Global Banks, Financial Shocks and International Business Cycles: Evidence from an Estimated Model

Transcription

1 Crawford School of Public Policy CAMA Cenre for Applied Macroeconomic Analysis Global Banks, Financial Shocks and Inernaional Business Cycles: Evidence from an Esimaed Model CAMA Working Paper 30/2013 May 2013 Rober Kollmann, ECARES, Universié Libre de Bruxelles, CEPR and Cenre for Applied Macroeconomic Analysis (CAMA), Ausralian Naional Universiy Absrac This paper esimaes a wo-counry model wih a global bank, using US and Euro Area (EA) daa, and Bayesian mehods. The esimaed model maches key US and EA business cycle saisics. Empirically, a model version wih a bank capial requiremen ouperforms a srucure wihou such a consrain. A loan loss originaing in one counry riggers a global oupu reducion. Banking shocks maer more for EA macro variables han for US real aciviy. Banking shocks accoun for abou 3%-5% of he uncondiional variance of US GDP and for 4%-14% of he variance of EA GDP. During he Grea Recession ( ), banking shocks accouned for abou 12%-20% of he fall in US and EA GDP, and for more han a hird of he fall in EA invesmen and employmen. THE AUSTRALIAN NATIONAL UNIVERSITY

2 Keywords financial crisis, global banking, real aciviy, invesmen, Bayesian economerics JEL Classificaion F36, F37, E44, G21 Address for correspondence: (E) The Cenre for Applied Macroeconomic Analysis in he Crawford School of Public Policy has been esablished o build srong links beween professional macroeconomiss. I provides a forum for qualiy macroeconomic research and discussion of policy issues beween academia, governmen and he privae secor. The Crawford School of Public Policy is he Ausralian Naional Universiy s public policy school, serving and influencing Ausralia, Asia and he Pacific hrough advanced policy research, graduae and execuive educaion, and policy impac. THE AUSTRALIAN NATIONAL UNIVERSITY

3 Global Banks, Financial Shocks and Inernaional Business Cycles: Evidence from an Esimaed Model Rober Kollmann () ECARES, Universié Libre de Bruxelles, CEPR and CAMA May 15, 2013 This paper esimaes a wo-counry model wih a global bank, using US and Euro Area (EA) daa, and Bayesian mehods. The esimaed model maches key US and EA business cycle saisics. Empirically, a model version wih a bank capial requiremen ouperforms a srucure wihou such a consrain. A loan loss originaing in one counry riggers a global oupu reducion. Banking shocks maer more for EA macro variables han for US real aciviy. Banking shocks accoun for abou 3%-5% of he uncondiional variance of US GDP and for 4%-14% of he variance of EA GDP. During he Grea Recession ( ), banking shocks accouned for abou 12%-20% of he fall in US and EA GDP, and for more han a hird of he fall in EA invesmen and employmen. Key words: financial crisis, global banking, real aciviy, invesmen, Bayesian economerics. JEL codes: F36, F37, E44, G21 () Address: R. Kollmann, ECARES, CP 114, Universié Libre de Bruxelles; 50 Av. Franklin Roosevel; B-1050 Brussels, Belgium; rober_kollmann@yahoo.com. I am graeful o Ken Wes and o wo referees, as well as o Alejandro Jusiniano, Mahias Pausian, Werner Roeger and Raf Wouers for advice. Mahias Pausian conribued o his projec a an early sage I also hank him for compuer code. Helpful suggesions were also received from Fabian Lipinsky, Giovanni Lombardo, Jochen Mankar, Olivier Pierrard, Gerno Müller, Alan Suherland, Chrisoph Thoenissen, Skander Van den Heuvel, Egon Zakrajšek and paricipans a workshops a he SWIM, SED and AEA meeings, Bonn, S.Gallen, London Business School, Cenral Bank of Luxembourg, Naional Bank of Belgium, Bank of France and Bank of England. I hank Egon Zakrajšek for providing me wih daa on he excess bond premium, US business loans and loan capaciy. 1

4 1. Inroducion The recen financial crisis began in US financial markes in 2007 and was quickly and srongly ransmied o Europe and oher pars of he world. The crisis revealed he fragiliy of major financial insiuions, and led o he wors global recession since he Grea Depression. These dramaic evens require a rehinking of he role of financial inermediaries for real aciviy. Before he financial crisis, sandard applied macro models absraced from financial inermediaries (e.g., Chrisiano e al. (2005)). The crisis revealed he sark limiaions of hose models. The crisis has simulaed much research ha incorporaes banks ino dynamic sochasic general equilibrium (DSGE) models. Given he global naure of he banking indusry, and of he financial crisis, ha research has frequenly focused on open economy models; see, for example, Devereux and Suherland (2011), Kamber and Thoenissen (2011), Kollmann, Enders and Müller (2011), Perri and Quadrini (2011) and Van Wincoop (2011). Closed economy DSGE models wih banks were, i.a., presened by Aikman and Pausian (2006), Van den Heuvel (2008), Gerali e al. (2010) and Meh and Moran (2010). 1 In his new class of DSGE models, bank capial is a key sae variable for he supply of credi, and for real aciviy; negaive shocks o bank capial are prediced o increase he spread beween banks lending and deposi raes, and o rigger a fall in bank credi and oupu; wih a globalized banking sysem, a loan loss in one counry can hus lead o a worldwide recession. So far, his new macro-banking lieraure has mainly used calibraed models a sysemaic empirical evaluaion, using economeric mehods, is necessary, o guide furher model building and policy. In order o provide an empirical assessmen of he role of banks as a source of shocks and as a ransmission channel in he global economy, he paper here esimaes (using Bayesian mehods) a wo-counry DSGE model wih a global bank. Quarerly US and Euro Area (EA) macro daa and banking daa (bank loans, bank 1 Oher open economy models wih banks can be found in Correa e al. (2010), Davis (2010), Nguyen (2011), Andreasen e al. (2010), Ueda (2011), Dedola and Lombardo (2012) and Lipinsky (2012), while closed economy DSGE models wih banks were also presened by Brunnermeier and Sannikov (2010), de Walque e al. (2010), Gerler and Karadi (2011), Gerler and Kiyoaki (2011), Iacoviello (2010), Del Negro e al. (2011), He and Krishnamurhy (2012), Dewacher and Wouers (2012) and Kollmann e al. (2012). 2

5 capial raio, loan spread) for he period 1990q1-2010q3 are used. 2 Specifically, I ake he Kollmann, Enders and Müller (2011) wo-counry model wih a banking secor o he daa ha srucure is used as i feaures a bank capial channel ha is broadly represenaive of hose in oher recen macro-banking models. The srucure builds on he Inernaional Real Business Cycle (RBC) lieraure, bu while sandard Inernaional RBC models assume direc fricionless inernaional borrowing and lending (e.g., Backus e al. (1992), Baxer and Crucini (1995), Kollmann (1996)), he model here assumes ha a global bank inermediaes beween savers and borrowers in he wo counries. Imporanly, he bank has o finance a fracion of her asses using equiy (own funds). This capial requiremen can reflec legal consrains and, more broadly, marke pressures. I implies ha he loan rae spread (relaive o he deposi rae) is a decreasing funcion of bank capial. The esimaed model assumes demand and supply shocks in home and foreign labor and good markes. In addiion, here are sochasic loan losses (defauls) in he wo counries, and shocks o he required (arge) bank capial raio henceforh, I refer o hese shocks as banking shocks. The esimaion resuls sugges ha he bank capial requiremen, and he banking shocks, maer for he dynamics of macro and banking variables. A model wih hese ingrediens ouperforms a model varian wihou an operaive bank capial requiremen (and wihou banking shocks)--he model wih he bank capial requiremen generaes prediced second momens of key macro and banking variables ha are closer o empirical momens (he marginal likelihood of ha model is markedly higher). In he presence of a bank capial requiremen, banking shocks induce synchronized responses of Home and Foreign real aciviy. Tha posiive inernaional ransmission mechanism is no presen in sandard Inernaional RBC models (wihou banks). According o baseline model esimaes, a one percenage poin fall in he global bank capial raio raises he loan rae spread by abou 20 basis poins. An unanicipaed US loan loss worh 1% of seady sae GDP lowers US and EA GDP by abou 0.10% and 2 Some previous papers have esimaed open economy DSGE models wihou banks; wo-counry models were esimaed by de Walque e al. (2005), Rabanal and Tuesa (2006) and Le e al. (2010) who also used UE and EA daa, and by Jacob and Peersman (2011). Small open economy models were esimaed by Adolfson e al. (2009) and Jusiniano and Preson (2010). 3

6 0.12%, respecively, on impac; a EA loan loss of equal size lowers US and EA GDP by 0.14% and 0.18%, respecively. A US loan loss hus lowers EA real aciviy more han US real aciviy. An unanicipaed increase in he required bank raio by one percenage poin lowers US and EA GDP by 0.10% and 0.11%, respecively, on impac. The esimaed banking model maches key cyclical properies of US and EA macro and banking variables. In paricular, i capures he fac ha US and EA loans are procyclical, while he loan spread is counercyclical. However, in he banking model here--as in sandard Inernaional RBC models (wihou banks)--a posiive shock o home TFP raises home GDP, bu lowers foreign GDP. Like sandard Inernaional RBC models, he presen model can only capure he high cross-counry correlaions of real aciviy seen in he daa, if TFP and oher non-banking shocks are highly posiively correlaed across counries. In mos model varians considered in his paper, banking shocks accoun for a non-negligible share of he variance of real aciviy. Specifically, banking shocks explain abou 3%-5% of he variance of US GDP, and 4%-14% of he variance of EA GDP. The variance share of invesmen accouned for by banking shocks is higher, especially for EA invesmen (above 20%). Thus, banking shocks maer more for EA real aciviy han for US real aciviy. US loan losses accoun for a greaer share of he variance of EA real aciviy han of he variance of US real aciviy. Exposure o US loan losses (via he global banking sysem) hus deepened he recen recession in he EA. Banking shocks conribued noiceably o he Grea Recession of , bu were no is dominan cause: banking shocks accouned for abou 12%-20% of he fall in US and EA GDP during he recession--bu hey accouned for more han a hird of he fall in EA invesmen and employmen. During he previous wo US recessions in he esimaion period ( and 2001), banking shocks accouned for a roughly similar share of he fall in US oupu, invesmen and employmen as in he recession. I consider several empirical measures of credi and lending spreads and find ha he key resuls are robus across he differen measures. This paper is complemenary o Gerali e al. (2010) who esimaed (using Euro Area daa) a closed economy New Keynesian macro model wih a banking secor ha faces a bank capial requiremen. The paper here differs (iner alia) from Gerali e al. by considering a real (flex-price) wo-counry world wih a global bank ha experiences loan 4

