R O M E Research On Money in the Economy

Transcription

1 R O M E Research On Money in he Economy No. 9-7 Ocober 29 The Imporance of Global Shocks for Naional Policymakers Rising Challenges for Cenral Banks Ansgar Belke and Andreas Rees ROME Discussion Paper Series Research on Money in he Economy (ROME) is a privae non-profi-oriened research nework of and for economiss, who generally are ineresed in moneary economics and especially are ineresed in he inerdependences beween he financial secor and he real economy. Furher informaion is available on ISSN

2 Research On Money in he Economy Discussion Paper Series ISSN No 29-5, Ocober 29 The Imporance of Global Shocks for Naional Policymakers Rising Challenges for Cenral Banks Ansgar Belke and Andreas Rees Prof. Dr. Ansgar Belke Dipl.-Vw. Andreas Rees Universiy of Duisburg-Essen HypoVereinsbank UniCredi Group Deparmen of Economics Chief Economis Universiaessr. 12 Kardinal-Faulhaber-Sraße 1 D Essen D-8333 München ansgar.belke@uni-due.de jens.klose@wiwinf.uni-due.de and Deusches Insiu für Wirschafsforschung (DIW) Mohrensraße 58 D-1117 Berlin and Insiue for he Sudy of Labor (IZA) Bonn Schaumburg-Lippe-Sr. 5 9 D Bonn The discussion paper represen he auhors personal opinions and do no necessarily reflec he views of DIW Berlin, he Insiue for he Sudy of Labor (IZA) Bonn and UniCredi Munich. NOTE: Working papers in he Research On Money in he Economy Discussion Paper Series are preliminary maerials circulaed o simulae discussion and criical commen. The analysis and conclusions se forh are hose of he auhor(s) and do no indicae concurrence by oher members of he research nework ROME. Any reproducion, publicaion and reprin in he form of a differen publicaion, wheher prined or produced elecronically, in whole or in par, is permied only wih he explici wrien auhorisaion of he auhor(s). References in publicaions o ROME Discussion Papers (oher han an acknowledgmen ha he wrier has had access o unpublished maerial) should be cleared wih he auhor(s) o proec he enaive characer of hese papers. As a general rule, ROME Discussion Papers are no ranslaed and are usually only available in he original language used by he conribuor(s). ROME Discussion Papers are published in PDF forma a Please direc any enquiries o he curren ROME coordinaor Prof. Dr. Albrech F. Michler, Heinrich-Heine-Universiy of Duesseldorf, Deparmen of Economics, Universiaessr. 1, Build , D-4225 Duesseldorf, Germany Tel.: ++49() Fax: ++49() helpdesk@rome-ne.org michler@uni-duesseldorf.de

3 Absrac We analyze he imporance of global shocks for he global economy and naional policy makers. More specifically, we invesigae wheher moneary policy has become less effecive in he wake of financial globalizaion. We also examine wheher here is increasing uncerainy for cenral banks due o globalizaion-driven changes in he naional economic srucure. A FAVAR framework is applied o derive srucural shocks on a worldwide level and heir impac on oher global and also naional variables. We esimae our macro model using quarerly daa from Q o Q4 27 for he G7 counries plus he euro area. According o our resuls, global liquidiy shocks are a driving force of he global economy and various naional economies. However, some oher shocks originaing in house prices, GDP, echnology and long-erm ineres rae shocks play a role a he global level as well. These resuls prove o be robus across differen specificaions. Srucural break ess indicae ha global liquidiy shocks have recenly become more imporan as a deerminan for house prices. In general, global variables have become more powerful over ime in driving naional variables. JEL-Classificaion: C22, E31, E32, F42 Keywords: Global shocks, inernaional business cycle; inernaional policy coordinaion and ransmission, facor augmened vecor auoregressive (FAVAR) models, common facors. Acknowledgemens: We are graeful for valuable commens from paricipans in he Meeing of Naional Economic Research Organizaions (NERO), Sepember 21, 29, OECD Headquarers, Paris. Corresponding auhor: Professor Dr. Ansgar Belke, Universiy of Duisburg-Essen, Deparmen of Economics, Essen, DIW Berlin and IZA Bonn phone: +49 ()21/ , fax: +49 ()21/ , ansgar.belke@uni-due.de

