that the postve statstcally sgnfcant relatonshp between non-nterest ncome and systemc rsk s drven by reverse causalty. More specfcally, banks ncrease ther nvolvement n non-tradtonal bank actvtes when they are systemcally more mportant. Systemcally mportant banks mght have an ncentve to engage n rsker actvtes, because they reap hgh profts when the bet turns out to be successful, whle shftng part of the losses to other banks n case rsks materalze (potentally because of a too-many-to-fal problem). In order to tests for ths possblty, we splt our sample nto two sub-perods. We then regress the change n t non-nterest ncome to nterest ncome rato over the perod on the change n systemc 13

15 rsk from 1996 to If reverse causalty s the problem, one should see a postve and statstcally sgnfcant relatonshp. We control for the bank-specfc varables as of year endng 2000, although none of our results change sgnfcantly f we use the change n these varables from 1996 to 2000 (not reported). The results of such a regresson are gven n Table 6. We fnd a statstcally nsgnfcant relatonshp between the change n non-nterest ncome to nterest ncome rato over the perod and the change n systemc rsk from 1996 to Ths suggests that our prevous results do not suffer from the problem of reverse causalty. *** Table 6 *** 4.3 Relatonshp of non-nterest ncome and the dfferent components of systemc rsk We now use the decomposton of systemc rsk nto ts three dfferent components (equaton (17)). Specfcally, we estmate the relatonshp of non-nterest ncome to tal rsk (alpha), exposure to fundamental macroeconomc and fnance factors (beta), and nterconnectedness (gamma). The results of these regressons are gven n Table 7. *** Table 7*** We frst examne the relatonshp of non-nterest ncome to a bank s tal rsk or alpha. We fnd that non-nterest ncome sgnfcantly ncreases tal rsk. A one standard devaton ncrease n non-nterest ncome (0.305) results n a 15.9% ncrease n a bank s tal rsk. These results are consstent wth those of Stroh (2004, 2006) who fnds a postve relatonshp between non-nterest ncome and total bank return volatlty. We next examne beta, the relatonshp of between nonnterest ncome and a bank s exposure to fundamental macroeconomc and fnance factors. We fnd that non-nterest ncome s statstcally nsgnfcantly related to beta, suggestng that nonnterest ncome does not lead to more severe co-movements wth macroeconomc and fnance factors. Ths results s nconsstent wth those of by DeJonghe (2010) who fnds a postve relatonshp. Fnally, we examne the relatonshp between non-nterest ncome and a bank s nterconnectedness or gamma. We fnd that non-nterest ncome s postvely related to gamma, suggestng that non-nterest ncome does lead to more systemc rsk due to nterconnectedness. A 14

16 one standard devaton ncrease n non-nterest ncome (0.305) results n a 60.4% ncrease n a bank s systemc rsk of beng nterconnected to other banks. From 2001 onwards, we can decompose the rato of non-nterest ncome to nterest ncome nto tradng ncome to nterest ncome, and nvestment bankng and venture captal ncome to nterest ncome, respectvely. Federal Reserve form FR Y-9C only gves these detaled data after Tradng ncome ncludes tradng revenue, net securtzaton ncome, gan (loss) of loan sales and gan (loss) of real estate sales. Investment bankng and venture captal ncludes nvestment bankng and advsory fees, brokerage commsson and venture captal revenue. We fnd n Table 8 that only tradng ncome s statstcally sgnfcantly related to CoVaR and ts components. Investment bankng and venture captal ncome s generally statstcally nsgnfcantly related to CoVaR and ts components. A one standard devaton shock to a bank s tradng ncome ncreases CoVaR by 2.43%, tal rsk alpha by 8.20%, and nterconnectedness gamma by 74.1%, respectvely. *** Table 8 *** 4.4 Relatonshp of tradng ncome or nvestment bankng/venture captal ncome and crss returns We now examne f there s a relatonshp n the levels of pre-crss tradng ncome or nvestment bankng/venture captal ncome and the bank s stock returns earned durng the crss. Dong so, allows us to predct (usng the dfferent components of non-nterest ncome) bank performance durng the crss perod. Gven that the exstng lterature has yet to defne a wellaccepted explct emprcal proxy for ex ante systemc rsk, dong so also mtgates the crtcsm that measures of systemc rsk are prone to severe measurement ssues. We specfcally examne f banks wth hgher tradng or nvestment bankng/venture captal ncome n the one-year before the crss had more negatve returns durng the crss. In dong so, we are lookng at a sort of predctve regresson. We run a regresson wth the bank s stock return durng the latest recesson perod (defned by NBER as December 2007 to June 2009) as the dependent varable, and categorze banks by ther tradng ncome (or nvestment bankng/venture captal ncome) nto four quartles n the year before the latest recesson (2006Q3-2007Q3). We use a dummy varable for the hghest quartle of each component of non-nterest ncome the results of whch are gven n Table 9. We fnd sze to be negatvely related to crss returns, confrmng 15