7 loss shocks and shocks o her required capial raio. By conras, he empirical analysis of Gerali e al. focuses on he role of shocks o bank mark-ups and o borrowers collaeral consrains (hese auhors do no consider loan loss shocks or shocks o he required bank capial raio). Secion 2 presens he model. Secion 3 discusses he economeric approach. Secion 4 describes key daa feaures. Secion 5 repors he esimaion resuls. Secion 6 concludes. 2. A wo-counry world wih a global financial inermediary As menioned above, his paper akes he heoreical wo-counry model of Kollmann e al. (2011) o US and EA daa. 3 In each of he wo counries, called Home (H) and Foreign (F), here is a represenaive worker, an enrepreneur and a governmen. A global bank collecs deposis from workers, and makes loans o enrepreneurs, in boh counries. The bank faces a capial requiremen: a fracion of bank asses has o be financed using he bank s own funds (equiy). Enrepreneurs produce a homogenous radable good ha is used for consumpion and for capial accumulaion. All agens are infiniely-lived. Markes are compeiive. Preferences and echnologies have he same srucure in boh counries. The following exposiion focuses hus on he Home counry. Foreign variables are denoed by an aserisk Preferences, echnologies, markes The Home worker The Home worker provides labor o he Home enrepreneur and invess her savings in one-period bank deposis. Her dae budge consrain is: where S C and C D T N DR, (1) S S D 1 N are he worker s consumpion and hours worked respecively. he real wage rae. D 1 is he bank deposi held by he saver a he end of period. is D R is 3 Governmens, and a rich se of (banking and non-banking) shocks are added o he srucure in Kollmann e al. (2011), in order o permi an empirical evaluaion of he bank capial channel, and provide esimaes of he conribuion of banking shocks o hisorical ime series for key macro aggregaes. Also, he presen paper allows for asymmeries beween counries, in order o capure differences beween he sensiiviy of he US and EA economies o banking shocks. 5

8 he gross ineres rae on deposis, beween -1 and. S dae expeced life-ime uiliy, V, is: V u( C ) u( D ) N E V, S S D N S S S T is a lump sum ax. The worker s wih x D ux () ( 1), 0 and 0. The worker s marginal disuiliy of labor, 0, is an exogenous random variable. N N will be referred o as he Home labor supply shock. Noe ha deposis provide uiliy o he worker (liquidiy services). This ensures ha, in equilibrium, he deposi rae is smaller han he loan rae, and ha workers hold deposis while enrepreneurs borrow. The worker s subjecive discoun facor is S S S S S S S decreasing in her fuure consumpion: 1 ( C 1), wih 0 ( C 1) 1, '( C 1) 0. The subjecive discoun facors of oher agens are likewise decreasing funcions of heir own consumpion. 4 Agens rea heir subjecive discoun facors as given, i.e. hey do no inernalize he effec of consumpion on he discoun facor I hus wrie he argumen of he subjecive discoun facor wih an upper-bar. I is assumed ha all agens have he same seady sae rae of ime preference, and he same risk aversion coefficien,. The Home worker maximizes her life-ime uiliy subjec o he period-by-budge consrain (1). Tha decision problem has hese firs-order condiions: S N u'( C ), (2) R E u'( C )/ u'( C ) u'( D )/ u'( C ) 1. (3) D S S S D S The Home enrepreneur The Home enrepreneur accumulaes physical capial and uses capial and local labor o produce oupu. Her echnology is Z K N 0 1, where Z, K and 1 ( ) ( ), N are oupu, capial and labor, respecively. Toal facor produciviy (TFP), 0, is an exogenous random variable. The law of moion of he capial sock is K 1 (1 ) K I, where 0 1is he capial depreciaion rae and I is gross invesmen. 0 is an exogenous random shock o invesmen efficiency (Fischer (2006), Jusiniano e al. 4 The endogenous discoun facors induce mean-reversion in individual wealh, and hus ensures saionariy (Kollmann (1991); Schmi-Grohé and Uribe (2003)). The numerical soluion mehod (local approximaion) and he esimaion mehod require saionariy. 6

9 (2008)). Gross invesmen is generaed using oupu. Le I ( I / I) be he amoun of oupu needed o generae I, where I is seady sae invesmen, and is an increasing, sricly convex funcion wih '(1) 1. Henceforh, variables wihou ime subscrips denoe seady sae values. The Home enrepreneur s period budge consrain is: where LR I I N d T L K N L E E 1 ( / ) 1 ( ) ( ), (4) L is a one-period bank loan received by he Home enrepreneur in period -1. is he gross ineres rae on ha loan, se a -1. In period, he Home enrepreneur L R defauls by an exogenous random amoun on he amoun LR ha she owes he L bank. T is a lump sum ax. d E is he enrepreneur s dividend income a. The E E enrepreneur consumes her dividend income. Her expeced lifeime uiliy a, V, is: E E E V u( d ) E V, wih 1 ( d 1) 1. Uiliy maximizaion by he enrepreneur E E E E 1 (subjec o (4)) yields hese firs-order condiions: (1 ) K N, (5) L E E E R 1E1u'( d 1)/ u'( d ) 1, (6) E ( u'( d )/ u'( d )){ K N q (1 )} / q 1, wih q '( I / I)/. E E E The Home governmen A dae, he Home governmen makes exogenous random oupu purchases financed using lump sum axes: G T W T E T B, where G ha are B T is a ax paid by he bank (see below). Each Home agen bears a consan share of he oal Home ax burden, equal o her share in Home seady sae consumpion: T G for i=w,e,b where i i i i is imeinvarian. In seing axes, he Home and Foreign governmens assume ha 50% of he banker s consumpion akes place in counry Home. The global bank The paper focuses on he role of bank capial for he ransmission of macroeconomic and financial shocks o global real aciviy. The paper herefore adops an aggregae 7

10 perspecive, and assumes a represenaive global bank ha may be hough of as he global financial sysem. 5 A, he global bank receives deposis D 1 and D from he 1 Home and Foreign workers, respecively, and makes loans L 1 and L 1 o Home and W Foreign enrepreneurs, respecively. Le D D D and L L L denoe W worldwide deposis and loans. The bank faces a capial requiremen: her dae capial L W D should no be smaller han a fracion of he bank s asses. 1 This may W W 1 1 reflec a legal requiremen (macro-prudenial policy) or, more broadly, marke pressures. To allow for ime-variaion in hese facors, I assume ha is a random variable (exogenous o he bank). 6 A sensiiviy analysis below considers a model varian wih a consan bank capial requiremen ( ). Bank capial requiremens are ofen jusified as limiing moral hazard in he presence of informaional fricions and deposi insurance (see Freixas and Roche (2008)). This issue is no modeled here. Insead, I ake he capial requiremen as given, and focus on is macroeconomic effecs. 7 I assume ha he bank can hold less capial han he required level, bu his is W W W cosly. Le x( L 1D1) L1 (1 ) W W L 1 D 1 denoe he bank s excess capial a W W he end of period. The bank bears a cos L ( x / L ) as a funcion of x, where L W L is he seady sae sock of loans. is a convex funcion ( '' 0) for which I assume: 5 Thus, he inerbank marke is no modeled here. Fricions in ha marke would maer for aggregae aciviy if hey affeced he oal flow of funds from savers o borrowers. The model here capures empirical flucuaions in he loan spread and in he oal volume of inermediaion. To invesigae he poenial role of an inerbank marke, I sudied a model varian wih a savings bank and an invesmen bank. The savings bank ges deposis from households, and lends o he invesmen bank (inerbank marke), which lends o firms. Each bank faces a capial requiremen and charges a loan spread. However, aggregae dynamics hinges on oal bank capial--hus ha se-up is observaionally equivalen o he represenaive-bank model. 6 The model could be used o evaluae he effecs of alernaive macro-prudenial governmen policies ha se as a funcion of he sae of he economy. However, policy comparisons, or he derivaion of opimal macro-prudenial policy rules, are beyond he scope of he paper. Mendicino and Punzi (2011) show ha he inroducion of macro-prudenial ools may have imporan implicaions for real aciviy and welfare. 7 See Meh and Moran (2010) for a closed economy DSGE model in which bank capial miigaes an agency problem beween banks and heir crediors. The auhors assume ha an exogenous fracion of bank earnings is kep for building bank capial, and ha bank capial is held in he form of physical capial; credi crunches originae in shocks o he depreciaion rae of bank-held physical capial. By conras, in he model here, banks make loans (bu do no hold physical capial), he fracion of reained bank earnings is endogenous (see below), and loan loss shocks are he main disurbance o bank capial. 8

11 ( x/ L W ) 0 for x 0; (0) 0. Thus, for x 0 he bank incurs a posiive cos; he cos is zero when he bank mees her capial requiremen. 8 W W A, he bank also bears an operaing cos ( D 1 L 1), where 0 is he (consan) real marginal cos of aking deposis and making loans. The bank s period budge consrain is: L D R ( D L ) L ( x/ L ) d T T L R D, (7) W W D W W W W B B B W L W where is he bank s oal loan loss, and T T is he oal ax paid by he bank B B (in he wo counries). B d is he dividend generaed by he bank a. (As he bank acs compeiively, loan raes and deposi raes are equaed across counries.) (7) implies ha bank capial a he end of period equals bank capial a he end of -1, plus reained bank earnings in. The banker consumes her dividend income, and selecs he pah of loans B and deposis o maximize her welfare. The banker s expeced life-ime uiliy a, V, is: B B B V u( d ) E V, wih 1 ( d 1) 1. B B B B 1 1 The banker s uiliy maximizaion problem has hese firs-order condiions: D B B B W R 1E1u'( d 1)/ u'( d ) 1 ' ( x/ L ), L B B B W R 1E1u'( d 1)/ u'( d ) 1 (1 ) '( x/ L ). A linear approximaion of hese Euler equaions (around x 0) gives: ' R L D ( / W ) 2 (0) (0) ( / W R xl xl ). (8) L D Hence, he loan rae spread R 1 R 1 is a funcion of he required capial raio and of he bank s excess capial, x. Noe ha if he bank raises deposis and loans by one uni, hen her operaing cos rises by 2 unis; excess bank capial falls by, which raises he ' '' Gerali e al. (2010) assume a quadraic cos funcion ( x) 2( x), 0, under which he bank also bears a cos when x 0. This funcion saisfies my assumpions. My seup is more general, as i allows for he possibiliy ha posiive excess capial generaes a convenience yield ( ( x ) 0 for x 0). Up o a linear approximaion (around x=0), boh specificaions yield idenical predicions; in paricular, he loan rae spread is decreasing in x if and only if ''(0) 0 (see below); he key assumpion is, hence, he convexiy of, which seems unconroversial. 9

12 W W W penaly L ( x / L ) by ( xl / ). The bank s Euler equaions imply ha he spread ' L D beween he loan rae and he deposi rae R 1 R 1 covers he marginal cos ' W 2 ( xl / ). Under sric convexiy of (i.e. 0), he marginal benefi of excess capial ' is a decreasing funcion of (excess) bank capial, which implies ha he loan rae spread is likewise a decreasing funcion of excess bank capial. The sensiiviy of he loan rae spread o changes in bank capial is governed by ''. Noe ha x/ W L cr, where cr ( W W 1 1)/ W L D L 1 is he bank s capial raio, i.e. he raio of bank equiy o bank asses. A one percenage poin rise in he capial raio '' lowers he loan rae spread by 4 percenage poins per annum (p.a.), while a one percenage poin increase in he required bank capial raio (holding consan he spread by 4[ '' '] percenage poins p.a.. '' cr ) raises Marke clearing Marke clearing for he oupu good requires: S S E E B W W W W ZZC C d d d I( I/ I) I ( I/ I ) GGL ( x/ L ) ( L 1D1). Forcing variables Seady sae TFP and invesmen efficiency are normalized o uniy ( 1). There are 11 forcing variables: Home and Foreign TFP (, ), invesmen efficiency (, ), governmen purchases ( G, G ) N N, labor supply shocks (, ), loan losses (, ) and he required bank capial raio ( ). I refer o he firs 8 shocks as nonbanking shocks, and o he las hree shocks as banking shocks. A large number of nonbanking shocks is assumed so ha he model has he poenial o capure imporan feaures of macro daa, even in he absence of banking shocks. Oher recen esimaed DSGE models likewise assume many shocks (e.g., Smes and Wouers (2007)). I consider wo specificaions of he law of moion of he forcing variables-- empirical resuls regarding he ransmission of banking shocks o real aciviy are broadly 10