4 -1-1. Inroducion Global economic inegraion has been spreading markedly in recen years. This is boh rue for goods and financial markes. Macroeconomic variables in one counry should herefore increasingly reflec evens occurring in he res of he world. Even he housing secor, which is usually regarded as a naional phenomenon, has seemingly become more synchronized across counries. As indicaed by he laes evens, srong rises in residenial propery prices in he US and some pars of Europe were followed by rapid declines. As naional economies become more inerconneced, a horough undersanding of he global economy and is effecs on domesic economic aciviy is crucial. The abiliy o gauge he iming and he magniude of inernaional spillovers is of paricular relevance, since i conribues o a beer assessmen of he developmen in one counry or region. The rapid speed of globalizaion on goods and financial markes is beneficial, bu may also have drawbacks for naional policymakers. Inernaional spillovers and global shocks can limi he auonomy of naional moneary and fiscal policy. For example, inernaional capial flows are influencing naional moneary condiions, hereby curailing he abiliy of cenral banks o influence naional real aciviy and prices. The quesions we are invesigaing in his conribuion are herefore hreefold. Firs, wha are he major shocks and ransmission channels which are driving he global economy? Second, o wha exen have global facors affeced he deerminaion of key macroeconomic variables in he G-7 counries? We quanify he speed and size of spillovers ha occur following a shock originaing from he global economy. Third, wha can naional economic policy do in he ligh of inernaional spillovers and wha should naional policymakers do? More specifically, we invesigae wheher here is increasing uncerainy for moneary policy in he wake of globalizaion and wheher here is a negaive ime rend in he effeciveness of naional moneary policy when rying o seer naional liquidiy. The paper proceeds as follows. In secion 2, we relae our conribuion o he lieraure and develop he global perspecive before we urn o he selecion of he daa and variables in secion 3. In secion 4, we briefly explain he Facor-Augmened Vecor Auoregression (FAVAR) mehodology. In secion 5, we display our esimaion resuls

5 -2- and he resuls of some ess for srucural breaks and some robusness checks. Secion 6 finally concludes wih some policy recommendaions. 2. Going global global variables and a global perspecive on shocks In his paper, we invesigae he co-movemens among some macro variables across he G7 and he euro-area counries wih he aim o uncover he common driving forces shaping inernaional macroeconomic dynamics and he feaures of heir ransmission mechanisms. For his purpose, we make use of a modified version of he Sock and Wason (25a) Facor-Augmened Vecor Auoregression (FAVAR) model. Our approach allows a more sraighforward economic inerpreaion of he unobservable global facors. Our work is, on he one hand, relaed o he lieraure on global VARs (GVARs) and, on he oher hand, o he research done on inernaional business cycle comovemen. 2.1 FAVARs, GVARs and inernaional business cycle co-movemen Le us firs delineae he noion of a SFAVAR agains ha of a GVAR model. A GVAR model is a compac model of he world economy designed o explicily esimae he economic and financial inerdependencies a naional and inernaional levels. Individual counry/region specific vecor error-correcing models are esimaed, where he domesic variables are relaed o corresponding foreign variables consruced exclusively o mach he inernaional rade paern of he counry under consideraion. The individual counry models are hen linked in a consisen manner so ha he GVAR model is solved for he world as a whole. The degree of regional inerdependence is invesigaed via generalized impulse response funcions ha porray he effecs of shocks o a given variable in a given counry/marke on he res of he world (Pesaran, Schuermann and Weiner, 24, and Dees, di Mauro, Pesaran and Smih, 27). In our FAVAR, he economeric approach is less complicaed and perhaps more sraighforward. All global variables are modeled as endogenous in a srucural VAR conex. Spillover effecs from global o naional variables are possible, since here is a direc link beween he global and he naional level via he facor loadings. Global forces are regarded as exogenous o

6 -3- domesic variables wih no feedback effecs possible. 1 Finally, in conras o Pesaran e al. (24), we choose he weighing in he consrucion of he common facors by using principal componens analysis. Anoher series of papers has empirically checked he exisence and significance of common paerns in he inernaional dynamics of macro variables. The focus has been on he changes over ime of business cycle synchronizaion across he mos imporan economies originaing in common global disurbances. Assessing inernaional business cycle co-movemen is mainly an empirical ask and he main drivers of he developmen may shif over ime. Usually, his lieraure has aken ino accoun only a limied se of real quaniies such as oupu, consumpion and invesmen, which is, however, someimes researched for a large number of counries (Sock and Wason 25). When a broader range of variables is included in he analysis, he focus of he lieraure ends o swich from he common driving forces of flucuaions o he spillover of shocks. Many auhors have invesigaed hese issues. 2 Mos of hem deeced a endency for naional business cycles o converge over he period of he second globalizaion. Aris and Okubo (28) provide a long-run hisorical perspecive which, by revisiing he era of he firs globalizaion before he Firs World War, demonsraes a endency for globalizaion o produce a high degree of synchronizaion in naional business cycles. Sock and Wason (25) conclude from heir analysis ha co-movemen has fallen during he period relaive o due o he absence of common shocks. In his paper, we adop a wider perspecive and sudy co-movemens among he G7-counries plus he euro area, using a larger daa se han previously employed in he lieraure, including boh real and nominal variables. In he seminal conribuion of Kose, Orok and Prasad (28), more han 1 counries are analyzed over he period. Flucuaions in economic aciviy are decomposed ino global, counry group and counry specific facors. During he second period of globalizaion ( ), business cycles have converged among he group of 1 No oo differen from our approach bu in conras o Bagliano and Morana (29) who model all variables as endogenous from he ouse, Dees, di Mauro, Pesaran and Smih (27) model each counry separaely, wih foreign variables reaed as weakly exogenous. 2 See for example Aris and Zhang (1997), Kose, Orok and Whieman (23), Aris, Krolzig and Toro (24), and Canova, Ciccarelli and Orega (26).