17 the common ntuton that large banks got nto trouble n the crss. We also fnd that glamor banks,.e., those wth hgh market-to-book ratos n the pre-crss perod contnued to have better stock returns durng the crss. Not surprsngly, those banks that had a hgher proporton of ther loans n real estate n the pre-crss perod earned lower stock returns durng the crss. Importantly, we fnd that banks wth hgher tradng ncome one-year before the recesson earned lower returns durng the crss perod. No such sgnfcant effect was found for nvestment bankng and venture captal ncome. *** Table 9 *** In columns (3) and (4) we add two control varables to our regressons that have suggested to have resulted n bad performance durng the crss. The frst varable s pre-crss short-term fundng, defned as debt whose maturty s less than one year dvded by the sum of short-term debt, long-term debt, deposts and other labltes (Fahlenbrach, Prlmeer, and Stulz 2012). The second varable s the rato of loan commtments to loan commtments and total loans (Shockley and Thakor 1997, Gatev, Schuermann, and Strahan 2009). In columns (3) and (4) we fnd that the pre-crss level of short term fundng s nsgnfcantly related to the bank s returns durng the crss, whereas the pre-crss level of loan commtments s negatvely related to crss returns. In column (4), we stll fnd that banks wth hgher tradng ncome one-year before the recesson earned lower returns durng the recesson perod. No such sgnfcant effect was found for nvestment bankng and venture captal ncome Robustness tests We run a number of robustness tests. Frst, we examne f our result s drven by the numerator (non-nterest ncome) and not the denomnator (net nterest ncome). In the frst two columns of Table 10, we re-estmate our regressons usng the rato of non-nterest ncome to total assets nstead of non-nterest ncome to nterest ncome. In column (1), we fnd that non-nterest ncome remans postvely related to systemc rsk when we also nclude net nterest ncome to 12 We also examned the 18 frms that were analyzed by the Federal Reserve for captal adequacy n late February 2009 under the Supervsory Captal Assessment Program (SCAP). Our sample sze was reduced to 15 as three frms were not commercal banks (Goldman Sachs, Morgan Stanley, and Amercan Express). Gven the small sample sze of 15 we dd not fnd any sgnfcant results (results not reported but avalable from the authors). 16

18 total assets as a separate regressor. We also fnd that the rato of nterest ncome to total assets s nsgnfcantly related to systemc rsk. In column (2), we once agan fnd a postve relatonshp for tradng ncome, and an nsgnfcant relatonshp for nvestment bankng and venture captal ncome. These results suggest that t s non-tradtonal ncome (namely, non-nterest ncome) that contrbutes adversely to systemc rsk, and not tradtonal ncome (namely, net nterest ncome). *** Tables 10 *** Second, we examne f our results are drven by sze. Although we controlled for sze as n ndependent varable n our prevous regressons, we drop the top 5% by asset sze of the largest banks n each quarter and redo our regressons. The results of such an analyss are gven n columns (3) and (4) of Table 10. Consstent wth our prevous results we fnd that non-nterest ncome s postvely related to systemc rsk. We agan fnd that tradng ncome (nvestment bankng and venture captal ncome) s postvely (nsgnfcantly) related to systemc rsk. Ths suggests that our prevous results are not drven by the largest banks. Thrd, we examne f our results hold f we use CRSP total market stock returns (by calculatng the value-weghted return of all stocks lsted n CRSP monthly database for each calendar quarter) as our proxy for market rsk rather than the value-weghted bank stock portfolo. We re-estmate our regressons the results of whch are gven n Table 11. We fnd that nonnterest ncome and tradng ncome are agan postvely related to systemc rsk, whereas nvestment bankng and venture captal ncome s nsgnfcantly related to systemc rsk. 13 These results suggest that non-tradtonal ncome (namely, non-nterest ncome or tradng ncome) contrbutes adversely to systemc rsk whether we use the bank portfolo or total market portfolo as our proxy for market rsk. *** Tables 11*** 13 The results also hold f CRSP equty returns are equally-weghted. 17