13 similar across he wo specificaions. The firs specificaion follows he empirical DSGE lieraure and assumes ha all 11 forcing variables are univariae AR(1) processes: z for variable z, wih 0 1 ln( z/ z) ln( z / z), (9), where z z 1 z is a normally disribued whie noise. The AR(1) parameers are esimaed joinly wih he behavioral parameers. Independence of shocks makes i easy o decompose he variance of endogenous variables ino componens associaed wih each shock. However, ha specificaion does no capure he correlaions beween empirical measures of he forcing variables (see below). I hus also consider a second specificaion wih correlaed shocks. Specifically, he second specificaion assumes ha he 8 non-banking shocks follow univariae AR(1) processes wih correlaed innovaions, while he laws of moion of loan losses and of he required bank capial raio are of his form: / Y / Y ln( Y/ Y), 1 / Y / Y ln( Y / Y ), (10) 1 (1 ) ln( Y / Y ), (11) W W 1 wih 0,, 1. Y, Y and Y Y Y are Home and Foreign GDP and world W GDP respecively. 9, and are independen normal whie noises. The vecor of bk innovaions (,, ) is assumed independen of he vecor of innovaions o he nbk 8 non-banking forcing variables, denoed by, a all leads and lags. To allow for correlaion beween he banking shocks, and beween hose shocks and he non-banking forcing variables, equaions (10),(11) assume ha (,, ) depends on GDP and hus is parly endogenous. The independence of and bk makes i sraighforward o nbk decompose he variance of he endogenous variables ino componens due o nbk and 9 W W The bank s operaing coss and he coss of excess bank capial L ( x/ L ) represen inpus used by he bank; hese coss hus have o be subraced from he enrepreneurs oupu when compuing GDP. I assume ha he resources used by Home banking L 1D 1) are purchased from he Home enrepreneur, and ha ( a fracion / W W W LL of he resource cos L ( x/ L ) is likewise purchased from he Home enrepreneur. Hence, W Home GDP is: YZ( L 1D1) L ( x/ L ). 11

14 , respecively. (The covariance marix of non-banking shocks is calibraed, while he bk parameers of (10),(11) are esimaed; see below.) 2.2 Model soluion A linear approximaion (around he deerminisic seady sae) is used o solve he model. The soluion can be expressed as s s, (12) where s is a vecor consising of saes and conrols chosen (or realized) in period, nbk bk expressed as deviaions from seady sae values. (, ) is he vecor of dae innovaions o he forcing variables. 1 and 2 are marices whose elemens are funcions of he model parameers. 3. Economeric approach The model is esimaed using quarerly ime series for 12 macro and banking variables, in 1990q1-2010q3: US and EA GDP, oal privae consumpion, invesmen, employmen, commercial bank credi (deflaed using he GDP deflaor), he loan rae spread of US commercial banks, and he capial raio of US commercial banks. US (EA) daa are aken as empirical counerpars of Home (Foreign) variables in he model. The baseline esimaes use daa on oal bank credi (o all secors) by US Commercial banks and by EA Moneary financial insiuions (MFI). Below, I also repor esimaion resuls ha use daa on credi o he business secor. (I use oal credi for he baseline esimaes, as ha variable accouns for a greaer share of bank asses.) The baseline measure of he US loan rae spread is he series commercial and indusrial loan raes spread over inended federal funds rae, from he Federal Reserve Board s (FRB) Survey of Terms of Business Lending (Table E.2). Daa on he EA loan rae spread are only available for he period since 2003q1; as shown in Figure 2, he available EA loan spread closely racks he US loan spread (correlaion in : 0.90). (The EA spread ploed in Fig. 2 is he difference beween he EA MFI loan rae and he EONIA rae.) I hus use he US loan rae spread as a measure of he global loan rae spread. The US Commercial bank capial raio is aken as a proxy for he capial raio of he global bank. The empirical bank 12

15 capial raio measure is consruced as (oal financial asses oal liabiliies)/oal financial asses, using daa from he Flow of Funds (FRB). See he Appendix for furher informaion on he empirical variables. In esimaion, he loan spread and he capial raio are demeaned, while he oher empirical variables are linearly derended in log-form. The number of daa series used for esimaion (12) exceeds he number of shocks (11). To avoid sochasic singulariy of he model, I assume ha observed variables conain measuremen error. Allowing for measuremen error also seems imporan because (especially) he empirical banking series migh be imperfec measures of he heoreical conceps. 10 obs The period daa used in esimaion, y, are a subse of he saes and conrols included in he vecor s (see (12)), and are measured wih error: where is a marix, and obs y s, (13) is a vecor of Gaussian i.i.d. measuremen errors ha are independen of he rue sae variables a all leads and lags. I use a Bayesian approach o esimae a subse of he parameers, while he remaining parameers are calibraed Preference, echnology and banking parameers I esimae he (scaled) curvaure of he bank capial penaly funcion 4 '', he curvaure of he invesmen cos funcion '' and he risk aversion coefficien,. The firs wo parameers do no affec he seady sae, bu are key for he dynamic properies of he model. In paricular, 4 '' (sensiiviy of he loan rae spread o changes in he bank capial raio) is crucial for he ransmission of banking shocks o real aciviy. The means and sandard deviaions of he prior disribuions of hese parameers are shown in Cols. (1)-(2) of Table 2. I se he mean of he prior disribuion of 4 '' a 0.2, a value consisen wih ime series regressions of he loan rae spread on aggregae bank capial repored by Kollmann e al. (2011). (As discussed below, I se he seady sae required bank capial raio a 11.17%.) Invesmen is excessively volaile when 10 To break he singulariy, measuremen error in jus one observable is sufficien. To deermine he presence of measuremen error empirically, I allow for i in all series. Assuming measuremen error jus in banking variables gives he same resuls abou he role of banking shocks. For recen empirical DSGE models ha explicily allow for measuremen error see Ireland (2004), Boivin and Giannoni (2006), Gali e al. (2011) and de Anonio (2011). Sizable esimaed measuremen error may sugges model misspecificaion (Canova (2007)). 13

16 he capial accumulaion echnology is linear ( '' 0), as hen inernaional capial flows respond very srongly o counry-specific shocks. I se he mean of he prior disribuion of '' a 1; for ha value, he raio of he sandard deviaion of invesmen divided by he sandard deviaion of GDP is abou 3 in he model varians discussed below, and hus roughly in he range of he relaive volailiy of EA invesmen, when he oher parameers are se a prior mean values. The mean of he prior disribuion of is se a uniy. (The prior disribuions of,4 '' and '' are Gamma disribuions wih sandard deviaions se a half he prior means. Thus a reasonably wide range of parameer values around he mean has non-negligible mass.) Calibraed parameers I calibrae he remaining srucural parameers so ha he seady sae maches long run properies of he daa. I would be difficul o esimae he calibraed parameers hrough he lens of he model, using he (derended) empirical ime series used for esimaion (see Smes and Wouers (2007)). One period in he model represens one quarer in calendar ime. As is sandard in he macro lieraure, he (quarerly) depreciaion rae of physical capial is se a δ= The elasiciy of oupu wih respec o capial is se a α=0.3, consisen wih long run average hisorical US and EA labor shares of abou 70%. The wo-counry model here absracs from US and EA rade wih hird counries; I hus use he sum of US governmen consumpion and of US ne expors o counries oher han he EA as an empirical measure of US auonomous spending, G ; EA auonomous spending is consruced analogously. During 1990q1-2010q3, US [EA] auonomous spending represened 14.2% of US GDP [21.2% of EA GDP], on average. I ake he US as he empirical counerpar of counry Home and se GY / 14.2%, G / Y 21.2%. Mos DSGE sudies calibrae he subjecive discoun facor o mach average hisorical reurns. I use he same approach. As menioned above, i is assumed ha all agens have he same seady sae subjecive discoun facor, here denoed by. is se so ha he seady sae loan rae maches he mean 1990q1-2010q3 US real loan rae. I use he ineres rae on commercial and indusrial loans made by all commercial banks 14

17 repored by he FRB (Survey of Terms of Business Lending, Table E.2) as a measure of he nominal loan rae, from which I subrac he quarerly growh rae of he US GDP deflaor o consruc he real loan rae. The average US real loan rae 1990q1-2010q3 was 3.440% p.a.. Accordingly, I se he (quarerly) seady sae subjecive discoun facor a L (as R 1, from he enrepreneur s Euler equaion (6)). I assume ha all agens subjecive discoun facors have he same elasiciy wih respec o consumpion, denoed by. I se a a small absolue value, 0.001, ha yields a saionary equilibrium, while generaing (essenially) he same shor run dynamics as a model varian wih a consan subjecive discoun facor. (Impulse responses over he firs 100 periods are very similar across model varians wih 0 and ) The sample mean (1990q1-2010q3) of he US loan rae spread was 2.161% p.a.. 11 I se he seady sae deposi raes in he model a 1.279% p.a., so ha he seady sae loan rae spread maches he mean hisorical spread, 2.161%. The mean EA loan spread was 2.01% in (see above), which is close o he seady sae spread assumed in he model calibraion. I se he seady sae acual and required bank capial raios a cr 11.17%, which corresponds o he average capial raio of US commercial banks during he sample D period (from Flow of Funds daa). The bank s Euler equaions imply R 1 ' and L R 1 (1 ) '. Given he seady sae deposi and loan raes, hese wo condiions pin down he bank s marginal operaing cos and he seady sae slope of he bank s penaly funcion ': 0.25%, ' 0.28%. cr implies ha seady sae excess bank capial is zero, x=0, i.e. W W L (1 ) D. 12 I se L(1 ) D and L(1 ) D, i.e. he seady sae raio of deposis o loans is he same in boh counries (as is consisen wih he daa). The raio of ousanding US commercial bank loans o annual US GDP was 11 As menioned above, he baseline measure of he US loan rae spread is he commercial and indusrial loan raes spread over inended federal funds rae. Using he rae on shor erm Cerificaes of Deposi as a measure of he bank s marginal funding coss yields a loan rae spread ha has a 0.75 correlaion wih he baseline spread, and a sample mean of 1.929% p.a., which is close o he assumed seady sae spread. 12 Seing seady sae excess capial a a non-zero value generaes he same behavior, provided he calibraion maches he same seady sae deposi and loan raes (as in he baseline calibraion wih x=0). 15

18 53% on average in 1990q1-2010q3, while he mean raio of he sock of EA MFI loans divided by annual EA GDP was 87%. Thus, he US has a noiceably lower loans/gdp raio han he EA. The calibraion reflecs his: I assume ha he seady sae raios of loans o annual GDP are 53% in counry Home, and 87% in Foreign. Finally, I assume ha boh counries have he same seady sae GDP, normalized a uniy: YY 1. These seady sae arges pin down he remaining preference parameers (he weighs of deposis in Home and Foreign workers uiliy funcions, D D,, and seady sae marginal disuiliies of labor, N N, ) Processes of forcing variable, measuremen error The firs specificaion of he law of moion of he forcing variables assumes ha hose variables follow independen AR(1) processes; I esimae he auocorrelaions of he 11 forcing variables, and he sandard deviaions of he 11 shock innovaions, hrough he lens of he model, using he Bayesian approach. 14 The second specificaion assumes correlaed forcing variables. The parameers of he laws of moion (10)-(11) of loan losses and of he required bank capial raio are esimaed via he model (using he Bayesian approach). 15 By conras, I calibrae he process governing non-banking shocks, as empirical measures of he non-banking shocks can easily be consruced, and as esimaion of he correlaion marix of hese shocks hrough he lens of he model would be challenging (given he large number of crosscorrelaions). Specifically, I se he auocorrelaions of he 8 non-banking forcing variables, and he covariance marix of hese variables, equal o he sample auocorrelaions and sample covariance marix of empirical measures of he non-banking shocks (1990q1-2010q3), respecively (see Table 1). Following Coeurdacier, Kollmann 13 (3) implies D (1 D )(( S R C / Y)/( DY / )) S. DY / is deermined by LY /, while C / Y is pinned down by raios of governmen purchases and invesmen o GDP. Y1 hen pins down 16 N (as N deermines he seady sae labor inpu). In seady sae, consumpion by he Home [Foreign] worker and he enrepreneur represen respecively 58.2% and 4.8% [52.3% and 3.5%] of domesic GDP, while he banker s consumpion represens 0.21% of world GDP. 14 The prior disribuions of he auocorrelaions [sandard deviaions of shock innovaions] have mean 0.5 [0.5%] and a sandard deviaion equal o 0.1. Using more diffuse priors leaves he resuls unchanged. 15 The prior mean of he sandard deviaions of, and is se a 0.5%; he prior mean of he auoregressive coefficiens,, is 0.5; he prior mean of he coefficiens of GDP, and, is 0.

