The Shifting Drivers of Global Liquidity

Transcription

1 Federal Reserve Bank of New York Staff Reports The Shftng Drvers of Global Lqudty Stefan Avdev Leonardo Gambacorta Lnda S. Goldberg Stefano Schaff Staff Report No. 819 June 2017 Ths paper presents prelmnary fndngs and s beng dstrbuted to economsts and other nterested readers solely to stmulate dscusson and elct comments. The vews expressed n ths paper are those of the authors and do not necessarly reflect the poston of the Federal Reserve Bank of New York, the Federal Reserve System, or the Bank for Internatonal Settlements. Any errors or omssons are the responsblty of the authors.

2 The Shftng Drvers of Global Lqudty Stefan Avdev, Leonardo Gambacorta, Lnda S. Goldberg, and Stefano Schaff Federal Reserve Bank of New York Staff Reports, no. 819 June 2017 JEL classfcaton: F34, G10, G21 Abstract The post-crss perod has seen a consderable shft n the composton and drvers of nternatonal bank lendng and nternatonal bond ssuance, the two man components of global lqudty. The senstvty of both types of flows to U.S. monetary polcy rose substantally n the mmedate aftermath of the global fnancal crss, peaked around the tme of the 2013 Federal Reserve taper tantrum, and then partally reverted toward pre-crss levels. Conversely, the responsveness of nternatonal bank lendng to global rsk condtons declned consderably after the crss and became smlar to that of nternatonal debt securtes. The ncreased senstvty of nternatonal bank flows to U.S. monetary polcy has been drven manly by post-crss changes n the behavor of natonal bankng systems, especally those that ex ante had banks that were less well captalzed. By contrast, the post-crss fall n the senstvty of nternatonal bank lendng to global rsk was manly owng to a compostonal effect, drven by ncreases n the lendng market shares of natonal bankng systems that were better captalzed. The post-2013 reversal n the senstvtes to U.S. monetary polcy partally reflects the expected dvergence n the monetary polces of the Unted States and other advanced economes, hghlghtng the senstvty of captal flows to the degree of commonalty of cycles and the stance of polcy. Moreover, global lqudty fluctuatons have largely been drven by polcy ntatves n credtor countres. Polces and prudental nstruments that renforced lendng banks captalzaton and stable fundng levels reduced the volatlty of nternatonal lendng flows. Key words: global lqudty, nternatonal bank lendng, nternatonal bond flows, captal flows Goldberg: Federal Reserve Bank of New York (emal: lnda.goldberg@ny.frb.org). Avdev, Gambacorta: Bank for Internatonal Settlements (emals: stefan.avdev@bs.org, leonardo.gambacorta@bs.org). Schaff: Boccon Unversty (emal: stefano.schaff@phd.unboccon.t). The authors thank Mattheu Bussère, Stn Claessens, Catherne Koch, Robert N. McCauley, Patrck McGure, Sergo Schmukler, Hyun Song Shn, Cédrc Tlle, Phlp Wooldrdge, and partcpants at the followng workshops, conferences, and semnars: the BIS-CGFS workshop Research on Global Fnancal Stablty: The Use of BIS Internatonal Bankng and Fnancal Statstcs (Basel, May 2016), the ECB-Federal Reserve Board-New York Fed Global Research Forum on Internatonal Macroeconomcs and Fnance (New York, November 2016), Fordham Unversty Macro Internatonal Fnance Workshop 2017 (New York, Aprl 2017), Chapman Conference on Money and Fnance: Systemc Rsk and the Organzaton of the Fnancal System (Los Angeles, May 2017), and a semnar at the Federal Reserve Bank of San Francsco. Lnda Goldberg developed parts of ths proect whle vstng the Bank for Internatonal Settlements under the Central Bank Research Fellowshp Programme. Bat-el Berger and Pamela Poglan provded excellent research assstance. The vews expressed n ths paper are those of the authors and do not necessarly reflect the poston of the Federal Reserve Bank of New York, the Federal Reserve System, or the Bank for Internatonal Settlements.

3 1. Introducton Internatonal captal flows channel fnancal resources across borders to both publc and prvate sector borrowers. Ther two largest debt components, cross-border bank loans and nternatonal bond ssuance, are the man elements of global lqudty (Bank for Internatonal Settlements, 2011a and 2011b). As such, they are mportant determnants of global fnancal condtons and worldwde economc actvty. The exstng emprcal lterature has establshed that global (push) and local (pull) factors are both mportant drvers of cross-border bank loans and nternatonal bond ssuance. Among pull factors, the lterature has dentfed recpent country output growth, soveregn credt rsk, and the degree of captal account openness. The most mportant global drvers dentfed by the lterature have been advanced economy monetary polces, global rsk averson and global output growth (e.g. Forbes and Warnock, 2012a; Mranda-Agrppno and Rey, 2015; and Cerutt, Claessens and Ratnovsk, 2017). Whle studes generally focus on dentfyng the drvers, they seldom examne the how and why of evolvng senstvtes to global factors. Yet, the structure and volatlty of crossborder bank loan and nternatonal bond flows clearly have changed consderably n the aftermath of the Global Fnancal Crss (GFC). In the mmedate aftermath of the crss, crossborder loans contracted sharply. Ths was followed by a feeble recovery and a second sharp contracton durng the peak of the euro area crss. By contrast, nternatonal bond ssuance was relatvely robust durng the post-crss perod. As a consequence, the composton of global lqudty has shfted away from cross-border bank loans and towards nternatonal bonds n what has been dubbed the second wave of global lqudty (Shn, 2013). Meanwhle, events such as the taper tantrum n 2013, when the Federal Reserve sgnalled t would start taperng ts bond buyng program, were marked by especally sharp changes n some captal flows to emergng markets (Khatwada, 2017). Graph 1 contans a summary of the behavour of the varous types of flow at the global level for bank and nonbank borrowers. An extensve lterature (revewed n the next secton) dscusses the vulnerablty of borrowng countres to nternatonal surges and retrenchments, and the potental polcy tools avalable for contanng excessve changes. In ths paper, we start wth the conecture that both the compostonal changes n the landscape of nternatonal fnancal flows and the more novel advanced economy monetary and regulatory nstruments have the potental to fundamentally alter global lqudty and ts drvers. We specfcally nvestgate the changes n senstvtes of the man components of global lqudty to global drvers durng the post-crss perod. We drll down nto observed changes, test for ther proxmate reasons, and dstngush between persstent versus transtory drvers. To acheve these ends, we draw on multple databases on global lqudty component flows from both borrower country and credtor country perspectves, dstngushng between nstrument types (nternatonal debt securtes versus nternatonal bank loans), and between borrowng sectors (bank versus non-bank). Usng the BIS Internatonal Debt Securtes (IDS) Statstcs and the BIS Locatonal Bankng Statstcs (LBS), we create a quarterly panel of nternatonal bank loan and bond flows to 64 recpent countres for the perod between 2000:Q1 and 2015:Q4. In addton, we utlse the BIS 1