19 5. Conclusons The recent fnancal crss showed that negatve externaltes from one bank to another created sgnfcant systemc rsk. Ths resulted n sgnfcant nfusons of funds from the Federal Reserve and the Treasury gven that depost takng and lendng make banks specal to nformaton-ntensve borrowers and for the bank lendng channel transmsson mechansm of monetary polcy. But banks have ncreasngly earned a hgher proporton of ther profts from non-nterest ncome from actvtes such as tradng, nvestment bankng, venture captal and advsory fees. Ths paper examnes the contrbuton of such non-nterest ncome to systemc bank rsk. Usng two promnent measures of systemc rsk ( CoVaR measure of Adran and Brunnermeer, forthcomng; MES measure of Acharya, Pedersen, Phlppon and Rchardson 2015), we fnd banks wth a hgher non-nterest ncome to nterest ncome rato have a hgher contrbuton to systemc rsk. Ths suggests that actvtes that are not tradtonally assocated wth banks (such as depost takng and lendng) are assocated wth larger systemc rsk. We also fnd that banks wth a hgher market-to-book rato, hgher proporton of C&I loans, lower proporton of agrcultural loans, and larger asset sze, ncrease systemc rsk. We fnd that non-nterest ncome sgnfcantly ncreases a bank s tal rsk and nterconnectedness. We do not fnd a statstcally sgnfcant relatonshp between non-nterest ncome and a bank s exposure to fundamental macroeconomc and fnance factors. Based on data avalablty after 2001, we are able to decompose non-nterest ncome nto two components, namely, tradng ncome, and nvestment bankng/venture captal ncome, respectvely. We fnd that only tradng ncome s postvely related to systemc rsk. A one standard devaton shock to a bank s tradng ncome ncreases CoVaR, tal rsk, and nterconnectedness. No statstcally sgnfcant relatonshp s found between nvestment bankng/venture captal ncome and systemc rsk. Fnally, we fnd that banks wth hgher tradng ncome one-year before the recesson earned lower returns durng the crss. No sgnfcant effect relatonshp s found for levels of nvestment bankng/venture captal ncome before the crss and stock returns durng the crss. 18

20 References Acharya, Vral, Iftekhar Hasan, and Anthony Saunders, 2006, Should Banks Be Dversfed? Evdence from Indvdual Bank Loan Portfolos, Journal of Busness 79, Acharya, Vral, Lasse Pedersen, Thomas Phlppon, and Mathew Rchardson, 2015, Measurng Systemc Rsk, NYU Stern Workng Paper. Allen, Lnda, Turan Bal, and Y Tang, 2012, Does Systemc Rsk n the Fnancal Sector Predct Future Economc Downturns?, Revew of Fnancal Studes 10, Adran, Tobas, and Markus Brunnermeer, 2016 forthcomng, CoVaR, Amercan Economc Revew. Bassett, Glbert and Roger Koenker, 1978, Asymptotc Theory of Least Absolute Error Regresson, Journal of the Amercan Statstcal Assocaton 73, Bernanke, Ben, 1983, Nonmonetary Effects of the Fnancal Crss n the Propagaton of the Great Depresson, Amercan Economc Revew 73, Bernanke, Ben, and Alan Blnder, 1988, Credt, Money, and Aggregate Demand, Amercan Economc Revew 78, Bllo, Monca, Mla Getmansky, Andrew Lo, and Lorana Pelzzon, 2012, Econometrc Measures of Connectedness and Systemc Rsk n the Fnance and Insurance Sectors, Journal of Fnancal Economcs 104, Bsas, Dmtros, Mark Flood, Andrew Lo, Stavros Valavans, 2012, "A Survey of Systemc Rsk Analytcs", Offce of Fnancal Research Workng Paper #0001. Brownlees, Chrstan, and Robert Engle, 2016, SRISK: A Condtonal Captal Shortfall Measure of Systemc Rsk, NYU Stern Workng Paper. Calomrs, Charles, and Charles Kahn, 1991, The Role of Demandable Debt n Structurng Optmal Bankng Arrangements, Amercan Economc Revew 81, Chan-Lau, Jorge, 2010, Regulatory Captal Charges for Too-Connected-to-Fal Insttutons: A Practcal Proposal, IMF Workng Paper 10/98. De Jonghe, Olver, 2010, Back to Bascs n Bankng? A Mcro-Analyss of Bankng System Stablty, Journal of Fnancal Intermedaton 19, Demrguc-Kunt, Asl and Harry Huznga, 2010, Bank Actvty and Fundng Strateges, Journal of Fnancal Economcs 98,