19 and Marin (2010), I use he raio of he CPI o he invesmen deflaor as a measure of invesmen efficiency. The empirical labor supply shock is consruced as N (1 )( Z / N )/ C, which follows from he firs order condiions (2),(5) when 1 (i.e. when equals is prior mean). Thus, he empirical labor supply shock is proporional o labor produciviy divided by consumpion. 16 As repored in Table 1, US invesmen efficiency, US auonomous spending and he US labor supply shock are more volaile han he corresponding EA variables. The cross-counry correlaions of TFP (0.51) and invesmen efficiency (0.84) are sizable. TFP is posiively correlaed wih invesmen efficiency; US and EA TFP are srongly negaively correlaed wih US auonomous spending (G), and negaively correlaed wih he US labor supply shock. All forcing variables are highly persisen (auocorrelaions in he range ). The prior means and prior sandard deviaions of he sandard deviaions of measuremen errors are se a 1/4 and 1/20, respecively, of he sandard deviaions of he corresponding (demeaned/derended) empirical series. 4. Daa plos and business cycle saisics. Figure 1 plos he (demeaned/derended) 12 empirical quarerly ime series (1990q1-2010q3) used in esimaion. Macro aggregaes co-move closely across he US and he EA he synchroniciy was especially high during he Grea Recession of 2007q4-2009q2 (as daed by he NBER). (Shaded areas in Figures indicae NBER recessions.) Relaive o rend, US oupu fell by 8.5%, during he recession, while EA oupu fell by 7.5%; US consumpion (-7.3%) and invesmen (-35.1%) fell more sharply han EA consumpion (-4.0%) and invesmen (-15.9%). US and EA bank lending grew srongly in he years before 2008, and hen decreased sharply. The loan rae spread fell during he hree years prior o he crisis, bu rose sharply during he Grea Recession. The bank capial raio exhibis relaively mild flucuaions--hroughou he sample period i says in a ±2% range around he sample mean of 11.17%. 16 Labor produciviy is consruced using GDP as a proxy for he enrepreneur s oupu Z. I also considered N an alernaive measure of he labor supply shock based on real wage rae daa: / C. Tha measure gives similar esimaes of model parameers and of he role of banking shocks. 17

20 Figure 3 plo he bank capial raio, ogeher wih he baseline loan spread series and wo oher spread measures ha are used for robusness checks below (all series in Figure 3 are demeaned). Excep for he period of he financial crisis, he bank capial raio and he baseline loan rae spread comove negaively. While he baseline loan rae spread rose, during he crisis (as menioned above), he bank capial raio has had a fla rend since abou i has been argued ha his may parly reflec accouning discreion, which has allowed banks o oversae he value of heir asses in he crisis (Huizinga and Laeven (2009)). The correlaion beween he bank capial raio and he baseline lending spread was during he period , and over he whole sample period. Figure 3 also plos he US series ne percenage of banks increasing spreads of loan raes over cos of funds, from he FRB Senior Loan Officer Opinion Survey on Bank Lending Pracices, SLOOS. (The series represens he percenage of banks increasing spreads minus he percenage of banks lowering spreads; he ploed series is scaled so ha is sandard deviaion equals ha of he baseline loan spread.) Tha series is posiively correlaed wih he baseline loan spread (correlaion 0.39 for ), and negaively correlaed wih he bank capial raio (-0.47 for ; for ). Also ploed in Figure 3 (see righ panel) is Gilchris and Zakrajšek s (2011a) excess US commercial bond premium, consruced by subracing expeced bond defaul probabiliies from he spread beween he yield on US commercial bonds and he yield on US Treasury bonds. As commercial banks are key players in he commercial bond marke, he commercial bond premium migh be informaive abou credi spreads/marke condiions. (Gilchris and Zakrajšek (2011a) argue ha an increase in he excess bond premium reflecs [...] a conracion of he supply of credi wih significan adverse consequences for he macroeconomy, p.31.) The excess bond premium oo is negaively correlaed wih he bank capial raio (correlaion: for ; for ). The bond premium is posiively correlaed wih he baseline loan rae spread (0.29) and wih he SLOOS ne percenage of banks increasing spreads (0.79). Overall, he daa are hus consisen wih he model s key predicion ha he spread is inversely relaed o he bank capial raio (see (8)). The absence of a pronounced inverse relaion during he crisis migh be due o he fac ha he measured bank capial raio oversaes he rue capial raio during he crisis (see discussion above), or ha he 18

21 required bank capial raio rose during he crisis (his could raionalize he observed increase in he loan rae spread, during he crisis, wihou a fall in he bank capial raio). The las Column of Table 3 repors momens of Hodrick-Presco (HP) filered quarerly macro and banking variables, for he US and he EA (1990q1-2010q3). (The smoohing parameer is se a 1600.) The sandard deviaion of GDP is very similar in he US (1.12%) and he EA (1.14%). Consumpion is less volaile han GDP, while invesmen is markedly more volaile han GDP. US invesmen is almos wice as volaile as EA invesmen. In boh counries, loans are more volaile han oupu, and he loan spread is counercyclical. Real aciviy and loans are posiively correlaed across he US and EA. 5. Esimaion resuls 5.1. Poserior parameer esimaes (Table 2) Columns (4)-(5) of Table 2 repor he mean and he sandard deviaion of he poserior parameer disribuion, for he model varian wih independen shocks. 17 Cols. (8)-(9) repor poserior parameer esimaes for he model varian wih correlaed shocks. The daa are informaive abou he esimaed parameers: in almos all cases, he poserior parameer disribuions have lower sandard deviaions han he prior disribuions; he poserior means ofen differ noiceably from he prior means (poserior means and modes are very close). Wih independen shocks, he poserior mean of 4 '' indicaes ha a 1 percenage poin increase in he bank capial raio leads o a 21 basis poin reducion in he annualized loan rae spread, and ha a 1 percenage poin rise in he required bank capial raio ( ) increases he loan rae spread by 19 basis poins p.a.. 18 In he model varian wih correlaed shocks, he poserior esimaes sugges ha he spread is abou wice as sensiive o changes in he bank capial raio: 4 '' The means and sandard deviaions of he poserior disribuions were generaed using he Random Walk Meropolis algorihm (see An and Schorfheide (2007)). 18 The poserior means of he spread sensiiviy 4 '' and risk aversion are close o prior means. I experimened wih larger prior means of hese parameer, and found ha he poserior means remain close o he poserior means repored in Table 2, which indicaes ha he daa are informaive abou hese parameers. 19

22 The poserior esimaes of he model wih independen shocks indicae ha EA loan loss shocks are roughly as volaile as US loan loss shocks he poserior modes of he sandard deviaions of and are 0.71% and 0.79%, respecively, in he model varian wih independen shocks, while he model wih correlaed shocks implies ha EA loan losses are more volaile han US loan losses. The required bank capial raio undergoes sizable flucuaions (poserior mode of sd. of : 0.5%-0.6%). The poserior means of he sandard deviaions of measuremen errors are mosly smaller han he prior means (an excepion is he measuremen error for he bank capial raio) Business cycle momens implied by poserior parameer esimaes (Table 3) Cols. (1)-(9) of Table 3 repor model-prediced momens of HP filered US and EA variables (compued a he poserior mode of he esimaed parameers), for he model varian wih independen shocks. Column (1) [labeled All ] assumes all 11 srucural shocks, and measuremen error. Cols. (2)-(9) consider momens generaed by differen subses of he srucural shocks, in isolaion, wihou measuremen error. Specifically, nbk Col. (2) [ NonBk ] assumes jus he 8 non-banking shocks, and Col. (3) [ Bnk ] bk assumes jus he 3 exogenous banking shocks. Cols. (4)-(9) assume jus a single ype of shock (Col.(4): jus TFP shocks; Col. (5): jus invesmen efficiency shocks; ec.). 19 The independen-shocks model wih all shocks and measuremen error generaes prediced sandard deviaions ha are mosly in he range of he empirical saisics. The prediced sandard deviaions of US GDP (1.14%) and of EA GDP (1.22%) are close o he empirical sandard deviaions (1.12%, 1.14%); see Columns (1) and (15). The model (wih all shocks) capures he fac ha invesmen is more volaile han GDP. The model also capures he high volailiy of US loans, bu i underpredics he volailiy of EA loans. I maches he procyclical behavior of he macro aggregaes, employmen and loans, and correcly predics ha he loan spread is counercyclical. However, he model 19 obs Using (12),(13), he model soluion for observables (wih measuremen error) can be wrien as: y nbk bk AL () BL () where AL () and BL () are lag polynomials. (The momens in Table 3 perain o HP obs, HP obs nbk bk filered series, y H() L y, where HL () is he HP filer.) By assumpion,, and are independen a all leads and lags. Thus he prediced variance of endogenous variables under all shocks and measuremen error is he sum of: (i) he variance wih jus non-banking shocks; (ii) he variance wih jus banking shocks; (ii) he variance of measuremen error. 20