4 Consoldated Bankng Statstcs (CBS) n order to assgn loans to specfc natonal lendng bankng systems. Usng Bankscope, we obtan nformaton on lendng bankng system balance sheet characterstcs. We also ncorporate a range of other data on prudental nstrument and monetary polcy developments, from the perspectve of both the borrowers and the credtor countres. Advanced economy monetary polces, as well as shadow measures that capture unconventonal polces, are also ncorporated nto the analyss. After replcatng the types of global factor and local factor results documented n pror studes, our analytcal contrbutons centre around three man sets of results. Our frst key result s that nternatonal captal flow senstvtes to global factors have changed consderably snce the GFC. Advanced economy monetary polcy, proxed by US monetary polcy, became a more potent drver of both cross-border loan and nternatonal bond flows. The estmated polcy mpacts peaked n 2013 and then partally retraced toward pre-crss levels whle remanng elevated. Meanwhle, the senstvty of cross-border bank loan flows to global rsk condtons declned consderably post-crss and became smlar to the respectve rsk senstvty observed for nternatonal bond flows. In fact, nternatonal bank loan and bond flows became more smlar n terms of ther responsveness to global factors after the GFC. Overall, aggregate global lqudty flows (the sum of nternatonal bank loan and bond flows) have become more senstve to US monetary polcy and less senstve to global rsk. The second set of results shows that post-crss shfts n senstvtes of nternatonal bank loan and bond flows to global factors, observed from the borrower perspectve, arse from a combnaton of changes n the country composton of lendng bankng systems and from changes n the behavour of the credtors nvolved n nternatonal fnancal flows. Workng across multple databases, we show an ncrease n the responsveness of flows from ndvdual lendng bankng systems to US monetary polcy. We also fnd evdence of a compostonal shft toward natonal lendng bankng systems wth lower senstvty to global rsk condtons. We drll deeper nto the type of varaton observed to nvestgate the contrbutons of a range of prudental measures, bank busness model features, and monetary regmes n the credtor countres. We fnd that the features of fnancal ntermedares that prevously have been shown to stablze domestc bank lendng response to lqudty rsk, lke bank captal ratos and depost fundng, also support expanson of nternatonal market share relatve to weaker peer country systems and help explan changng behavours. Natonal bankng systems that were better captalzed before the GFC experenced smaller post-crss rses n senstvty to US monetary polcy and larger ncreases n nternatonal lendng shares. Hgher ex-ante shares of deposts n total fundng and of locally booked clams n total foregn clams were also assocated wth larger ncreases n nternatonal lendng market shares. Tghter local reserve requrements pre-crss were assocated wth relatve expansons of nternatonal market shares n the post crss perod. Even post-gfc there has been a sgnfcant evoluton of credtor senstvtes to global rsk and US monetary polcy. Wthn the post-gfc perod, we te ths evoluton to the roles of bankng sector performance metrcs and relatve monetary polcy stances across advanced economes. Hgh senstvtes to US monetary polcy post-gfc are ted to the relatve path of expected US monetary polcy vs-à-vs that of other maor advanced economes, proxed usng two-year nterest rate futures data. In partcular, senstvty n cross-border loans s enhanced n the years mmedately followng the crss, consstent wth an nterpretaton that 2

5 US monetary polcy served as a stronger ndcator of global monetary polcy n a perod of low growth across advanced economes. Ths effect unwound as a percepton took hold of greater polcy dvergence across advanced economes startng n Proxes for the busness models of credtor bankng systems played less of a role of n ths evoluton. These results contrbute mportantly to both research and polcy debates around global lqudty and local stablzaton. One pertnent queston s whether the enhanced dversfcaton across fnancng types wll have dfferent consequences n the case of future stress epsodes, as well as n normal perods. Ths s an especally pertnent ssue f, ex-ante, bank loan and debt securtes fnancng agents are subect to dstnct degrees of leverage and balance sheet constrants. We show that pre-crss borrowers experenced more global factor senstvty n cross border loans than n nternatonal debt securtes. As a consequence, nternatonal debt securtes remaned relatvely robust durng the global fnancal crss. Post-crss, the senstvtes of both types of fnancng have become more smlar. Another queston s how stablzaton challenges across countres borrowng nternatonally evolve n post-crss perods and when the synchronzaton of busness cycles across countres s enhanced. The range of evdence we provde across econometrc exercses suggests that the large ncreases n senstvtes to US monetary polcy post GFC may have been a transtory phenomenon, whereas the declnes n global lqudty senstvty to rsk measures may be more persstent. At least n nternatonal bank flows, behavoural changes n the perod mmedately followng the GFC were drven largely by the convergence n advanced economy monetary polces. These transtory effects gradually weakened when the monetary polces of advanced economes started to dverge n More persstent effects may come from the ncreased market shares of better-captalzed lendng bankng systems, whose nternatonal lendng tends to be less responsve to fluctuatons n global rsk condtons. The mplcatons would be that evolvng global drvers also change the scope for monetary autonomy and prudental polces optons for borrowng countres. Moreover, a potentally mportant consequence of the focus on captal and stable fundng n credtor countres are reduced ampltudes of global lqudty surges and waves as observed by borrowers. The remander of the paper s organsed nto sx sectons. Secton 2 revews relevant fndngs of the exstng lterature on global lqudty and ts drvers, also focusng on dfferences between banks and non-banks as credtors and debtors. Secton 3 presents the econometrc methodology that we employ n respectve emprcal nvestgatons. Secton 4 descrbes the data. Secton 5 provdes the emprcal results and related dscusson. Secton 6 presents robustness tests. Secton 7 concludes. 2. Prevous lterature Global lqudty and drvers have been explored n a number of related studes. The most extensve prevous lterature s on nternatonal captal flows. The second strand of lterature s more explctly focused on global lqudty, nternatonal debt securtes versus loans, and constrant dfferences across banks and non-banks. The thrd thread of lterature addresses nternatonal monetary polcy spllovers, coverng the transmsson channels through banks 3