7 -4- advanced economies and among he group of emerging marke economies. Bu a he same ime, he relaive imporance of he global facor has declined. Hence, here is evidence of cyclical convergence wihin each group, bu for decoupling beween hem. Overall, here has been lile change in he degree of inernaional synchronizaion as measured by he join conribuion of he global and group-specific facors. However, his feaure is quie consisen wih an increased imporance of common shocks as a driving force of inernaional oupu flucuaions: he smaller magniude of he shocks occurred since he early 198s can explain a broadly consan paern of correlaions among GDP growh raes across counries (Bagliano and Morana, 29, p. 432). In conras o our sudy which imposes some srucure on he global and naional level, Kose, Orok and Prasad (28) do no make use of any srucural model. Moreover, hey confine hemselves o he use of facor analysis. In our sudy, we move one sep furher by using FAVAR analysis, i.e. by inegraing facor analysis ino a VAR framework. As a sylized fac from he lieraure, common componens appear o play a larger role in business cycles in hose advanced economies which are he focus of our paper. In conras, counry specific facors are relaively more imporan for emerging marke economies (Kose, Orok and Whieman, 23). One reason for his resul migh be ha many emerging marke economies in conras o he indusrialized counries have only reached inermediae levels of financial inegraion, i.e. hey have no been able o achieve improved risk sharing over he globalizaion period (Kose, Prasad and Terrones, 27). Our work is relaed o ha conduced by Bagliano and Morana (29). In heir paper, inernaional co-movemens among a se of key real and nominal macroeconomic variables in he US, UK, Canada, Japan and he euro area have been invesigaed for he period, using a facor vecor auoregressive approach. They deliver empirical evidence ha co-movemens in macroeconomic variables do no only concern real aciviy, bu are an imporan feaure also of sock marke reurns, inflaion raes, ineres raes and, o a smaller exen, moneary aggregaes. Boh common sources of shocks and similar ransmission mechanisms explain inernaional co-movemens, wih he only excepion of Japan, where he idiosyncraic feaures seem o dominae. Finally,

8 -5- concerning he origin of global shocks, evidence of boh global supply-side and demandside disurbances is found. However, our work differs from Bagliano and Morana (29) in hree major aspecs. Firs, we explicily analyze he srucural relaionship beween various global variables in order o ge a beer undersanding of he world economy. This par is missing in Bagliano and Morana (29). Insead, he auhors concenrae on four global facors which hey label inflaion facor, oupu growh facor, sock reurn facor and oil price facor. For example, he inflaion facor conains no only inflaion raes bu also ineres raes as well as moneary aggregaes. In our approach, he relaionships beween such common forces are explicily disenangled via a srucural model on a global level. We use a oal of seven global facors, including a house price facor. The laer enables us o invesigae o wha exen oher global variables conribued o he srong rise in propery prices unil he sar of he financial crisis. Second, we examine spillover effecs from global o naional variables using srucural VARs for he G-7 counries plus he euro area. In conras o Bagliano and Morana (29), we do no esimae one model wih all naional variables aken as endogenous bu implemen separae VARs for each counry or region. Hence, we inenionally neglec feedback effecs beween counries. Bagliano and Morana idenify srucural idiosyncraic shocks by imposing exclusion resricions on heir conemporaneous impac on all naional variables across counries, esimaing a oal of 43 parameers in each equaion (including he four facors). Our main moivaion o apply an approach differen from heirs is ha modeling boh global and all naional variables in one complee srucural framework would have been oo cosly in erms of degrees of freedom. Hence, he ypical empirical rade-off beween using a sufficien number of srucural facors on a global level and esablishing srucural links beween naional variables for various counries emerges once again. 3 3 We use seven facors wih a lag lengh of 2, hereby leading o he esimae of foureen coefficiens on he global level (plus a consan and a deerminisic componen). Our sample covers he period from Q o Q4 27. Bagliano and Morana base heir specificaion on four facors wih only one lag (sample Q1 198 Q2 25).