21 DeYoung, Robert, and Karn Roland, 2001, Product Mx and Earnngs Volatlty at Commercal Banks: Evdence from a Degree of Total Leverage Model, Journal of Fnancal Intermedaton 10, Damond, Douglas, 1984, Fnancal Intermedaton and Delegated Montorng, Revew of Economc Studes 51, Fahlenbrach, Rudger, Robert Prlmeer, and Rene Stulz, 2011, Ths Tme Is the Same: Usng Bank Performance n 1998 to Explan Bank Performance Durng the Recent Fnancal Crss, Journal of Fnance 67, Fama, Eugene, 1965, The Behavor of Stock Market Prces, Journal of Busness 37, Fama, Eugene, 1985, What's Dfferent About Banks?, Journal of Monetary Economcs 15, Fang, Lly, Vctora Ivashna, and Josh Lerner, 2013, Combnng Bankng wth Prvate Equty Investng, Revew of Fnancal Studes 26, Fraser, Donald, Jeff Madura, and Robert Wegand, 2002, Sources of Bank Interest Rate Rsk, Fnancal Revew 37, Gatev, Evan, Tl Schuermann. And Phl Strahan, 2009, Managng Bank Lqudty Rsk: How Depost Lendng-Lendng Synerges Vary wth Market Condtons, Revew of Fnancal Studes 22, Gauther, Celne, Alfred Lehar, and Moez Souss, 2012, Macroprudental Captal Requrements and Systemc Rsk, Journal of Fnancal Intermedaton 21, Gglo, Stefano, Bryan Kelly and Seth Prutt, 2016, Systemc Rsk and the Macroeconomy: An Emprcal Evaluaton, Journal of Fnancal Economcs 119, Gorton, Gary, and George Pennacch, 1990, Fnancal Intermedares and Lqudty Creaton, Journal of Fnance 45, Gray, Dale and Andreas Jobst, 2009, Tal Dependence Measures of Systemc Rsk Usng Equty Optons Data Implcatons for Fnancal Stablty,, IMF workng paper. Huang, Xn, Hao Zhou, and Habn Zhu, 2009, A Framework for Assessng the Systemc Rsk of Major Fnancal Insttutons, Journal of Bankng and Fnance 33, James, Chrstopher, 1987, Some Evdence on the Unqueness of Bank Loans, Journal of Fnancal Economcs 19, Joron, Phlppe, 2006, Value at Rsk: The New Benchmark for Managng Fnancal Rsk, McGraw-Hll. 20

22 Kashyap, Anl, Raghuram Rajan, and Jeremy Sten, 2002, Banks as Lqudty Provders: An Explanaton for the Coexstence of Lendng and Depost-Takng, Journal of Fnance 57, Kashyap, Anl, Jeremy Sten, and Davd Wlcox, 1993, Monetary Polcy and Credt Condtons: Evdence from the Composton of External Fnance, Amercan Economc Revew 83, Koenker, Roger, 2005, Quantle Regresson, Cambrdge Unversty Press. Koenker, Roger, and Glbert Bassett, 1978, Regresson Quantles, Econometrca 46, Koenker, Roger, and Kevn Hallock, 2001, Quantle Regresson, Journal of Economc Perspectves 15, Lehar, Alfred, 2005, Measurng Systemc Rsk: A Rsk Management Approach, Journal of Bankng and Fnance 29, Offcer, R. R., 1972, The Dstrbuton of Stock Returns, Journal of the Amercan Statstcal Assocaton 67, Praetz, Peter, 1972, The Dstrbuton of Share Prce Changes, Journal of Busness 45, Shlefer, Andre, and Robert Vshny, 2010, Unstable Bankng, Journal of Fnancal Economcs 97, Shockley, Rchard and Anjan Thakor, 1997, Bank Loan Commtment Contracts: Data, Theory, and Tests, Journal of Money, Credt, and Bankng 4, Song, Fenghua and Anjan Thakor, 2007, Relatonshp Bankng, Fraglty, and the Asset-Lablty Matchng Problem, Revew of Fnancal Studes 6, Sten Jeremy, 1998, An Adverse-Selecton Model of Bank Asset and Lablty Management wth Implcatons for the Transmsson of Monetary Polcy, RAND Journal of Economcs 29, Stroh, Kevn, 2004, Dversfcaton n Bankng: Is Nonnterest Income the Answer?, Journal of Money, Credt and Bankng 36, Stroh, Kevn, 2006, A Portfolo Vew of Bankng wth Interest and Nonnterest Actvtes, Journal of Money, Credt, and Bankng 38, Tarashev, Nkola, Claudo Boro, and Kostas Tsatsarons, 2010, Attrbutng Systemc Rsk to Indvdual Insttutons: Methodology and Polcy Applcatons, BIS Workng Papers 308. Wagner, Wolf, 2010, Dversfcaton at Fnancal Insttutons and Systemc Crses, Journal of Fnancal Intermedaton 19,