6 and captal markets, nterest rate and asset prce co-movements, and broader ssues around the structure of the nternatonal monetary system and polcy nstrument avalablty. The large lterature on the drvers of captal flows focuses most extensvely on emergng markets, and more recently consders advanced economes also as destnatons of captal. Surges n cross-border flows to EMEs reflect mproved macroeconomc fundamentals of the borrowng country (pull and local factors) and more favourable global condtons of a prmarly cyclcal nature (push and global factors). Examples of such studes nclude those by Calvo et al. (1993), Ghosh and Ostry (1993), Fernandez-Aras (1996), Taylor and Sarno (1997), and Chuhan et al. (1998). 1 The emphass of the lterature then shfted specfcally to understandng gross (as opposed to net) nternatonal flows and dstngushng across dfferent nsttutonal partcpants. Portes and Rey (2005) show that nformaton frctons and technology matter for the relatve stablty of gross flows. Broner et al. (2013) show the hgher volatlty n gross flows than n net flows, specfcally n the context of busness cycles and crses. Forbes and Warnock (2012b) present a systematc framework for analysng captal flows whereby extreme epsodes are classfed nto four categores: surges, stops, flght and retrenchment. Ths work carefully documents how the most extreme captal flows epsodes are drven by global factors, notably global rsk averson. Mles-Ferrett and Tlle (2011) document heterogenety n the behavour of varous captal flows components durng the Global Fnancal Crss, emphaszng the domnant contracton of nternatonal bankng flows and the relatve stablty of foregn drect nvestment. Post-crss declnes n bank-based cross-border lendng, partcularly by euro area banks, have been descrbed n some analyss as fnancal deglobalzaton (Rose and Weladek, 2011; Forbes et al. 2015) or the great cross-border bank deleveragng (Cerutt and Claessens, 2017; Bussère et al., 2016). The explanatons provded nclude weaker economc actvty; captal controls and the slower pace of lberalzaton; deleveragng, and rsk averson (CGFS 2011). Related research uses mcro-bankng data to explore nternatonal fnancal lnkages. Cetorell and Goldberg (2012a), workng wth bank-specfc data, show that contractons n nternatonal lendng by global banks durng the crss were related to balance sheet shocks through holdngs of asset-backed commercal paper. Contractons n some cases are shown to be magnfed by polcy nterventons. Across UK banks, prudental polces and unconventonal monetary polcy n the form of a fundng for lendng scheme ontly contrbuted to a retrenchment of cross-border lendng wth dfferental effects across banks (Forbes, Renhardt and Weladek, 2017). More broadly, across countres prudental polcy effects on nternatonal bank flows were assocated wth contractons n some cases and expansons elsewhere (Buch and Goldberg, 2017). The composton of lendng bankng systems, as some countres wth banks that were well-captalzed pre-crss expanded nternatonal actvtes post-crss, as occurred for Canada among others (Damar and Mordel, 2017). The actual channels of transmsson that drve these co-movements have been dentfed by heterogenety across bank-balance sheets. Cetorell and Goldberg (2012b) show bank transmsson through nternal captal markets and heterogenety n shock transmsson to countres dependng on ther global bank-specfc mportance n lendng and fundng 1 See Koepke (2015) for a comprehensve summary of the lterature n the drvers of captal flows to EMEs. 4

7 actvty. Bruno and Shn (2015b) pont to a drect role for balance sheet valuaton by banks, as monetary polcy spllovers drve cross-border bank captal flows and the US dollar exchange rate through the bankng sector. Bruno and Shn (2015a) demonstrate that epsodes of apprecaton of the US dollar are assocated wth deleveragng of global banks and an overall tghtenng of global fnancal condtons. Banks also have been shown to have shfted ther treatment of ther soveregn exposures pre- versus post-crss, later havng more rsk assgned to these postons and posng dfferent constrants (Acharya et al., 2013; Farh and Trole, 2016; De Grauwe and J, 2013). Banks have more pronounced bank lendng channel responses to lqudty rsk when they have low levels of captalzaton and low depost fundng shares (Cornett et al. 2011; Buch and Goldberg, 2015). A parallel and rapdly growng lterature on the man drvers of global lqudty emphaszes specfc global fnancal factors. Mranda-Agrppno and Rey (2015) argue that one global factor explans an mportant part of the varance of a large cross secton of returns of rsky assets and nterpret ths factor as tme-varyng market-wde rsk averson lnked to US monetary polcy. Smlar man drvers are documented by Cerutt, Claessens and Ratnovsk (2017) and nclude US monetary polcy, global uncertanty (proxed by the VIX), the exchange rate value of the US dollar, and European bank condtons. Non-bank fnancal ntermedatons also have recently receved attenton. Another type of shft occurred n the composton of nternatonal captal flows worldwde, as the frst phase of global lqudty through banks was replaced to some degree by a second phase of global lqudty through corporate bond fnancng, partcularly for emergng market borrowers (Shn, 2013). 2 Chung et al. (2016) lnk the evoluton of global monetary aggregates to the fnancal actvtes of non-fnancal corporatons (NFCs), wth the non-core labltes of NFCs reflectng global credt condtons and predctng global trade and growth. McCauley et al. (2015) fnd that unconventonal monetary polcy contrbuted to shftng the balance of dollar credt transmsson from global banks to global bond nvestors, demonstratng a negatve relatonshp between the term premum on 10-year Treasury bonds and nternatonal bond ssuance durng the post-crss perod. Fnally, the long lterature on nternatonal monetary polcy spllovers, wth ts focus on short-term nterest rate co-movement and the constrants on stablzaton polces posed by the nternatonal monetary trlemma, s drectly relevant for analyss of global lqudty drvers. Shambaugh, Obstfeld and Taylor (2005) show that monetary polcy rates across a large sample of countres can closely track advanced economy polcy rates, partcularly wth the rates of countres playng a central role n the nternatonal monetary system. The form of exchange rate and monetary regmes n place nfluences the degree of co-movement, although greater near term autonomy can come from some restrctons on nternatonal captal movements (Klen and Shambaugh, 2008) and lower levels of bankng globalzaton (Goldberg 2013). Prudental polcy nstruments are argued to potentally afford countres 2 These observatons pertan to volumes of cross-border flows, not to co-movements of asset prces. Durng ths same broad perod, co-movements n nternatonal asset prces contnue to be at least as strong and senstve to global rsk sentment and lqudty condtons as pre-crss state. Ths type of evdence does not support de-globalzaton. 5