9 -6- Third, we are examining wheher srucural breaks emerge on a global level bu also in he relaion beween global and naional variables. In our view, his is of special ineres for naional policy-makers, as hey are supposed o reac in a imely and consisen manner when naional economies are hi by global shocks. 2.2 Global variables and a "heory" of global shocks Le us now urn o why we make use of specific global variables and o adhere o a heory of global shocks in he conex of our paper. In mos of he lieraure, here is a focus on commodiy prices and real GDP when defining global shocks. However, we would like o argue ha i can also make sense o hink in erms of oher macroeconomic and financial variables as global ones. More specifically, we invesigae wheher and o wha exen inernaional co-movemens concern no only real economic aciviy bu also nominal variables like, for example, moneary aggregaes and ineres raes and provide an economic inerpreaion for he sources of common dynamics. We feel legiimized o do so by keeping an eye on he main empirical paern of he mos recen financial and economic crisis which is commonly modeled as a global demand shock and has been characerized by a synchronous downurn of house prices (Federal Reserve Bank of Kansas Ciy, 29). Our main focus in his paper is on he concep of global moneary liquidiy and of global shor-erm ineres raes. Neverheless, we also find significan global house price shocks, echnology shocks and long-erm ineres rae shocks. Le us firs address global moneary liquidiy. I is usually argued ha here is a global money marke only under a fixed exchange rae sysem. By pegging he value of domesic currency o a foreign currency, cenral banks make foreign currency a perfec subsiue for domesic currency on he supply side. Should he moneary auhoriy in one counry increase he money supply, he domesic money supply would exceed domesic money demand. As a resul, money flows ou hrough he balance of paymens. The domesic balance-of-paymens defici mus be mached by a balance-of-paymens surplus abroad. Thus, money supplies abroad mus also increase. Flexible exchange raes are assumed o eliminae his source of moneary inerdependence. Naional money becomes a non-radable asse whose relaive price (he exchange rae) is assumed o be deermined freely in foreign exchange markes. In

10 -7- conras wih he fixed exchange rae sysem in which each counry s money supply was endogenous, i is deermined exogenously by moneary auhoriies. Under flexible raes, he balance of paymens is always zero, i.e. here is no ne money flow beween cenral banks. Flexible raes herefore make currencies perfec non-subsiues on he supply side. However, he insular propery of floaing exchange raes may break down when here is currency subsiuion. Some invesors migh consider domesic and foreign currencies as relaively close subsiues. Currency subsiuion suggess ha he demand for domesic money is dependen on exernal facors. If domesic residens hold porfolios conaining boh foreign and domesic asses and reallocae hese porfolios according o changes in he relaive opporuniy coss of hese asses, foreign moneary shocks will aler he relaive coss of holding a given porfolio. This in urn induces residens o reallocae heir porfolios beween domesic and foreign asses. The readjusmen of currency holdings enables moneary shocks o be ransmied via money demand from one economy o anoher even in a world of flexible exchange raes. In addiion, Rüffer and Sracca (26) argue ha apar from currency subsiuion money has o be characerized as endogenous. For his purpose, hey modify he sandard porfolio balance model by assuming ha he key rae of he cenral bank is exogenous and domesic money holding is endogenous, i.e. money-demand driven (and dependen on foreign ineres raes, as invesors hold foreign bonds). Hence, one can hink of direc spillover effec in moneary aggregaes from abroad if he naional money supply is endogenous, i.e. driven by money demand and no by money supply via cenral banks. Consider he case of an expansionary moneary policy in he foreign counry which we capure by a reducion in foreign ineres raes. As a firs consequence, foreign invesors shif ou of foreign bonds and ino foreign money due o he reduced opporuniy coss of moneary balances. As a second sep, he reducion in foreign ineres raes raises he relaive araciveness of domesic money and bonds for domesic agens. If he elasiciy of he demand for money wih respec o he foreign ineres rae is larger han he elasiciy of he demand for domesic bonds, hen he money holdings in he domesic counry increase.

11 -8- Moneary liquidiy spillovers across counries could have serious implicaions for cenral banks and naional macroeconomic and financial variables as well. For example, moneary liquidiy spillovers may lead o a global cycle in house prices. The same could be rue for a common paern in inflaion and share prices. According o Lane and Milesi-Ferrei (27), he dramaic increase in inernaional financial inegraion has been one of he salien global economic developmens in recen years. Counries have accumulaed subsanial cross-border holdings and here have been sizable shifs in he composiion of asse and liabiliy posiions. The size of counries exernal porfolios is now such ha flucuaions in exchange raes and asse prices cause very significan reallocaions of wealh across counries. This creaes a huge poenial for inernaional capial flows, hereby influencing moneary condiions in oher counries as well. We now briefly address our concep of a "global shor-erm ineres rae". Of course, here is no world cenral bank which ses he shor-erm ineres rae for many counries. However, from his quie rivial insigh, i canno be concluded ha ineres rae shocks canno be idenified on a global level. If a global cycle exiss in he world economy and in housing markes, he money marke raes should move synchronously as well, since naional moneary policies reac direcly or indirecly o global developmens: direcly, in he sense ha global variables ener he cenral bank s reacion funcion (via world GDP and/or global excess liquidiy); indirecly, if moneary policy does no reac unil global variables have an effec on naional paerns (global excess liquidiy spills over ino naional moneary aggregaes). Our "global ineres rae shocks" can hen be inerpreed as unexpeced changes in his reacion paern. In he lieraure, common beliefs and peer pressure are menioned as addiional reasons for a similar reacion paern of cenral banks, i.e. naional moneary policies reac quie similarly across borders via meeings and he exchange of informaion among cenral banks (implici policy coordinaion). For example, he esablishmen of inflaion argeing regimes in he majoriy of indusrialized counries migh serve as an example of "common beliefs" in his respec.