23 wih all (independen) shocks predics cross-counry correlaions of GDP (-0.26), invesmen (-0.02) and employmen (-0.24) ha are negaive, and hus markedly below he empirical (posiive) correlaions. Bu noe ha he prediced cross-counry consumpion correlaion (0.24) is posiive, and hus much closer o he empirical correlaion (0.39). 20 Taken in isolaion, TFP shocks and labor supply shocks induce by far he larges flucuaions in real aciviy (prediced sandard deviaions of US GDP wih jus hese shocks: 0.78% and 0.62%, respecively). The prediced sandard deviaions of US GDP wih jus loan loss shocks (0.18%) and wih jus shocks o he required bank capial raio (0.08%) are noiceably lower. Wih jus (uncorrelaed) TFP shocks, and jus labor supply shocks, GDP is negaively correlaed across counries. This is due o he fac ha hese shocks are negaively ransmied inernaionally e.g. a posiive shock o Home TFP raises Home GDP, bu lowers Foreign GDP (see below). By conras, governmen purchases shocks and he banking shocks induce srong posiive cross-counry oupu correlaions. Noice ha he banking shocks induce a srong negaive correlaion beween he loan rae spread and GDP. Prediced momens generaed by he model varian wih correlaed shocks are repored in Cols. (12)-(14) of Table 3. The prediced sandard deviaions and crosscorrelaions wih domesic GDP are mosly in he same range as he corresponding momens generaed by he srucure wih independen shocks. 21 However, he varian wih correlaed shocks generaes sizable posiive cross-counry correlaions of oupu (0.45) and invesmen (0.31) ha are close o he empirical correlaions; he prediced cross-counry consumpion correlaion (0.75) exceeds he empirical correlaion. Sandard 20 In open-economy DSGE models, he prediced cross-counry consumpion correlaion exceeds he crosscounry oupu correlaion, provided inernaionally raded asses allow he residens of differen counries o (parially) share heir consumpion risk (Kollmann (1996, 2012)). Consumpion would be perfecly correlaed across he wo counries, if complee financial markes exised. 21 In independen-shocks model, loan losses and he required capial raio are negaively correlaed wih GDP (as a rise in losses and in he required capial raio lowers oupu; see below). In he correlaed-shocks model varian, he poserior esimaes of he response coefficiens of loan losses and he required bank capial raio o GDP (, ) are boh posiive bu close o zero (0.07); see (10)-(11). This implies ha he required bank capial raio is posiively correlaed wih loan losses. The implied correlaions of US GDP wih he US loan loss and he world loan loss are 0.18 and -0.04, respecively; EA GDP is prediced o be negaively correlaed wih he EA loan loss (-0.17) and he world loan loss (-0.16). Imposing a zero response coefficiens of he US loan loss o GDP yields a negaive correlaion beween he US loan loss and US GDP, bu i barely affecs esimaes of remaining parameers and oher prediced momens. 21

24 Inernaional RBC models wihou banks (e.g., Backus, e al. (1992), Baxer and Crucini (1995), Kollmann (1996)) oo generae cross-counry correlaions of real aciviy ha are lower han he empirical cross-counry correlaions, unless shocks are highly correlaed across counries, and he same also holds for varians of he presen model wihou an operaive bank capial requiremen ( '' 0); see below Variance shares accouned for by banking shocks (Table 4) Table 4 repors he % shares of he prediced variances of HP filered endogenous variables (wih measuremen error) ha are accouned for by he non-banking shocks nbk (see rows labeled NonBk ), and by he banking shocks bk (rows labeled Bank ); he remainder represens he conribuion of measuremen error o he prediced variance. (The variance shares are compued a he poserior mode of he esimaed parameers.) According o he model varian wih independen shocks, he banking shocks accoun for a 3.1% share of US GDP variance, bu explain larger shares of he variances of US invesmen (6.1% share), employmen (6.3%) and loans (41.0%); see Panel (a1) of Table 4. Banking shocks accoun for greaer variance shares of EA variables--gdp: 4.0%; invesmen: 22.6%; employmen: 7.8%; loans: 72.0%. Thus, roughly one-fifh of he variance of EA invesmen is due o banking shocks, according o he model wih independen shocks. Banking shocks accoun for 59.7% of he variance of he bank capial raio, and for 84.7% of he variance of he loan rae spread. 22 The esimaion resuls also show ha loan loss shocks are more imporan drivers of real aciviy han shocks o he required bank capial raio; he laer explain merely 0.5% of he variances of US and EA GDP, in he model varian wih independen shocks US EA (see rows labeled ' ',' ' and ' ' in Panel (a1) of Table 4). In he model varian wih correlaed shocks, he variance shares of real aciviy induced by he hree join banking shocks are imes greaer han in he esimaed model wih independen shocks; see Panel (b) of Table 4. Banking shocks now accoun for 5.5% (14.2%) of he variance of US (EA) GDP, and for 10.6% (53%) of he variance 22 Non-banking shocks explain only a negligible share of he variance of he bank capial raio and he loan rae spread. A sizable share of he variance of he bank capial raio (40.1%) is hus accouned for by measuremen error. 22

25 of US (EA) invesmen. Inuiively, when he processes of he non-banking shocks are calibraed, hen he esimaed banking shock processes bear a greaer burden of fiing he daa (and hus accoun for greaer variance shares). These findings sugges ha banking shocks accoun for a non-negligible share of he variance of real aciviy. Specifically, banking shocks accoun for 3%-5% of he uncondiional variance of US GDP, and for 4%-14% of he variance of EA GDP. The variance shares of invesmen accouned for by banking shocks are higher, especially for he EA (above 20%). 23 EA real aciviy hus depends more on banking shocks han US real aciviy. Ineresingly, US loan losses accoun for a greaer share of he variance of EA real aciviy han of he variance of US real aciviy. This finding is in line wih Helbling e al. (2011) who argue, based on vecor auoregressions, ha US credi supply shocks accoun for a greaer share of flucuaions in global GDP han of US GDP. A robusness analysis below confirms he findings discussed in his Secion Impulse responses (Table 5) Impulse responses (repored in Table 5) help o undersand he model s mechanics, and he prediced business cycle momens. The impulse responses are compued a he poserior mode of esimaed model parameers, for he baseline model wih independen shocks. Each impulse response focuses on an isolaed innovaion, assuming ha all oher exogenous innovaions are zero. (To save space, Table 5 does no show responses o EA non-banking shocks hose responses are qualiaively similar o he responses o US non-banking shocks.) A posiive innovaion o Home TFP raises Home GDP and invesmen, bu leads o a fall in Foreign GDP. The shock raises he income of he Home worker; hus ha worker saves more, and her holdings of bank deposis increase--i.e. he bank s deb rises, which lowers he bank capial raio. The deposi rae falls (due o he greaer supply of 23 The esimaes here sugges a more imporan role for banking shocks han illusraive calibraions in Kollmann e al. (2011), according o which banking shocks accoun for less han 0.2% of he variance of real aciviy. This greaer role is due o he fac ha he esimaed sensiiviy of he loan spread o bank capial here, and he esimaed variance of loan losses, are larger han corresponding parameers assumed in hose calibraions. 24 Nolan and Thoenissen (2009) and Jermann and Quadrini (2012) use closed economy models wih collaeral-consrained firms (bu wihou banks) o consruc esimaes of shocks o firms funding consrains. The auhors argue ha hose shocks can explain up o half of he variance of US GDP. By conras, he model here assumes ha only he bank faces a capial requiremen. 23

26 deposis), and so does he loan rae however, he loan rae spread rises. The Foreign worker responds o he fall in he deposi rae by consuming more, and working less, and hence Foreign GDP falls. (Foreign invesmen rises slighly, due o he fall in he loan rae.) Counry-specific labor supply shocks likewise drive Home and Foreign GDP in opposie direcions. By conras, banking shocks induce responses of real aciviy (and of loans) ha are common across he wo counries. For example, a loan loss in one counry lowers he global bank s capial raio, which riggers a rise in he loan rae spread he deposi rae falls, while he loan rae rises. In response o his, loans, invesmen and GDP fall in boh counries. A rise in he required capial raio ( ) likewise raises he loan rae spread; on impac, his oo lowers loans, invesmen and real aciviy in boh counries. Noe also ha banking shocks drive he loan spread and oupu in opposie direcions. According o he model varian wih independen shocks, an unanicipaed US loan loss worh 1% of seady sae quarerly US GDP reduces he bank capial raio by 14.9 basis poins, on impac, and i lowers US and EA quarerly GDP by, respecively, 0.10% and 0.12% (on impac). An unanicipaed EA loan loss of he same size lowers US and EA GDP by 0.14% and 0.18%, respecively. Thus, EA GDP is more sensiive o domesic and foreign loss shock han US GDP. A US loan loss lowers EA GDP more han US GDP. An unanicipaed increase in he required bank raio by one percenage poin lowers US and EA GDP by 0.10% and 0.11%, respecively, on impac Decomposing hisorical ime series (Figures 6 and 7) Figure 6 plos he esimaed conribuions of he banking shocks and of US and EA nonbanking shocks o he hisorical ime series, implied by he model varian wih independen shocks. 25 Thick coninuous lines show he hisorical daa; he hin coninuous lines indicae he conribuion of banking shocks, while he dashed-doed and dashed lines represen he conribuions of US and EA non-banking shocks, respecively. The hisorical decomposiion yields a picure ha is consisen wih he variance 25 The decomposiion is compued a he poserior mode of he esimaed parameers. Using smoohed shocks and measuremen errors, each hisorical series can be expressed as he sum of: (i) a base rajecory (dynamic effecs of predeermined saes in he iniial period) plus measuremen error; (ii) conribuions of each exogenous shock. Figures 6 and 7 shows daa and shock conribuions. 24

27 decomposiions. Banking shocks maer more for EA GDP han for US GDP. During he Grea Recession of 2007q4-2009q2, banking shocks accoun for a 1.0 percenage poin [1.2 pp.] fall in US [EA] GDP i.e. he banking shocks capure 12% [16%] of he 8.5 pp. [7.5 pp.] fall in US [EA] GDP, relaive o rend. Banking shocks also capure 15% [35%] of he fall in US [EA] invesmen, and 19% [56%] of he fall in US [EA] employmen, during he recession. Thus, more han 1/3 of he fall in EA invesmen and employmen is accouned for by he banking shocks. In he previous US recession (2001q1-2001q4), banking shocks accouned for 11% of he fall in US oupu and invesmen, and for 21% [29%] of he fall in EA oupu [invesmen]. During he 1990q3-1991q1 US recession, banking shocks accouned for 6%, 10% and 16%, respecively, of he fall in US GDP, invesmen and employmen (he EA did no experience a recession in ). Figure 6 shows ha he oupu componens accouned for by he domesic nonbanking shocks rack hisorical US and EA GDP very closely. This resul parallels he finding by de Walque e al. (2005) and Le e al. (2010) ha domesic macro shocks are he main drivers of US and EA GDP. Foreign non-banking shocks had a sabilizing effec on domesic real aciviy; eg, during he recession, EA non-banking shocks had a posiive influence on US GDP, and hus miigaed he US recession. This reflecs he fac ha, in he model here, TFP shocks and labor supply shocks are negaively ransmied inernaionally (see above). Figure 7 repors hisorical decomposiions implied by he model varian wih correlaed shocks. Tha varian aribues larger shares of he drop in real aciviy during he 2007q4-2009q2 recession o banking shocks, namely 17% [21%] of he fall in US [EA] GDP, 24% [54%] of he fall in US [EA] invesmen, and 30% [79%] of he reducion in US [EA] employmen The role of he bank capial requiremen and of banking shocks The presence of an operaive bank capial requiremen '' 0 is key for he ransmission of banking shocks o domesic and foreign real aciviy. Banking shocks have a negligible effec on real aciviy, bu remain imporan drivers of loans and he bank capial raio, when '' 0. (An esimaed model varian wih '' 0 predics ha banking shocks 25