8 relef from nternatonal captal flow movements (Rey, 2013), although the evdence to date on consequences for flows through global banks s mxed and not yet large n magntude. 3 Collectvely these papers show the mportance for global lqudty and nterest rate comovements of constrants on dfferent types of nsttutons, the shfts n mportance of these nsttutons, and the scope for polcy responses to nternatonal fnancal flows. Our study takes an ntegrated approach by studyng the flows through banks and non-banks as borrowers and lenders. We analyse the effects of key global lqudty drvers, ncludng rsk and advanced economy monetary polcy, and tests conectures about why and how these effects change over tme. Our analyss tes together the dynamsm n the effects of dfferent global factors, showng the roles of mcro-bankng characterstcs, composton of credtors, monetary polcy regmes, and prudental polces of both borrowers and credtors. 3. Emprcal strategy The emprcal strategy mplemented has three man parts. The startng pont s the nternatonal captal flow and global lqudty specfcaton whereby nternatonal fnancal flows are explaned by global (push) and country-specfc (pull) drvers. We replcate fndngs from that lterature as a baselne before delvng nto dfferences n senstvtes to global (and other) factors over tme, as well as across dfferent borrower groups (banks and non-banks) and across dfferent types of fnancng nstrument (nternatonal clams and nternatonal debt securtes). After havng dentfed sgnfcant changes n patterns pre- and post- global fnancal crss, the second part of the emprcal strategy focuses on the pre- versus post patterns of changes n global lqudty senstvty to global factors, wth a specfc set of tests for changes n the composton versus the behavour of credtors. The patterns of composton and behavour then are related to ex ante balance sheet condtons and regulatory polces of lendng bankng systems. The last part of the emprcal strategy relates perod-by-perod tme varaton n the effects of advanced economy monetary polcy and rsk senstvty to evolvng credtor bank balance sheet characterstcs and to degrees of dvergence across monetary polces of advanced economes. 3.1 Baselne analyss The baselne model for global and local factors n nternatonal captal flows follows the lterature by ntroducng push global factors and pull local factors, and s gven by: GrRateY t = β 1 ΔFFR t + β 2 logvix t + β 3 ΔlogGlobalGDP t +β 4 ΔSovRatng t + β 5 ChnnIto t + β 6 ΔlogGDP t + μ + ε t (1) where denotes country and t s tme. Our baselne specfcaton consders the ssue of nternatonal captal flows and global lqudty drvers from the perspectve of the borrowng country. Global lqudty s dvded nto component cross-border flows by nstrument and by type of borrower, wth these components explored separately and n aggregate. For our 3 Extensve dscusson and cross-country evdence s provded n the March 2017 volume of the Internatonal Journal of Central Bankng n whch a range of country and cross country studes document experences through global banks and hosted afflates of foregn banks. Buch and Goldberg (2017) provde a meta-analyss of fndngs. 6

9 analyss, Y t can be cross-border loans - to all sectors, to banks, to non-banks - or nternatonal debt securtes - ssued by all sectors, by banks or by non-banks. As s standard n the lterature, the model s expressed n statonary varables to avod problems of spurous correlatons. The nternatonal flows on the left-hand sde of the equaton are expressed n growth rates. The rght-hand-sde of the equaton contans three global lqudty drvers - the US federal funds rate (as a gauge for the stance of US monetary polcy), the VIX (as a measure global rsk condtons) and global GDP (as an ndcator of global economc actvty). As the US federal funds rate does not reflect all of the monetary polcy nterventons for the post GFC perod, we use the Wu-Xa shadow rate measure (Wu and Xa, 2016) as a proxy to reflect both conventonal and unconventonal monetary polces. 4 The local factors correspondng to borrowng country and flow type nclude soveregn credt ratngs, the Chnn-Ito ndex of fnancal openness (Chnn and Ito, 2008) and local GDP growth. The latter measures overall economc performance. Soveregn ratngs proxy the role of country rsk and the perceved credtworthness of borrowers by country. The Chnn-Ito ndex gauges the degree of captal account openness. The Fed funds rate and the soveregn ratngs are n frst dfferences, whle local and global GDP are n growth rates. The Chnn-Ito ndex s n levels and the VIX enters the equaton n logs. 5 The model s estmated under the assumpton that the two key global lqudty drvers, the Fed funds rate and the VIX, are exogenous when controllng for local and global GDP, government ratngs and degree of fnancal openness. As both anecdotal evdence and the lterature dscusson of phases of fnancal globalzaton hnt at the presence of a possble structural break around the global fnancal crss, we modfy the full tme perod approaches of the lterature and allow for shfts n the drvers of global lqudty. Rather than exogenously mposng a partcular break date, we conduct a formal search for an endogenous structural break n the parameters of the model. Usng the tools developed n Ba (1994, 1997), Kurozum (2002) and Carron--Slvestre and Sansó (2006), for each quarter T startng n 2007:Q1, we estmate the followng equaton: where GrRate Y t = β X t + μ + I(t T)(κ + γ X t ) + ε t X t = (ΔFFRt, logvix t, ΔlogGlobalGDP t,δloggdp t, ΔSovRatngt, ChnnItot ) (2) and I(t T) s an ndcator functon that takes the value 1 when t T and 0 otherwse. Notce that for each canddate break date T, all the parameters of equaton (2) are dfferent. For each type of cross-border flow Y and each quarter T we can compute the sum of squared resduals of the regresson n order to get a sequence {SSR Y T } T 2007:Q1. The most lkely canddate for the break s the date that mnmzes the sequence, hence maxmzng the ft of Y the model: T break = argmn T 2007:Q1 {SSR Y T }. Y Once we detect the endogenous date for the break (T break ), we re-estmate the baselne model wth the approprate break dummy and use a Wald test on κ and γ to determne whether the break s statstcally sgnfcant. The vector β captures the senstvtes 4 As there are multple shadow polcy rates avalable n the lterature, we perform extensve robustness checks usng alternatve ndcators of U.S. monetary polcy. The man fndngs are robust to alternatve proxes. 5 The Chnn-Ito ndex s only avalable at an annual frequency. We have tested the robustness of the results by usng a quarterly lnear nterpolaton of the Chnn-Ito ndex and by elmnatng the ndex from the regressons. In both cases, the man results of the study reman qualtatvely smlar. 7