12 -9-3. Daa and variables In our FAVAR analysis, we use quarerly ime series from Q o Q4 27 for he G7-counries plus he euro area, i.e. he US, Japan, he UK and Canada and he EMU. Hence, 48.9% of global GDP in 27 is represened in our empirical analysis. 4 In principle, one could argue ha emerging markes have become increasingly imporan for he global economy and inernaional financial markes. However, we oped for a focus on major indusrialized counries for hree reasons. Firs, he majoriy of emerging markes have a fixed exchange rae regime which makes moneary spillover effec likely, according o he radiional rilemma view. The moivaion of our sudy is o examine wheher such moneary spillovers can also occur despie flexible exchange raes. Second, here are daa availabiliy problems even for bigger emerging markes. Third, as already menioned in secion 2, common componens appear o play a larger role in business cycles in advanced economies han in emerging marke economies, where counryspecific facors end o be more imporan. Why are global liquidiy shocks imporan in our conex and by far no arifac? Some criics migh argue ha global liquidiy, as measured in one currency, can only change in quaniaive erms if one assumes a fixed exchange rae sysem worldwide. Noe, however, ha inernaional liquidiy spillover effecs may occur regardless of he exchange rae sysem. Under pegged exchange rae regimes, official foreign exchange inervenions resul in a ransmission of moneary policy shocks from one counry o anoher. In a sysem of flexible exchange raes, he validiy of he "uncovered ineres rae pariy" (UIP) relaionship should in heory preven cross-border moneary spillover. According o his heory, he expeced appreciaion of he low-yielding currency in erms of he high-yielding currency should be equal o he difference beween (risk-adjused) ineres raes in he wo economies. However, he violaion of he UIP - ofen referred o as he forward premium puzzle - is a common empirical finding in he lieraure on macroeconomics and finance. The enduring exisence of carry rades can be aken as evidence ha exchange raes diverge from fundamenals for lenghy periods, as he exposure of a carry rade posiion 4 Own calculaions based on IMF PPP daa.

13 -1- involves a be ha UIP does no hold over he invesmen period. Moreover, currency subsiuion may enable inernaional liquidiy spillovers in a framework of flexible exchange raes. These inernaional adjusmens of money holdings allow he ransmission of moneary shocks from one economy o anoher (via money demand) even in a sysem of flexible exchange raes. The ime series for he G7 plus he euro zone are drawn from a variey of sources, including various naional saisical offices, cenral banks and he OECD. Mos ime series are provided by professional daabanks like Thomson Daasream, and Bloomberg. Since EMU series a a quarerly frequency are ofen available only for a relaively shor ime-span, we parly rely on he Area-Wide Model daabase by Fagan, Henry and Mesre (21) who provide backdaed ime series. Hence, some cauion is warraned, as here are mehodological differences across euro-zone counries in collecing he daa. Moreover, daa availabiliy on a naional level becomes increasingly scarce when moving back in ime. Hisorical house prices for he EMU sem from Gros (27). We esimae our baseline FAVAR model including he following variables for he G7-counries plus he non G7-member euro-area counries: real GDP, four inflaion measures (CPI, PPI, impor prices and GDP deflaor), 3-monh shor-erm ineres raes, broad moneary aggregaes (ypically, M2 or M3), wo commodiy price indices (HWWI and CRB), house prices and share prices (naional MSCI). A complee daa lis wih sources can be found in Table A7 in he appendix. 4. The FAVAR mehodology The empirical value added from our approach sems from he use of a Facor Augmened VAR (FAVAR). A FAVAR is he combinaion of a sandard VAR model wih facor analysis. Global variables were derived by using facor analysis. We regard his procedure as superior compared o a simple aggregaion of naional variables for wo reasons. Simple sum aggregaion implicily assumes ha he included naional variables are perfec subsiues. However, given differences in naional measuremen, his does no need o be rue. For example, he moneary aggregae M2 in he US is no a perfec