23 Wong, Alfred, and Tom Fong, 2011, Analyzng Interconnectvty among Economcs, Emergng Markets Revew 12, Zhou, Chen, 2009, Are Banks too Bg to Fal? Measurng Systemc Importance of Fnancal Insttutons, Internatonal Journal of Central Bankng 6,

24 Fgure 1. Average non-nterest ncome to nterest ncome rato and CoVaR Includng bank holdng companes that were nvestment banks pror to Equally Weghted N2I Market Value Weghted N2I Market Value Weghted CoVaR dcovar Year Excludng bank holdng companes that were nvestment banks pror to These nvestment banks are AIG, Amercan Express, Amerprse, Frst Amercan Corp., Frst Marblehead, Frankln Templeton, Goldman Sachs, Morgan Stanley, Raymond James Fnancal, Se Investment, Stfel Fnancal, and T. Rowe Prce Equally Weghted N2I Market Value Weghted N2I Market Value Weghted CoVaR dcovar Year

25 Table 1: Rato of non-nterest ncome to nterest ncome for the 10-largest banks Bank Name Ctgroup Bank of Amerca Chase Wachova Wells Fargo Suntrust US Bank Natonal Cty Bank of New York Mellon PNC Fnancal Captal One Fnancal 0.22 State Street 2.93 Average Non-nterest ncome rato to nterest ncome rato (N2I) s defned below and the data are taken from the Federal Reserve Bank reportng form FR Y9C: Nonnterest Income BHCK4079 N2I Net Interest Income BHCK4107 Ctgroup was Ctbank n 1989 before the merger wth Travelers Group. Bank of Amerca was called BankAmerca n 1989 before the merger wth NatonsBank. US Bank was Frst Bank System n 1989 before the combnaton wth Colorado Natonal Bank and West One Bank. Bank of New York Mellon was called Bank of New York n 1989 before the merger wth Mellon Fnancal. 24

26 Table 2: Varable defntons and sources Varable Name Calculaton Sources CoVaR Fnancal nsttuton s contrbuton to systemc rsk From equaton (15) MES Margnal expected shortfall From equaton (18) R R s Weekly equty return of ndvdual bank Weekly equty return of all banks MV MV t t 1 j 1 MV MV t 1 j t 1 R CRSP Daly Stocks, Compustat Fundamentals Quarterly CRSP Daly Stocks, Compustat Fundamentals Quarterly Market-to-book Market value of equty / book value of equty CRSP Daly Stocks, Compustat Fundamentals Quarterly Market value of equty Prce shares outstandng CRSP Daly Stocks Leverage Total assets / book value of equty Compustat Fundamentals Quarterly Logarthm of total book assets Log (total assets) U.S. Federal Reserve FRY-9C Report Non-nterest ncome to nterest ncome Tradng ncome to nterest ncome IBVC ncome to nterest ncome Non-nterest ncome / nterest Income Tradng ncome ncludes tradng revenue, net securtzaton ncome, gan(loss) of loan sales and gan(loss) of real estate sales. (2001 onwards) IBVC ncome ncludes nvestment bankng/advsory fee, brokerage commsson and venture captal revenue. (2001 onwards) U.S. Federal Reserve FRY-9C Report U.S. Federal Reserve FRY-9C Report U.S. Federal Reserve FRY-9C Report 25