10 of nternatonal fnancal flows to the drvers n X t before the break. The sum β + γ captures the post-break senstvtes. Gven our specal nterest n the senstvtes of nternatonal loan and bond flows to US monetary polcy and global rsk condtons, we then conduct an addtonal closer nvestgaton of the evoluton of the respectve estmated coeffcents. In partcular, we examne the hypothess that the post-crss paths of the above senstvtes may have been altered before and after the 2013 taper tantrum. For ths purpose we sequentally estmate equaton (2) wth the approprate break date, startng wth the sample 2000:Q1 2013:Q1 and addng one quarter at tme untl we reach our full sample (2000:Q1 2015:Q4). Ths procedure generates a dstnct set of parameter estmates for each sample-end quarter from 2013:Q1 through 2015:Q4. Ths allows us to track how senstvtes to US monetary polcy and global rsk condtons have evolved durng that perod. As wth the baselne analyss, for ths approach to tme varaton Y t can be cross-border loans to all sectors, to banks, to nonbanks, can be nternatonal debt securtes ssued by all sectors, by banks or by non-banks. 3.2 Decomposng the post-crss shfts n senstvtes The global lqudty seres exploraton descrbed n secton 3.1 utlzes gross flows data from the perspectve of borrowers n countres ndexed by. As the specfcatons ntroduce controls for local drvers of lqudty, the evoluton of estmated global factor coeffcents β 1 and β 2, on advanced economy monetary polcy and rsk, are assocated wth credtors. In partcular, changes n estmated β 1 and β 2 are attrbutable to a combnaton of shfts n the composton of nternatonal credtors (a compostonal component) and shfts n the senstvty of flows from country credtors vs-a-vs advanced economy monetary polcy and rsk metrcs (a behavoural component). For any class of credtor and borrower type, the aggregate senstvtes of nternatonal bank lendng flows to global factors ( β 1 and β 2 ) can be expressed as weghted averages of the natonal credtor-specfc senstvtes to global factors (β 1 and β 2 ). Whle ths observaton s general, as we have and analyse detaled nformaton from the perspectve of credtor bankng systems at the country level (but do not have credtor data for nternatonal debt securtes fnancng, our dervaton of the decomposton takes the perspectve of nternatonal bank lendng. S t S t 1 We start by re-wrtng specfcaton (1) as: 1 = β 1 ΔFFR t + β 2 logvix t + β 3 ΔlogGlobalGDP t + β 4 ΔSovRatng t + β 5 ChnnIto t + β 6 ΔlogGDP t + μ + ε t (3) where S t s the outstandng stock of nternatonal bank lendng to the resdents of country at the end of perod t. The natonal bankng system-specfc counterpart to specfcaton (1) s then wrtten as: 8

11 S t,, S t 1 1 = β 1 ΔFFR t + β 2 logvix t + β 3 ΔlogGlobalGDP t + β 4 ΔSovRatng t + β 5 ChnnIto t + β 6 ΔlogGDP t + μ, + ε t, (4) Expandng and smplfyng yelds: S t S t 1 1 =, S t. S t 1 1 = ( S, t, S t 1 ), S t 1 S t 1, 1 = {( S, t, S t 1 1) w, t 1 } (5), where the weght for each bankng system w t 1, S = t 1, S t 1 equals the respectve share of the outstandng stock of flows for whch t accounts. Combnng (4) and (5), the baselne regresson specfcaton mples that the senstvtes to the federal funds rate (β 1 ) and to the VIX (β 2 ) can be expressed as weghted averages of the respectve senstvtes (β 1 ) and (β 2 ) for the ndvdual lendng natonal bankng systems: 6, β 1 = {w 1,t 1 β 1 } and β 2 = {w 2,t 1 β 2 }. (6) The compostonal component s captured by the w 1 s and w 2 s and the behavoral component s captured by the β 1 s and the β 2 s. The compostonal factors w 1 s and w 2 s are drectly observable and can be obtaned from the data on blateral nternatonal clams. Meanwhle, the behavoral factors β 1 s and β 2 s are estmated usng a varant of the baselne specfcaton. Usng ths approach, we pvot from the borrowng country perspectve taken n secton 3.1 to nstead usng data from the credtor country perspectve. The BIS consoldated bankng statstcs (CBS) s a dataset from the credtor country perspectve that contans nformaton on banks nternatonal clams defned as the sum of cross-border clams and local clams denomnated n foregn currences. The data s blateral and contans nformaton on the natonalty of the lendng banks and on the resdence of the borrower. By lendng country and for estmaton perods correspondng to those defned n our frst stage of the analyss (pre-break perod and post break perods), the baselne model s estmated smlarly to model (1), to generate lendng country specfc estmates of behavoural factors β 1 s and β 2 s. Thus, we observe the credtor country hstory of changes n senstvtes and the precson of estmates of those senstvtes for global lqudty flows through nternatonal banks to both bank and non-bank counterpartes., 3.3 Identfyng the determnants of the post-crss behavoural and compostonal changes The thrd man emprcal element of our analyss s a further nvestgaton nto the changng drvers of global lqudty. We conduct a dff-n-dff analyss that compares the pre- and postcrss senstvtes to global factors and the shares of natonal bankng system lenders. The 6 A detaled explanaton of the decomposton of post-crss shfts n senstvtes s reported n Annex A. 9