28 explain merely 0.002% of he variance of HP filered GDP and invesmen, bu beween 1/3 and 2/3 of he variances of loans and he bank capial raio.) Columns (6)-(7) of Table 2 repors poserior parameer esimaes for an independen-shocks model varian wihou an operaive bank capial requiremen ( '' 0), and wihou banking shocks. 26 Tha model varian resembles sandard Inernaional RBC models wih incomplee financial markes in which jus an uncondiional bond can be raded inernaionally (see, e.g., Baxer and Crucini (1995), Kollmann (1996)). Columns (10)-(11) of Table 3 repor he implied business cycle momens. Table 3 shows ha he baseline independen-shocks model (wih operaive bank capial requiremen and banking shocks) generaes business cycles momens ha are, mosly, closer o he empirical momens han he varian wihou he operaive bank capial requiremen (and wihou banking shocks); see Columns (1) and (10) of Table 3. For example, prediced sandard deviaions of US and EA GDP in he baseline model (1.14% and 1.22%, respecively ) are larger han in he srucure wihou he operaive bank capial requiremen (1.01%, 0.95%), and closer o he empirical sandard deviaions (1.12%, 1.14%). Noe ha he independen-shocks model wih '' 0 oo generaes prediced cross-counry correlaions of GDP (-0.16), invesmen (-0.10) and employmen (-0.10) ha are markedly below he empirical cross-counry correlaions (like he baseline srucure). 27 Model fi can be evaluaed using he marginal likelihood. 28 The log marginal likelihood (LML) of he baseline model wih independen shocks is , while he LML of he model varian wihou he operaive bank capial consrain and wihou 26 The priors for parameer (oher han '') are he same as in he baseline model. The poserior esimaes of mos parameers are similar o he esimaes in he baseline model. 27 Surprisingly, he prediced cross-counry correlaions of GDP and employmen wih '' 0 (wihou banking shocks) are slighly higher han in he baseline model wih '' 0 (wih banking shocks). This is i.a. due o he fac ha esimaed sandard deviaions of TFP shocks (which are negaively ransmied across counries) are lower when '' 0. Holding consan he disribuion of non-banking shocks (and parameers oher han '') a he esimaes for he baseline model, bu seing '' 0 and eliminaing he banking shocks lowers he prediced cross-counry correlaions of GDP and employmen o and -0.32, respecively (compared o and -0.24, respecively in he baseline model). 28 The marginal likelihood (ML) measures he ou-of-sample predicive abiliy of he model (Geweke (2001)). The MLs repored here were compued using a Laplace approximaion (Canova (2007)). The Geweke (1999) harmonic mean esimaor, based on parameer draws from he Meropolis algorihm, yields very similar MLs. 26

29 banking shocks is This implies a Bayes facor (raio of poserior odds o prior odds) of e ha massively favors he baseline model. The model varian wih an operaive bank capial requiremen, bu wihou banking shocks has a LML of ; a model varian wihou an operaive bank capial requiremen bu wih banking shocks has a LML of This suggess ha boh he operaive bank capial requiremen and he banking shocks help he model capure he join dynamics of he macro and banking variables used in esimaion. The presence of hese model ingrediens also helps o beer explain he 8 US and EA macro variables used in esimaion. For hese 8 macro variables, he baseline model has a LML of , while he model varian wihou an operaive bank capial requiremen and wihou banking shocks has a LML of Similar findings obain for he srucure wih correlaed shocks. Wih correlaed shocks, he model varian wih an operaive bank capial requiremen and banking shocks has a LML of , while a varian wihou bank capial requiremen and wihou banking shocks has a LML of ; in erms of jus he 8 US and EA macro variables, he corresponding LMLs are ( '' 0, wih banking shocks) and ( '' 0, no banking shocks), respecively, i.e. he model wih he operaive bank capial requiremen and banking shocks is clearly favored Furher robusness checks Panel (a2) of Table 4 repors variance decomposiions for an independen-shocks model varian wih an operaive bank capial requiremen, in which he required bank capial raio is consan,, so ha loan losses are he only banking shocks. In ha varian, banking shocks explain somewha greaer shares of he variance of real aciviy han in he baseline model; however, he varian has a markedly lower LML ( ) han he baseline model. As a furher robusness check, I re-esimaed he model varian wih independen shocks using oher empirical measures of he loan rae spread and of bank loans. Panel (a3) of Table 4 repors he resuling esimaes of variance shares explained by banking shocks. These variance shares are broadly in he same range as he baseline shares discussed above (Panel (a1)). (Poserior parameer esimaes obained from he alernaive daa ses are in he same range as he baseline esimaes, and are hus no repored) 27

30 In Panel (a3.1) of Table 4, he baseline loan rae spread is replaced by he series ne percenage of banks increasing spreads of loan raes over cos of funds (from SLOOS), while Panel (a3.2) uses he Gilchris-Zakrajšek (2011a) excess bond premium series in lieu of he baseline loan rae spread. The imporance of banking shocks in explaining real aciviy flucuaions rises somewha when hose alernaive spread measures are used o esimae he model. (The same resul holds when he SLOOS series ne percenage of banks ighening lending sandards is used insead of he baseline lending spread.) For example, when he Gilchris-Zakrajšek excess bond premium is used, abou 5% of he variance of US and EA GDP is aribued o banking shocks. 29 In Panel (a3.3) of Table 4, oal bank credi is replaced by US and EA bank loans o he non-financial business secor, while Panel (a3.4) uses Gilchris and Zakrajšek s (2011b) measure of US business lending capaciy in lieu of US oal credi. 30 Figures 4 and 5 plo hese series. Business loans are highly posiively correlaed wih oal loans, bu more volaile, especially in he US. US lending capaciy fell earlier and much more sharply han oal lending, during he recession. The business lending measures yields smaller variance shares due o banking shocks han he baseline model varian, while he lending capaciy measure yields roughly similar variance shares as he baseline varian Conclusion This paper has esimaed a wo-counry model wih a global banking sysem, using US and Euro Area (EA) daa (1990q1-2010q3), and Bayesian mehods. The esimaed model maches key US and EA business cycle saisics. Empirically, a model version wih an 29 As poined ou by a referee, he lending rae spread may be affeced by facors ha are no capured by he model, such as credi risk, liquidiy ensions and banking compeiion (incorporaing hese facors ino he heory would be a useful ask for fuure research). The fac ha he SLOOS index and he Gilchris- Zakrajšek bond premium explain a slighly greaer share of he variance of real aciviy migh indicae ha hese spread measures capure beer hose oher deerminans of credi supply. 30 In he US, many business loans are offered under prior commimen (credi lines); hence, business loans respond wih a lag o shocks o bank funding. The Gilchris and Zakrajšek business lending capaciy measure is defined as he sum of loans ousanding and of unused commercial bank lending commimens-- he auhors argue ha his variable is more informaive (han loans ousanding) for idenifying loan supply shifs (no comparable measure exiss for he EA). 31 I also esimaed he model using an alernaive measure of he bank capial raio--he capial raio of US Securiies Brokers and Dealers (insead of he capial raio of US commercial banks). Resuls are robus o using his measure. Kollmann and Zeugner (2012) analyze he capial raio dynamics of differen subsecors of he finance indusry. 28

31 operaive bank capial consrain ouperforms a srucure wihou such a consrain. Banking shocks accoun for 3%-5% of he uncondiional variance of US GDP, and for 4%-14% of he variance of EA GDP. The variance shares of invesmen accouned for by banking shocks are higher, especially for he EA (20% or more). Thus, EA real aciviy depends more on banking shocks han US real aciviy. US loan losses accoun for a greaer share of he variance of EA real aciviy han of he variance of US real aciviy. During he Grea Recession ( ), banking shocks accouned for abou 12%-20% of he fall in US and EA GDP, and for more han a hird of he fall in EA invesmen and employmen. 29

32 DATA APPENDIX A.1 Baseline daa se used for esimaion US GDP, privae consumpion (oal), invesmen (all a consan prices): from US Naional Income and Produc Accouns (Bureau of Economic Analysis, BEA); he invesmen series include privae and governmen invesmen. US employmen: Toal nonfarm payrolls: all employees (Bureau of Labor Saisics) US bank loans: ousanding oal bank credi by Commercial Bank, deflaed using GDP deflaor (from June 2011 Flow of Funds, Table L109). US bank capial raio: (oal financial asses-oal liabiliies)/(oal financial asses) for Commercial Banks (from June 2011 Flow of Funds, Table L109). US loan rae spread: Commercial and indusrial loan raes spread over inended federal funds rae ( All loans series, Survey of Terms of Business Lending, Table E.2, Federal Reserve Board, June 2011). EA GDP, privae consumpion (oal), invesmen (all a consan prices): from ECB Area-Wide Model (AWM) daabase (10 h updae, Sepember 2010). EA employmen: from AWM daabase. EA bank loans: MFI loans o privae secor (from ECB monhly bullein), deflaed using he GDP deflaor. A.2 Oher variables (used for esimaion of model varians) Excess bond premium: spread beween he yield on US commercial bonds and he yield on Treasury bonds, minus expeced bond defaul probabiliies, as consruced by Gilchris and Zakrajšek (2011a) using daa for a panel of individual bonds. Ne percenage of banks increasing spreads of loan raes over cos of funds : percenage of banks increasing spreads minus he percenage of banks lowering spreads, from he Senior Loan Officer Opinion Survey on Bank Lending Pracices, SLOOS (Federal Reserve Board). SLOOS repors a series (ne percenages of banks raising spreads) for loans o large and middle-marke firms and one for loans o small firms. The wo series are very similar (correlaion: 0.95). I use he average of he wo series. US business loans: ousanding commercial bank loans o he non-financial business secor, consruced by Gilchris and Zakrajšek (2011b). EA business loans: MFI loans o non-financial corporaions(nfc), from ECB monhly bullein, deflaed using he GDP deflaor. US business lending capaciy: ousanding commercial bank loans plus unused commercial bank lending commimens (credi lines) o he non-financial business secor, consruced by Gilchris and Zakrajšek (2011b). A.3 Oher variables (used for model calibraion) Auonomous spending (G): governmen purchases plus ne expors o hird counries (deflaed using GDP deflaor). Daa sources: AWM, BEA and ECB monhly bullein. Invesmen efficiency: measured as raio of CPI o Gross Invesmen Deflaor (BEA and AWM). All series are quarerly and seasonally adjused (when relevan) 30

33 REFERENCES Adolfson, M., S. Laséen, J. Lindé and M. Villani, Bayesian Esimaion of an Open Economy DSGE Model wih Incomplee Pass-Through. Journal of Inernaional Economics 72, Aikman, D. and M. Pausian, Bank Capial, Asse Prices and Moneary Policy. Working Paper No. 305, Bank of England. An, S. and F. Schorfheide, Bayesian Analysis for DSGE Models. Economeric Reviews 26, Andreasen, M., J. Sondergaard and M. Pausian, Porfolio Linkages, Financial Shocks and Inernaional Business Cycles. Working Paper, Bank of England. Backus. D., P. Kehoe, and F. Kydland, 1992, Inernaional Real Business Cycles. Journal of Poliical Economy 100, Baxer, M., and M. Crucini, Business Cycles and he Asse Srucure of Foreign Trade. Inernaional Economic Review 36, Boivin, J., and M. Giannoni, DSGE Models in a Daa Rich Environmen. Working Paper No , Naional Bureau of Economic Research. Brunnermeier, M. and Y. Sannikov, A Macroeconomic Model wih a Financial Secor. Working Paper, Princeon Universiy. Canova, F., Mehods for Applied Macroeconomic Research. Princeon Universiy Press, Princeon. Chrisiano, L., M. Eichenbaum and C. Evans, Nominal Rigidiies and he Dynamic Effecs of a Shock o Moneary Policy. Journal of Poliical Economy 113, Coeurdacier, N., R. Kollmann and P. Marin, Inernaional Porfolios, Capial Accumulaion and he Dynamics of Capial Flows. Journal of Inernaional Economics 80, Correa, R., H. Sapriza and A. Zlae, Inernaional Banks, he Inerbank Marke, and he Cross-Border Transmission of Business Cycles. Working Paper, Inernaional Finance Secion, Federal Reserve Board. Davis, S., The Adverse Feedback Loop and he Effecs of Risk in boh he Real and Financial Secors. Working Paper No. 66, Globalizaion and Moneary Policy Insiue. de Anonio Liedo, D., Wha are Shocks Capuring in DSGE Modeling? Noise Versus Srucure. Working Paper, Naional Bank of Belgium. Dedola, L. and G. Lombardo, Financial Fricions, Financial Inegraion and he Inernaional Propagaion of Shocks., Economic Policy, Del Negro, M., G. Eggersson, A. Ferrero, A. and N. Kiyoaki, The Grea Escape? A Quaniaive Evaluaion of he Fed s Liquidiy Faciliies. Saff Repor no. 520, Federal Reserve Bank of New York. Devereux, M. and A. Suherland, Evaluaing Inernaional Financial Inegraion Under Leverage Consrains. European Economic Review 55, Dewacher, H. and R. Wouers, Endogenous Risk in a DSGE Model wih Capial- Consrained Financial Inermediaries. Working Paper 235, Naional Bank of Belgium. de Walque, G, O. Pierrard and A. Rouabah, Financial (In)Sabiliy, Supervision and Liquidiy Injecion: A Dynamic General Equilibrium Approach. Economic Journal 120, de Walque, G., F. Smes and R. Wouers, An Esimaed Two-Counry DSGE Model for he Euro Area and he US Economy. Working Paper, Naional Bank of Belgium. 31