14 -11- subsiue for M3 in he euro zone. In conras, facors are laen variables which measure he "underlying process". The raionale is ha he behavior of several (naional) variables is driven by a few common (global) forces, he facors. Hence, he laer can provide an exhausive summary of he informaion included in daa across counries and regions. Moreover, and in conras o simple aggregaion, facor analysis allows he disincion beween common forces and idiosyncraic shocks, i.e. he amoun in he measured daa which is no considered o be par of he underlying global forces. Idiosyncraic componens mean ha he measured variables can include changes which are exclusively he resul of a naional daa-generaing process. Insead, if global variables are derived by aggregaion, he disincion beween idiosyncraic and global shocks is blurred. Any idiosyncraic shock semming from one (major) counry will ineviably influence he global aggregae and will herefore be couned as a common shock across counries. Global and idiosyncraic shocks are presumed o have he same influence, alhough idiosyncraic shocks should no influence he common movemen in an economic sense. The use of he FAVAR mehodology is especially appealing in he ligh of spillovers and global shocks, since i allows boh examining he ineracion of global variables and heir effecs on naional variables. For example, i is possible o derive he impac of a global liquidiy shock on global GDP, global commodiy, house and share prices, global inflaion, ec. Differen ypes of shocks can herefore be pu o some kind of a "horse-race": Is a global liquidiy shock more imporan for he global economy han a global commodiy price or global ineres rae shock? A he same ime, he response of every naional variable included in he respecive global facor (naional CPI, naional money supply, ineres rae, share price, ec.) due o a global shock (liquidiy, GDP, commodiy, ec.) can be examined. The problem in expanding he exernal secor (in a VAR model) is ha here is a rapid increase in he number of parameers ha need o be esimaed wih he addiion of each economy as well as he addiion of each secor of he respecive economy. The esablished approach adoped o circumven his problem of over-parameerizaion is o

15 -12- specify laen facors which capure he overall dynamic feaures of he inernaional economies (Fry, 24). Recen work wih Facor-Augmened Vecor Auoregression (FAVAR) suggess ha sandard VAR analysis can be improved by incorporaing he informaion in a large number of macroeconomic series. A general formulaion of he dynamic facor model is X i ) i( L f i wih i X as he observed daa for he macroeconomic ime series i a ime for i 1,..., N and 1,..., T. If he lag polynomials (L) are assumed o have i finie orders, he equaion can be wrien in saic form: X F. Hence, he facor F can be hough of as a weighed average of he variables in a daa se. The facor loadings, i.e. he weighs, can be eiher posiive or negaive and reflec how correlaed each variable is wih he facor. F, and are no direcly observable and have o be esimaed. To separae facors from idiosyncraic disurbances, he following idenifying assumpions are made (Jusiniano, 23): i j Orhogonaliy of idiosyncraic errors, i.e. ( ) i, j 1,..., N and i j. Usually, he assumpion of no cross-correlaion is relaxed. The model is hen said o have an approximae facor srucure. Orhogonaliy of facors, i.e. be correlaed in ime. j k f f k, j 1,..., K and k j. However, facors can i Idiosyncraic errors are orhogonal o facors, i.e. f j i 1,..., N and j 1,..., K. These assumpions imply ha all co-movemens across variables are aribued exclusively o a se of orhogonal facors. Sock and Wason (1998, 22) show ha he facors in a model of he form X F v can be consisenly esimaed by principal componen analysis when he ime series dimension (T) and he cross-secion dimension (N) are large. The facors are exraced in a sequenial fashion, wih he firs facor explaining he mos variaion in he daa se, he second facor explaining he nex mos variaion (no explained by he firs facor), and so on.

16 -13- The dynamic facor model in VAR form (FAVAR) can be obained by combining facor analysis (equaion 1) wih a VAR model (equaion 2): (1) X F D( L) X v (2) F ( L) F 1 G 1 where q r qp wih r saic facors F and q dynamic facors. is a n r marix, D(L) is a n n marix lag polynomial of order p, (L) is a r r marix lag polynomial of order p, G is r q, is a r -variae vecor of global shocks driving he common facors, v equals a n -variae vecor of idiosyncraic shocks. Subsiuing he facor evoluion equaion (2) ino equaion (1) and collecing erms yields he complee FAVAR form: F ( L) F 1 F (3) X ( L) D( L) X 1 X where F X I G. v The FAVAR conains he exclusion resricion implied by facor analysis, i.e. X does no predic F given ( L ) F 1. Resricions of his form closely resemble he assumpion of exogenous world variables which are used o idenify global shocks in open economy VARs. The FAVAR can be esimaed via a wo-sep principal componen approach. In he firs sep, he common componens principal componens of F are esimaed using he firs r X. In he second sep, a VAR is esimaed on hese common componens. Bernanke, Boivin and Eliasz (25) poin ou ha his wo-sep approach implies he presence of generaed regressors in he second sep. However, he uncerainy in he facor esimaes should be negligible when N is large relaive o T. se In he firs sep, esimaes of he common facors are obained by dividing he daa X ino caegories of variables. These caegories are capuring differen dimensions of 1 2 I he economy across counries ( X, X,..., X ) : economic aciviy as refleced by real GDP; inflaion which include consumer prices, producer prices, impor prices and he GDP deflaor; commodiy prices which include he HWWI and he CRB commodiy price index in domesic currency; house prices; moneary liquidiy which include broad