12 analyss consders whch pre-crss characterstcs of bankng systems and polces are assocated wth changes to outcomes post-crss. Among the potental drvers of the structural changes n credtor country shares n global lqudty and credtor behavours are bank balance sheet condtons and regulatory polces. The bank condtons nclude the ntal (pre-crss) level of bank captalzaton and other ex-ante balance sheet characterstcs. Bank captal acts as a buffer aganst contngences trggered by monetary polcy shocks and can lmt the effect of a credt crunch n a crss characterzed by ncreased global uncertanty and volatlty (Gambacorta and Shn, 2016). We construct ex ante balance sheet characterstcs at the natonal bankng system level by usng bank-level Bankscope data, aggregated at the country level and correspondng to averages over dates ncluded n specfc estmaton perods. We test whether the shfts n senstvtes and weghts have been drven by the use of specfc prudental polcy tools usng the IBRN Prudental Instruments Database from the perspectve of lendng countres as descrbed n Cerutt, Correa, Forentno and Segalla (2017). Ths database covers prudental nstruments such as captal requrements, loan-to-value lmts, local currency reserve requrements, and nterbank exposure lmts. We test for the man drvers of the shft n senstvtes to global factors by estmatng regressons n whch the changes n the estmated coeffcents are regressed on a set of precrss varables. In partcular, we estmate the followng regressons: (β 1,k,PostBreak (β 2,k,PostBreak β 1,k,PreBreak ) = γ 1 F ζ 1 P θ 1,k + ε 1,k (7) β 2,k,PreBreak ) = γ 2 F ζ 2 P θ 2,k + ε 2,k (8) where (β 1,k,PostBreak (β 2,k,PostBreak β 1,k,PreBreak ) s the dfference n coeffcents for ΔFFR and ) s the dfference n the coeffcents for logvix taken from β 2,k,PreBreak equaton (4), estmated for lendng country and borrowng sector k (banks, non-bank prvate sector and publc sector). θ 1,k and θ 2,k are vectors of borrowng sector fxed effects. The vector F 2008 ncludes two bankng system ndcators: ) the captal-to-asset rato; ) the average bank sze. The vector P 2008 represents the prudental stance and t ncludes a range of prudental nstruments. We use pre-break characterstcs n order to lmt endogenety ssues. Snce the dependent varable n those regressons s a functon of estmated coeffcents, each wth an assocated standard error around t, we use meta-regressons technques 7. We lkewse examne the drvers of the shfts n lendng bankng system weghts n flows to bank and non-bank borrowers, applyng a smlar regresson specfcaton to (w k,postbreak w k,prebreak ): w k,postbreak w k,prebreak = γ w F ζ w P χ w B θ w,k + ε w,k (9) 7 The meta-regresson allows for resdual statstcal heterogenety n the results of dfferent estmaton (between -study varance) by assumng that the true effects follow a normal dstrbuton around the lnear predctor (Stanley and Jarrell, 1989). The metaregresson can be formally defned as: y θ ~N(θ, σ 2 ), where θ ~ N(x β,τ 2 ) therefore: y ~N(x β, σ 2 + τ 2 ), where β s the vector of estmated effects of study characterstcs. Ths type of equaton s estmated by weghted least-squares, n whch the weght of each estmated coeffcent depends nversely of ts varance and corresponds to the nverse of the sum of two standard devatons (σ 2, τ 2 ). 10

13 where B 2008 s a vector contanng the followng addtonal pre-break bankng system ndcators: ) the depost-to-total fundng rato, ) the rato of net nterest ncome to total ncome, ) the rato of local clams to foregn clams. 3.4 Examnng the drvers of the evoluton of post-crss senstvtes The fnal part of our emprcal analyss takes a tme seres approach n order to dentfy the man drvers of the evoluton of post-crss senstvtes. We conecture that ths evoluton may be nfluenced by the overall advanced economy monetary polcy stance and by the characterstcs of the credtor natonal bankng systems. One possble drver of the post-crss evoluton n senstvtes could be the degree of monetary polcy convergence among advanced economes (AEs). The reacton to U.S. monetary polcy as a global lqudty drver could be especally pronounced f t s a sgnal for a broader based set of expansonary polces across AE countres. In the perod between the global fnancal crss and the 2013 Fed taper tantrum, there was consderable convergence between the monetary polces of advanced economes, all of whch were conductng varous forms of quanttatve easng to stmulate the real economy. In 2013 the Federal Reserve sgnalled that t would start taperng ts bond buyng program. As the central banks of other advanced economes, most notably the European Central Bank and the Bank of Japan, dd not follow sut, the monetary polces of advanced economes dverged through the end of our estmaton perod n Thus, we conecture that the senstvtes of the man global lqudty components to of US monetary polcy could be stronger durng the convergence perod and weaker as polcy dverges. We also conecture that bankng system characterstcs, such as lenders domnant busness models and proftablty, may have also drven the post-crss evoluton n senstvtes. Insttutons engagng manly n commercal bankng actvtes have lower costs and more stable profts than those more heavly nvolved n captal market actvtes, manly tradng. Also, retal bankng has ganed ground post-crss, reversng a pre-crss trend (Roengptya et al., 2014). 8 The wllngness to lend to rsker counterpartes s partcularly strong n low nterest rate envronments, especally when these are lkely to be sustaned. Thus, we conecture that reach for yeld behavours, a push factor n global lqudty, may be stronger for the bankng systems that had more depressed proftablty and return on assets. Ths last secton of emprcal results thus examnes the relevance of lendng bankng system characterstcs and the degree of dvergence among advanced economes monetary polces by nteractng the respectve varables wth the coeffcents of ΔFFR and logvix n equaton (2). The resultng model s: GrRate Y t = β X t + [ν + η ( ΔFFR logvix ) ] poldv t + [ρ + ξ ( ΔFFR logvix ) ] Prof t + μ t t Y +I(t T break ) {κ + γ X t + [ω + χ ( ΔFFR logvix ) ] poldv t + [δ + ψ ( ΔFFR logvix ) ] Prof t } + ε t t t (10) 8 The lnk between busness models and lendng s developed, among others, n Lamers et al. (2016) and Martnez-Mera and Repullo (2015). 11