34 Fisher, J., The Dynamic Effecs of Neural and Invesmen-Specific Technology Shocks. Journal of Poliical Economy 114, Freixas, X. and J.-C. Roche, The Microeconomics of Banking. MIT Press, Cambridge. Galí, J., F. Smes and R. Wouers, Unemploymen in an Esimaed New Keynesian Model. Working Paper 17084, Naional Bureau of Economic Research. Gerali, Andrea, Sefano Neri, Luca Sessa and Federico Signorei Credi and Banking in a DSGE Model of he Euro Area. Journal of Money, Credi and Banking 42, Gerler, M. and P. Karradi, A Model of Unconvenional Moneary Policy. Journal of Moneary Economics 58, Gerler, M. and N. Kiyoaki, Financial Inermediaion and Credi Policy in Business Cycle Analysis. In: Handbook of Moneary Economics (B. Friedman and M. Woodford, eds.), Vol. 3A, , Elsevier: Amserdam. Geweke, J., Using Simulaion Mehods for Bayesian Economeric Models: Inference, Developmen and Communicaion. Economeric Reviews 18, , Bayesian Economerics and Forecasing. Journal of Economerics 100, Gilchris, S. and E. Zakrajšek, 2011a. Credi Spreads and Business Cycle Flucuaions. Working Paper No , Naional Bureau of Economic Research , 2011b. Bank Lending and Credi Supply Shocks, Working Paper, Boson Universiy. He, Z. and A. Krishnamurhy, A Macroeconomic Framework for Quanifying Sysemic Risk. Working Paper, Universiy of Chicago and Norhwesern Universiy. Helbling, T., R. Huidrom, M.A. Kose and C. Orok, European Economic Review 55, Huizinga, H. and L. Laeven, Accouning Discreion of Banks During a Financial Crisis. Discussion Papers 7381, Cenre for Economic Policy Research. Iacoviello, M., Financial Business Cycles. Working Paper, Boson College. Ireland, P., A Mehod for Taking Models o he Daa. Journal of Economic Dynamics and Conrol 28, Jacob, P. and G. Peersman, Dissecing he Dynamics of he US Trade Balance in an Esimaed Equilibrium Model. Working Paper, Ghen Universiy. Jermann, U. and V. Quadrini, Macroeconomic Effecs of Financial Shocks, American Economic Review 102, Jusiniano, A, and B. Preson, Can Srucural Small Open Economy Models Accoun for he Influence of Foreign Disurbances? Journal of Inernaional Economics 81, Jusiniano, A, G. Primiceri and A. Tambaloi, Invesmen Shocks and Business Cycles. Journal of Moneary Economics 57, Kamber, G. and C. Thoenissen, Financial Inermediaion and he Inernaional Business Cycle: he Case of a Small Counry wih Big Risk. CAMA Working Paper 22/2011. Kollmann, R., Essays on Inernaional Business Cycles. PhD Disseraion, Universiy of Chicago. Kollmann, R., Incomplee Asse Markes and he Cross-Counry Consumpion Correlaion Puzzle, Journal of Economic Dynamics and Conrol 20,

35 Kollmann, R., Z. Enders and G. Müller, Global Banking and Inernaional Business Cycles. European Economic Review 55, Kollmann, R. and S. Zeugner, Leverage as a Predicor of Real Aciviy and Volailiy. Journal or Economic Dynamics and Conrol 36, Kollmann, R., Limied Asse Marke Paricipaion and he Consumpion-Real Exchange Rae Anomaly. Canadian Journal of Economics 45, Kollmann, R., W. Roeger and J. in Veld, Fiscal Policy in a Financial Crisis: Sandard Policy versus Bank Rescue Measures. American Economic Review (Papers and Proceedings) 102, 1-7. Le, V., D. Meenagh, P. Minford and M. Wickens, Two Orhogonal Coninens? Tesing a Two-Counry DSGE Model of he US and he EU Using Direc Inference. Open Economies Review 21, Lipinsky, F., Imbalances in he Euro Area and Macroprudenial Policies, Working Paper, London Business School. Meh, C.A., Moran, K., The Role of Bank Capial in he Propagaion of Shocks. Journal of Economic Dynamics and Conrol 34, Mendicino, C. and M. Punzi, Sabilizaion Policy and Boom-Bus Cycles: Moneary and Macro-Prudenial Rules. Bank of Porugal Economic Bullein 17, Nguyen, H., Inernaional Crisis Transmission and Asymmeric Recoveries. Working Paper, World Bank. Nolan, C. and C. Thoenissen, Financial Shocks and he US Business Cycle. Journal of Moneary Economics 56, Perri, F. and V. Quadrini, Inernaional Recessions. Working Paper, Universiy of Minnesoa. Rabanal, P. and V. Tuesa, Euro-Dollar Real Exchange Rae Dynamics in an Esimaed Two-Counry Model: Wha is Imporan and Wha is No? Working Paper 06/177, Inernaional Moneary Fund. Schmi-Grohé, S. and M. Uribe, Closing Small Open Economy Models. Journal of Inernaional Economics 61, Smes, F. and R. Wouers, Shocks and Fricions in US Business Cycles: A Bayesian DSGE Approach. American Economic Review 97, Ueda, K., Banking Globalizaion and Inernaional Business Cycles. Working Paper, Insiue for Moneary and Economic Sudies, Bank of Japan. Van den Heuvel, S., The Welfare Cos of Bank Capial Requiremens. Journal of Moneary Economics 55, Van Wincoop, E., Inernaional Conagion Through Leveraged Financial Insiuions. Working Paper, Universiy of Virginia. 33

36 Table 1. Time series properies of non-bank forcing variables (1990q1-2010q3) (a) Sandard deviaions (in %, diagonal) and cross-correlaions (off-diagonal elemens) US TFP EA TFP US Ieff EA Ieff US G EA G US LabS EA LabS US TFP EA TFP US Ieff EA Ieff US G EA G US LabS EA LabS 1.83 (b) Auocorrelaions US TFP EA TFP US Ieff EA Ieff US G EA G US LabS EA LabS (c) Sandard deviaions of innovaions (in %) US TFP EA TFP US Ieff EA Ieff US G EA G US LabS EA LabS Noe: The Table repors sample momens of empirical measures of US and Euro Area (EA) non-banking forcing variables (in linearly derended log form). Panel (a) repors % sandard deviaions (on main diagonal), and crosscorrelaions (off-diagonal elemens). Panel (b) repors firs-order auocorrelaions. Panel (c) repors % sandard deviaions of residuals of univariae AR(1) equaions fied o each variable. TFP: oal facor produciviy ( ); Ieff: invesmen efficiency ( ); G: auonomous spending (governmen N consumpion plus ne expors o hird counries); LabS: Labor supply shock ( ). Log TFP is esimaed as ln( Y) 0.7ln( N) where Y and N are GDP and employmen, respecively. The esimae of invesmen efficiency is he raio of he CPI o he invesmen deflaor. The esimae of he log labor supply shock is ln( Y) ln( N) ln( C), where C is privae consumpion. 34

37 Table 2. Model parameers: prior disribuion and poserior disribuion for hree model varians Poserior disribuion Model varians wih independen shocks Model varian wih '' 0, no correlaed shocks Prior disribuion '' 0 banking shocks '' 0 Parameer Mean Sd Disr. Mean Sd Mean Sd Mean Sd (1) (2) (3) (4) (5) (6) (7) (8) (9) Behavioral parameers 4 '' G '' G G Sandard deviaions (%) of innovaions o forcing variables IG IG IG IG G IG G IG IG IG IG IG IG Auocorrelaions of forcing variables B B B B G B G B B B B B B Feedback parameers of loan losses and required capial raio o GDP N N Sandard deviaions (%) of measuremen errors GDP US IG GDP EA IG C US IG C EA IG I US IG I EA IG N US IG N EA IG

38 Table 2.-- coninued Loans US IG Loans EA IG Bnk cap.raio IG Loan spread IG Log marginal likelihood Noes: Cols. (1) and (2) shows he means and sandard deviaions of he prior disribuion for model parameers lised in he lefmos Column. Col. (3) indicaes he disribuion funcion of he prior (B: Bea; G: Gamma; IG: Invered Gamma; N: Normal). Cols. (4)-(9): saisics of poserior parameer disribuion (means, sandard deviaions), for differen model varians. Cols. (4)-(5): model varian wih independen forcing variables, operaive bank capial requiremen ( '' 0) and banking shocks. Cols. (6)-(7): model varian wih independen shocks, no operaive bank capial requiremen ( '' 0), no banking shocks. Cols. (8)-(9): model varian wih correlaed shocks, operaive bank capial requiremen ( '' 0) and banking shocks. Enries repored as --- in Cols. (4)-(7) represen parameers of he model varian wih independen shocks ha are se a zero. Enries repored as --- in Cols. (8)-(9) represen parameers of he model varian wih correlaed shocks ha are calibraed; specifically, I se he auocorrelaions of he 8 non-banking forcing variables, and he covariance marix of hese variables, equal o he sample auocorrelaions and sample covariance marix of empirical measures of he non-banking shocks (1990q1-2010q3), respecively (from Table 1); see Secion 3.2. Parameer definiions: 4 '': sensiiviy of loan rae spread o bank capial raio; a 1 percenage poin increase in he bank capial raio lowers he lending rae spread by 4 '' percenage poins ( is he seady sae bank capial requiremen). '': curvaure of invesmen cos funcion. : coefficien of relaive risk aversion z z, : sandard deviaion of innovaion o forcing variable z, and auocorrelaion of z, wih z represening he following shocks--, : Home and Foreign TFP;, : Home and Foreign invesmen efficiency; GG, : governmen purchases;, : Home and Foreign labor supply;, : Home and Foreign loan loss; : required bank capial raio. [ ]: feedback coefficien of Home and Foreign loan loss [required bank capial raio] o GDP; see equaions (10),(11). The esimaion uses quarerly ime series on 12 ime series: US and EA GDP, consumpion, invesmen, employmen and real loans; he US commercial bank lending rae spread and he US commercial bank capial raio. The loan spread and he capial raio are demeaned; oher empirical variables are linearly derended in log form. See Figure 1 for daa plos and Daa Appendix for definiions of variables and daa sources. Sample period: 1990q1-2010q3 (83 periods). Poserior disribuions are compued using he Random Walk Meropolis algorihm (250,000 draws of which he firs 50,000 were discarded). 36