17 -14- moneary aggregaes, like M2 or M3; shor-erm ineres raes which include 3M ineres raes; and share prices which include he MSCI share price index in domesic currency. Esimaes for he global facors are obained as he firs principal componen for each sub-se (caegory) of series. Each segmen of X is herefore explained by exacly one facor. For example, global moneary liquidiy is esimaed as he firs principal componen from he se of moneary aggregaes in he G-7 counries plus he euro zone. (4) 1 X 1 2 X I X I 1 1 F e 2 2 F e I I F e 1 2 I In he second sep, a VAR wih he esimaed ( F, F,..., F ) is implemened. The innovaions in he VAR model can be idenified by applying sandard procedures, like a Cholesky decomposiion. However, a simple recursive ordering may no be appropriae in an inernaional conex. Given is low level of flexibiliy, one has ineviably o make exreme assumpions abou he ineracion beween global variables. For example, if one pus global money before he global shor-erm ineres rae, he ineres elasiciy of global money demand is consrained o be zero. In conras, if he global ineres rae is predeermined for global money, money supply s ineres elasiciy is assumed o be zero (Leeper and Roush, 23). As a resul, he derived moneary policy shocks migh be conaminaed. Non-recursive schemes allow for more general conemporaneous ineracions among variables han recursive orderings. A global srucural FAVAR (SFAVAR) posulaes more reasonable economic srucures and reflecs beer he complexiies of inernaional policy-making. We see wo major advanages by implemening a FAVAR in our conex. Firs, i is possible o derive srucural shocks on a worldwide level and heir impac on oher global variables. For example, global money demand shocks can be disenangled from global money supply shocks. Second, i is possible o derive he effecs of each global shock on specific naional variables (cf. equaion 3). Hence, impulse response funcions can be consruced for any variable included in he informaional daa se ( X ).

18 Empirical resuls 5.1. Baseline FAVAR The common facors are esimaed by principal componen analysis. As is common in facor model applicaions, he variables are iniially de-rended and sandardized. Apar from he shor-erm ineres raes, all X are log firs-differenced. For he shor-erm ineres raes only, firs differences are calculaed. Aferwards, all he de-rended variables are sandardized so ha each of hem has a mean of zero and a variance of one. Oherwise, he resuls would have been sysemaically affeced by cross-counry differences in variabiliy. For each global variable, he proporion of he oal variance of he series aribuable o each principal componen is calculaed. For he firs principal componen (PC1) o suiably qualify as a facor capuring inernaional co-movemen, one imporan condiion mus be me. PC1 should explain a sufficienly large fracion of he oal variance of he relevan daa se in comparison o he remaining principal componens of higher order. As can be seen in he able below, he requiremens are me for all of he common facors. For example, in he case of global money, a significan par of he oal variance (48.6%) can be aribued o he firs principal componen. In conras, PC2 accouns for only 18%. Even for he inflaion facor which includes four differen measures of inflaion for each counry or region (CPI, PPI, impor prices and he GDP deflaor), he firs principal componen s share is 31.8% compared o 16.9% for PC2. Wih 55.9% and 74.3%, he PC1 s share for commodiy and share prices is clearly he highes. Table 1 - Share of variance explained by firs hree principal componens PC1 PC2 PC3 Real GDP Inflaion Commodiy prices House prices Broad money M ineres rae Share price Noe: Calculaions based on firs sandardized differences

19 -16- We conclude ha common forces exis which qualify as global facors in he global economy and on inernaional financial markes. The seven global facors are herefore used for esimaing a SFAVAR in levels including a consan and a ime rend. 5 Since he global facors have been obained in firs-differences, hey are "re-consruced" in levels by seing each global facor equal o zero in he firs quarer of 1984 and calculaing he cumulaive sum of he firs principal componens. In order o idenify he global srucural shocks, he following assumpions are made: (5) 1 a a a a a a 1 a a a 1 a a a 1 a a a a a YR u PI u CP a 37 u HP u M u SR u SP 1 u YR PI CP HP M SR SP wih as he vecor of errors in he reduced-form equaions and u as he global srucural shocks. The global GDP (YR) and he global house price facor (HP) are no influenced conemporaneously by any oher global variable. The global inflaion facor (PI) is affeced conemporaneously only by he commodiy price facor (CP). The laer is affeced a he same ime by all oher common facors apar from inflaion. The global money facor is influenced conemporaneously by he global GDP and inflaion facor and by he shor-erm ineres rae (SR) as well. For he reacion funcion of cenral banks' worldwide, i is assumed ha hey reac conemporaneously o commodiy prices, he global house price facor and global money, bu no o he global aciviy and inflaion facor due o ime lags in publicaion. Given he forward-looking naure of financial markes, share prices (SP) respond o all oher global variables a he same ime. To deermine he lag lengh, we apply he usual crieria such as he Likelihood Raio es, he Final Predicion Error, he Akaike informaion crierion, he Schwarz crierion and he Hannan-Quinn crierion. Mos of he crieria poin a a lag lengh of wo, which is also sufficien o avoid serial correlaion among he residuals and seems o be 5 Since we now impose some srucure on he global economy, we inroduce he noion of a SFAVAR insead ha of a pure FAVAR analysis.