14 where poldv t s a proxy for the monetary polcy dvergence between the US and other advanced economes; Prof t s a weghted average of a proxy for the busness models of bankng systems lendng to country. The weghts are gven by equaton (5). The tmevaryng, borrowng-country-specfc post-crss senstvty to ΔFFR s captured by β 1 + η 1 poldv t + ξ 1 Prof t + γ1 + χ 1 poldv t + ψ 1 Prof t. 4. Data We utlze three databases to capture the man components of global lqudty: the BIS Locatonal Bankng Statstcs (LBS), the BIS Internatonal Debt Securtes Statstcs (IDSS), and BIS Consoldated Bankng Statstcs (CBS). The BIS LBS captures the outstandng clams and labltes of nternatonally actve banks located n 44 BIS LBS reportng countres 9 aganst counterpartes resdng n more than 200 countres. Banks record ther postons on an unconsoldated bass, ncludng ntragroup postons to capture nternatonal flows between offces of the same bankng group. The data are compled followng prncples that are consstent wth balance of payments statstcs. The LBS statstcs capture around 95% of all cross-border nterbank busness (Bank for Internatonal Settlements, 2015). At the same tme, the counterparty sector breakdown avalable n the BIS LBS enables us also to dstngush between cross-border bank lendng to bank and non-bank borrowers. These data seres capture the nternatonal flows to bank and non-bank borrowers from the borrower perspectve. The BIS CBS are manly used n our analyss credtor for the bank perspectve. We use them to compute lender-specfc senstvtes to the global factors, as well as the relatve mportance of lendng countres for a gven borrowng country n terms of captal flows. The BIS IDSS data capture borrowng n money and bond markets. Internatonal debt securtes (IDS) are defned as those ssued n a market other than the local market of the country where the borrower resdes (Gruć and Wooldrdge, 2012). They encompass what market partcpants have tradtonally referred to as foregn bonds and eurobonds. The sample used for the emprcal analyss conssts of quarterly data from Q to Q On the borrowng sde, we focus on a set of 64 countres, whch ncludes both, Advanced Economes (AEs) and Emergng Market Economes (EMEs). On the bank lendng sde, we use data on the postons of all 44 BIS LBS and 31 CBS reportng countres. 10 The typcal lenders and borrowers connected by each flow type dffer consderably n composton and sze, as llustrated wthn Table 1. Cross-border loans are typcally suppled by nternatonally-actve banks, whch tend to be relatvely large. Meanwhle, the credtors n nternatonal debt securtes markets are usually non-bank fnancal ntermedares, such as penson funds, nsurance companes, money market mutual funds, and hedge funds. The varaton on the borrower sde s even greater. Internatonal bond ssuance by non-banks tends to be domnated by soveregns and large non-fnancal corporates. The latter are also mportant players on the borrowng sde of the cross-border bank loan market, whch also channels funds to export/mport frms and leveraged non-bank fnancals. 9 The complete lst of BIS LBS reportng countres s provded at 10 The complete lsts of all borrowng countres and lendng natonal bankng systems are avalable n Annex B. 12

15 There are three global factors n our analyss. Global real GDP growth measures global economc actvty. The second global factor s changes n stance of US monetary polcy. The emprcal lterature dscussed n Secton 2 manly corresponds to the perod pror to the ntroducton of unconventonal monetary polcy and exclusvely uses a short-term polcy rate. However, ths approach may not be approprate for the full post-crss perod. Monetary polcy at the zero lower bound s a defnng feature of the post-crss perod, and changes n communcatons, nterest on effect reserves and quantty easng actons became more nstrumental. Accordngly, there s extensve debate on whether a sngle metrc can convncngly serve as a suffcent statstc for changes n U.S. monetary condtons. For our purposes, we use the Wu-Xa polcy measure (Wu and Xa, 2016) as a baselne for the US FFR. By ths construct, the effectve US Federal Funds target rate s used pror to Q and the Wu-Xa estmates of the shadow Federal Funds rate are used from Q through end of 2015 (Graph 2, left panel). Whle a number of alternatve shadow rates have been constructed, all shadow rates are senstve to the underlyng modellng assumptons utlzed. Changes n shadow polcy rates have been documented to be hghly correlated across a number of the alternatve shadow rate seres. We conduct robustness analyss (secton 6) usng alternatves, showng the robustness of our results. The thrd global factor s a measure of global rsk condtons. Followng the global lqudty lterature, we utlze the VIX ndex of the mpled volatlty n S&P500 stock ndex opton prces from Chcago Board Optons Exchange (CBOE) (Graph 2, rght panel). Three borrowng country varables (pull factors) are ncluded n baselne specfcatons: local real GDP growth, soveregn ratngs, and the degree of fnancal openness. For each borrowng country, the soveregn ratngs varable s defned as the average ratngs across the three maor credt ratngs agences (S&P, Moody s and Ftch). The degree of fnancal openness s captured by the Chnn-Ito ndex (Chnn and Ito, 2008), normalzed between 0 and 1. The IBRN Prudental Instruments dataset covers wdely-used prudental nstruments, keepng track of the ntensty of ther usage n 64 countres between 2000 and 2014 at a quarterly frequency. The nstruments that are covered are: general captal requrements, sector-specfc captal requrements (splt nto real estate credt, consumer credt, and other), nterbank exposure lmts, concentraton lmts, loan-to-value (LTV) rato lmts, and (local currency and foregn currency) reserve requrements. We focus on the three prudental polcy nstruments that have been shown to have the largest mpact on nternatonal bank lendng: loan-to-value rato caps, captal requrements and local currency reserve requrements (Avdev et al., 2017; and Buch and Goldberg, 2017). 11 The balance sheet characterstcs of natonal bankng systems are constructed usng Bankscope data. We obtan the balance sheet tems of nterest for the set of nternatonally actve banks that report to the BIS consoldated bankng statstcs, and then aggregate banklevel characterstcs to natonal bankng system-wde varables, usng weghted averages across the ndvdual banks of a gven natonalty. Data are adusted for mergers and acqustons to correct for balance sheet umps that are unrelated to lendng (Bre et al., 11 Cerutt et al. (2017) provde an extensve dscusson of the propertes of the quarterly changes n these prudental nstruments and the cumulatve changes over tme. 13