39 Table 3. Business cycle saisics generaed by hree model varians Model varians wih independen shocks Model varian wih correlaed shocks '' 0 '' 0, no Inves. Loan banking shocks '' 0 Shocks: All NonBk Bnk TFP Eff. G LabS Loss All NonBk All NonBk Bnk Daa (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (13) (14) (15) (a) US momens Sandard deviaions (in %) GDP Consumpion Invesmen Employmen Loans Bank cap raio Loan spread Correlaions wih domesic GDP Consumpion Invesmen Employmen Loans Bank cap raio Loan spread (b) EA momens Sandard deviaions (in %) GDP Consumpion Invesmen Employmen Loans Bank cap raio Loan spread Correlaions wih domesic GDP Consumpion Invesmen Employmen Loans Bank cap raio Loan spread (c) Cross-counry correlaions GDP Consumpion Invesmen Employmen Loans Loan spread Noe: Momens of HP filered model variables (compued for poserior mode of model parameers) are shown. The bank capial raio is expressed in facional unis. The loan rae spread is in fracional unis per annum. Oher variables are normalized by seady sae values. Cols. (1)-(11) perain o model varians wih independen shocks. Cols. (1)-(9): model varian wih operaive bank capial requiremen ( '' 0) and banking shocks. Col. (1) [ All ] assumes all 11 join shocks and measuremen error. In Cols. (2)-(9), subses of shocks used in isolaion, wihou measuremen error (model no re-esimaed). Col. (2) [ NonBk ]: jus 8 non-banking shocks; Col. (3) [ Bnk ]: jus 3 banking shocks. Col. (4) [ TFP ]: jus Home and Foreign TFP shocks; Col. (5) [ Inves.Eff ]: jus invesmen efficiency shocks; Col. (6) [ G ]: jus governmen purchases shocks; Col.(7) [ LabS ]: jus labor supply shocks; Col.(8) [ Loan Loss ]: jus loan loss shocks; Col. (9) [' '] : jus shock o required bank capial raio. Cols. (10)-(11): model varian wihou operaive bank capial requiremen ( '' 0) and wihou banking shocks. Col. (10) [ All ]: all non-banking shocks and measuremen error; Col. (11) [ NonBk ]: jus non-banking shocks, no measuremen error. Cols. (12)-(14): model varian wih correlaed shocks, operaive bank capial requiremen ( '' 0) and banking shocks. Col. (12) [ All ]: all 11 shocks and measuremen error. Col. (13) [ NonBk ]: jus 8 non-banking shocks; Col. (14) [ Bnk ]: jus 3 banking shocks. Col. (15): empirical momens of HP filered daa (GDP, consumpion, invesmen, employmen and loans are logged before filering). Saisics for he EA bank capial raio use he US capial raio as a proxy. Sample: 1990q1-2010q3 (excep for EA loan spread: 2003q1-10q3). 37

40 Table 4. Share (%) of variance of HP filered variables due o non-banking shocks and o banking shocks GDP Consumpion Invesmen Employmen Loans Bank Cap. Loan US EA US EA US EA US EA US EA Raio Spread (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (a) Model varians wih independen forcing processes (a1) Baseline specificaion, '' 0 [4 '' 0.21; LML= ] NonBk NonBk US NonBk EA Bank US EA (a2) Model version wih consan required bank capial raio, [4 '' 0.69; LML= ] NonBk Bank (a3) Esimaes based on alernaive loan rae spread and bank loans daa (a3.1) Loan rae spread replaced by survey of loan officer spread measure [4 '' 0.21; LML= ] NonBk Bank (a3.2) Loan spread replaced by Gilchris-Zakrajsek excess bond premium [4 '' 0.30; LML= ] NonBk Bank (a3.3) Toal bank loans replaced by business loans [4 '' 0.18; LML= ] NonBk Bank (a3.4) US business lending capaciy; EA business loans [4 '' 0.10; LML= ] NonBk Bank (b) Model varian wih correlaed forcing variables [4 '' 0.48; LML= ] NonBk Bank US EA Noe: This Table repors shares of he variances of endogenous variables (HP-filered) ha are accouned for by differen ypes of shocks (based on model predicions evaluaed a poserior modes of model parameers). Panel (a) perains o varians of he model wih independen forcing variables. (a1): baseline model; (a2): model wih consan required bank capial raio; (a3): esimaes of he model based on alernaive measures of spreads and loans. (a3.1): baseline loan rae spread replaced by ne percenage of banks increasing spreads of loan raes over cos of funds from he US senior loan officer opinion survey, SLOOS. (a3.2): baseline loan rae spread replaced by he Gilchris and Zakrajsek (2011a) excess bond premium series. (a3.3): he baseline loan series for US and EA replaced by loans o he non-financial business secor. (a3.4): baseline US loan series replaced by US business lending capaciy measure of Gilchris and Zakrajsek (2011b); baseline EA loan series replaced by EA bank lending o he non-financial business secor. Panel (b) perains o he model varian wih correlaed forcing variables. 38

41 Table 4 cd. The rows labeled NonBk show shares of he variances of HP filered variables accouned by he non-banking forcing variables: US and Euro Area [EA] TFP, invesmen efficiency, governmen purchases and labor supply shocks). Rows labeled NonBk US [ NonBk EA ]: variance shares accouned for by US [EA] non-banking shocks. Rows labeled Bank : variances shares accouned by he 3 banking shocks. US EA Rows labeled, and : variance shares accouned for by shocks o Home loan losses, o Foreign loan losses and o he required bank capial raio, respecively. Decomposiions are compued a he poserior mode of he esimaed parameers. The variance shares are shown for he variables lised above Cols. (1)-(12). Cols. (1)-(2): US and EA GDP; Cols. (3)-(4): US and EA consumpion; Cols. (5)-(6): US and EA invesmen; Cols. (7)-(8): US and EA employmen; Cols. (9)-(10): US and EA loans; Col. (11): bank capial raio; Col. (12): Loan rae spread. 4 '': a 1 percenage poin increase in he bank capial raio lowers he lending rae spread by 4 '' percenage poins. LML: log marginal likelihood 39

42 Table 5. Baseline model wih independen shocks: dynamic responses o innovaions GDP Consumpion Invesmen Employmen Loans Bank Cap. Loan Horizon US EA US EA US EA US EA US EA Raio Spread (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (11) (12) (a) US TFP shock (1%) (b) US invesmen efficiency shock (1%) (c) US governmen purchases shock (1%) (d) US labor supply shock (1%) (e) US loan loss shock (1% of seady sae quarerly GDP) (f) EA loan loss shock (1% of seady sae quarerly GDP) (g) Shock o required bank capial raio (1 percenage poin) Noe: The Table shows dynamic responses o exogenous shocks, afer 0, 4 and 20 quarers (see lef-mos column labeled Horizon ), of he variables lised a he op of he Table. The responses are compued a he poserior mode of he esimaed parameers. In each case, an isolaed innovaion is considered, assuming ha all oher exogenous innovaions are zero. Panel (a): 1% innovaion o Home TFP ( ); Panel (b): 1% innovaion o US invesmen efficiency ( ); Panel (c): 1% innovaion o N governmen purchases ( G ); Panel (d): 1% innovaion o US labor supply preference parameer ( ); Panel (e): innovaion o Home loan loss ( ) worh 1% of seady sae quarerly GDP; Panel (f) innovaion o Foreign loan loss ( ) worh 1% of seady sae quarerly GDP; Panel (g) innovaion ha raises required bank capial raio ( ) by 1 percenage poin. Cols. (1)-(2): Responses of US and EA GDP; Cols. (3)-(4): US and EA consumpion; Cols. (5)-(6): US and EA invesmen; Cols. (7)-(8): US and EA employmen; Cols. (9)-(10): US and EA loans; Col. (11): bank capial raio; Col. (12): Loan rae spread. Responses of he bank capial raio are in basis poins. Responses of he loan spread are in basis poins per annum. Oher responses are in percenage poins of seady sae values. 40

43 Fig. 1. Time series used in esimaion Noe: The Figure shows he US and Euro Area (EA) ime series (1990q1-2010q3) used in esimaion. Loans represen oal credi by US commercial banks and EA Moneary Financial Insiuions (MFI). The US bank capial raio is he capial raio of US commercial banks. Bank capial raio and loan spread (p.a.) are demeaned. Oher variables are logged and derended. Shaded areas indicae US recessions (NBER daes). Fig. 2. US and Euro Area (EA) loan rae spreads Noe: Loan spreads (p.a.) are no demeaned. Sample period for US (EA) loan spread: 1990q1-2010q3 (2003q1-2010q3) Shaded areas: US recessions (NBER daes). Fig. 3. US bank capial, US loan spreads and US excess bond premium Noe: In boh panels, he solid line shows he demeaned US bank capial raio. The lef panel also plos he demeaned baseline US loan spread (dashed line) and he demeaned ne percenage of US banks increasing spread, from Survey of Senior Loan Officers Opinion Survey (doed line). (The SLOOS series is scaled so ha is sandard deviaion equals ha of he baseline loan spread.) The righ panel plos he demeaned US excess commercial bond premium of Gilchris and Zakrajsek, 2011a (dashed line). Sample period: 1990q1-2010q3. Shaded areas: US recessions (NBER daes). 41

44 Fig. 4. US commercial banks: oal loans, business loans and business lending capaciy Noe: The solid line shows oal US bank credi (baseline measure); dashed line: business lending; dashed-doed line: US business lending capaciy (Gilchris and Zakrajsek (2011b). All series are linearly derended in log form. Sample period: 1990q1-2010q3. Shaded areas: US recessions (NBER daes). Figure 5. EA banks: oal loans and business loans Noe: The solid line shows oal EA bank credi (baseline measure); dashed line: loans o non-financial corporaions. Boh series are linearly derended in log form. Sample period: 1990q1-2010q3. Shaded areas: US recessions (NBER daes). 42

45 Figure 6. Hisorical decomposiions, model wih independen shocks (baseline specificaion) Noe: Based on he baseline model wih independen shocks (using poserior mode of esimaed parameers), he Figure shows he hisorical conribuions of banking shocks (hin solid lines), US non-banking shocks (dashed-doed lines), Euro Area (EA) non-banking shocks (dashed lines) o hisorical series, 1990q2-2010q3 (hick solid lines). The hisorical bank capial raio and loan rae spread series (p.a.) are demeaned, he oher hisorical series are linearly derended in log form. Shaded areas: US recessions (NBER daes). 43

46 Figure 7. Hisorical decomposiions, model wih correlaed shocks Noe: Based on he model varian wih correlaed shocks (poserior mode of esimaed parameers), he Figure shows he hisorical conribuions of banking shocks (hin solid lines), US non-banking shocks (dashed-doed lines), Euro Area (EA) nonbanking shocks (dashed lines) o hisorical series, 1990q2-2010q3 (hick solid lines). The hisorical bank capial raio and loan rae spread series (p.a.) are demeaned, he oher hisorical series are linearly derended in log form. Shaded areas: US recessions (NBER daes). 44