16 2013). We gather data on ) captal to total assets, ) average bank sze, ) deposts to total assets, v) net nterest ncome over total ncome, v) net nterest ncome to total assets. Two bank busness model measures are consdered: ) an ncome dversfcaton rato (defned as net nterest ncome to total ncome); ) net nterest ncome to total assets. The frst ndcator ranges from 0 to 1 and ndcates the fracton of a bank s proftablty that derves from tradtonal ntermedaton actvty (.e. lendng and deposts). If a bank has a large porton of non-nterest ncome (tradng ncome, fees and commssons for servces) than ths ndcator tends lower values. The second ndcator s the return per unt of assets that derves from tradtonal ntermedaton actvty. It represents the proftablty of ntermedated assets that s obtaned by the bank gettng deposts and supplyng loans. As a proxy for monetary polcy dvergence among advanced economes, we take the dfference between the 2-year futures on the polcy rate for the Unted States and the average of the 2-year futures for the Unted Kngdom, Swtzerland, Japan and a group of core Eurozone countres (Austra, Belgum, Germany, Fnland, France, the Netherlands, Span) Evdence on global lqudty drvers 5.1 Baselne results Our emprcal nvestgaton begns wth the baselne specfcaton n equaton (1) as a way to replcate pror global lqudty results. The estmated coeffcents for the entre sample 2000:Q1 2015:Q4, presented n Table 2, are largely n lne wth those obtaned n the exstng lterature. Usng data on nternatonal bank flows and nternatonal debt securtes ssuance from the bank and non-bank debtor perspectve, the results from the baselne model ndcate that an ncrease n global rsk condtons (measured by the VIX) has a negatve and strongly statstcally sgnfcant effect on all flows we examne. The US federal funds rate has a sharply negatve mpact on cross-border bank loans. Its estmated mpact on nternatonal debt securtes s also negatve, albet only margnally statstcally sgnfcant. Local factors are also statstcally sgnfcant drvers. Borrowng countres wth h gher GDP growth rates and wth better soveregn credt ratngs tend to attract more cross-border loans. Meanwhle, the degree of fnancal openness, as reflected n the Chnn-Ito ndex, has a postve (and statstcally sgnfcant) effect on the nternatonal bond flows, especally to banks. As descrbed n Secton 3.1. we formally test whether the above estmated coeffcents from equaton (1) are stable over tme. Rather than exogenously mposng an ad-hoc break date, we test for ts presence and exact tmng endogenously. We fnd that the most lkely break date for both cross-border loan flows and nternatonal bond flows s 2009:Q1. Wald (or Chow) tests on the coeffcents κ and γ n equaton (2) ndcate that the break s statstcally sgnfcant for the global lqudty components that we examne Summary statstcs for the explanatory varables used n our emprcal analyss are presented n Table C1 n Annex C. 13 Test results are avalable upon request. 14

17 Table 3 summarzes the estmated senstvtes to the man global drvers (the VIX and the federal funds rate) durng the pre-break and the post-break perods, respectvely. Two sets of estmates are provded for the post-crss perod one for the full sample (endng n Q4:2015) and one for a sub-sample endng n Q1:2013. The latter set of results allows us to examne whether the 2013 Fed taper tantrum marked a turnng pont n the post-crss senstvtes to global factors. The results confrm that the relatonshp between the man global factors and nternatonal captal flows has changed profoundly snce the Global Fnancal Crss. Senstvtes of all flow types to US monetary polcy ncreased sharply between the GFC and the taper tantrum. Ths s true for all flow types we examne and for all borrowng sectors. The mpact of US monetary polcy on cross-border loans, whch was already negatve and statstcally sgnfcant durng the pre-crss perod, rose even further n the mmedate aftermath of the GFC. More concretely, whle pror to the crss a 25-bass pont declne n the federal funds rate was assocated wth an 80-bass pont rse n the quarterly growth rate of cross-border bank lendng, n the aftermath of the crss was assocated wth a 202-bass pont ncrease n the same growth rate. The respectve negatve mpact on nternatonal bond ssuance, whch was not statstcally sgnfcant pror to the crss, also ncreased consderably after the GFC. In quanttatve terms, the mpact of a 25-bass pont drop n the federal funds rate on nternatonal bond ssuance surged from 36 bass ponts before the crss to 202 bass ponts after the crss. Followng the 2013 taper tantrum, senstvtes to US monetary polcy reverted towards ther respectve pre-crss levels, especally for cross-border loans. Once the sample s extended to nclude the post-taper tantrum perod, a 25-bass pont declne n the federal funds rate becomes assocated wth a 92-bass pont ncrease n the quarterly growth rate of cross-border bank loans, a level of responsveness much closer to the one observed pror to the crss. The responsveness of nternatonal bond flows also reverted, but not by as much. Nonetheless, all the mpact of changes n US monetary polcy on all flow types remaned negatve and hghly statstcally sgnfcant. The post-crss perod also s characterzed by a substantal declne n the senstvty of cross-border bank loans to global rsk condtons. The estmated effect of the VIX, whch was negatve and hghly statstcally sgnfcant pror to the crss, weakened consderably by Q1:2013 and even became nsgnfcant by the end of the sample (n Q4:2015). Whereas pror to the crss a one-standard devaton ncrease n the VIX was assocated wth a 134-bass pont contracton n the quarterly growth rate of cross-border bank lendng, after the crss t only caused an 11-bass pont declne n the same growth rate. The post-crss evoluton of the senstvtes of aggregate global lqudty flows (.e. the sum of nternatonal bank loans and bond flows) was n lne wth the respectve evolutons for the man global lqudty components. Namely, the responsveness of aggregate flows to US monetary polcy rose sharply between the GFC and the 2013 taper tantrum. It subsequently reverted towards pre-crss levels, but remaned at relatvely hgh levels. By contrast, aggregate global lqudty flows became much less senstve to global rsk snce the GFC. There has been some convergence n the global factor senstvtes of the two man global lqudty components. Table 4 shows the dfference between the senstvtes of cross- 15