20 -17- appropriae in order o esimae a model which is parsimonious where possible. The LR es for overidenified VARs suggess ha our shor-run resricions canno be rejeced a any convenional confidence level. The saisic is equal o 5.1 and he corresponding p- value is.17. The major resul of our srucural facor augmened vecor auoregression (SFVAR) model is ha global liquidiy shocks are driving forces of he global economy and various naional economies. Moreover, he oucomes of our empirical analysis are in line wih economic heory, since frequenly emerging puzzles as, for example, he "price puzzle" and he "liquidiy puzzle" do no appear in our case. As can be seen in he chars below, a global liquidiy shock has a significan posiive impac on global GDP afer six quarers. As always, he solid line in each char represens he response o a one-sandard deviaion shock, again measured in sandard deviaions. The dashed lines represen he 95% confidence inervals boosrapped by ourselves based on a sandard residual boosrap procedure wih 5 draws (Enders, 24). Furhermore, global inflaion responds significanly wih a considerable ime lag of 11 quarers o a global liquidiy shock. However, in conras o our findings for global GDP, he inflaionary effec is far more persisen. Srong responses can also be found for he common house price and he shor-erm ineres rae facor. The global house price facor rises srongly and persisenly wihou any delay. This may indicae ha excess liquidiy on a worldwide level has conribued o he phase of excepionally high increases in residenial propery prices across counries. Global liquidiy shocks also lead o a marked liquidiy effec, driving shor-erm ineres raes down by up o one sandard deviaion. All in all, our impulse response analysis seems o confirm he resuls found by Rüffer and Sracca (26) and Belke, Orh and Sezer (28) on he basis of global VARs. Common liquidiy disurbances are influencing major macro variables on an inernaional level. However, wih respec o asse prices, our resuls show some marked differences versus previous empirical work. We are no able o find any significan impac boh on commodiy and share prices, whereas Belke, Orh and Sezer (28) repor a significan response of commodiy prices afer a global liquidiy shock. A recen

21 -18- sudy conduced by Alessi, Deken (29) finds ha global measures of liquidiy, like he M1 gap and he privae credi gap, are useful early warning indicaors for aggregae asse price booms in OECD counries. Their asse price measure includes house prices as well as commercial propery and share prices. Char 1 - Impulse response analysis in SFAVAR (global liquidiy shock) 2.5 SFAVAR: Global liquidiy -> global GDP 2. SFAVAR: Global liquidiy -> global inflaion SFAVAR: Global liquidiy -> global house price SFAVAR: Global liquidiy -> global 3M ineres rae SFAVAR: Global liquidiy -> commodiy price SFAVAR: Global liquidiy -> global share price

22 -19- Apar from a common liquidiy driving force, global demand and house price shocks prove o be imporan as well. For example, a sudden change in he global GDP disurbance leads o emporary higher inflaion from quarer wo o quarer eleven. Significan responses of common facors can also be found afer a global house price shock. In conras, he common shor-erm ineres rae is no a driving bu a driven force in he global economy. For example, an ineres rae shock does no rigger any significan response of he common house price facor. Insead, a sudden change in he disurbance of residenial propery prices leads o an increasing shor-erm ineres rae. The same insignifican resuls are found for he inflaion facor and commodiy and share prices (see Table A1 for more resuls). Char 2 - Impulse response analysis in SFAVAR (various global shocks) 2. SFAVAR: Global GDP -> global inflaion SFAVAR: Global house price -> global GDP SFAVAR: Global 3M ineres rae -> global house price 1.2 SFAVAR: Global house price -> global 3M ineres rae

23 -2- In order o invesigae he effecs of global and idiosyncraic srucural shocks on naional variables, we esimae equaion (3), using a separae VAR for each counry or region. The variables in levels are again re-consruced by calculaing he cumulaive sum of he sandardized naional variables. This is necessary o be fully consisen wih our approach on he global level. Since a naional economy can be hi by boh global and idiosyncraic shocks, one has o disinguish beween hese wo effecs. We do so by regressing X which consiss of global and idiosyncraic disurbances on he global srucural shocks ( u ) derived from equaion (5). On he basis of he obained idiosyncraic componens, resricions can be imposed for exac idenificaion on a naional level (Bagliano and Morana, 29). Again, we refrain from using a Cholesky decomposiion and implemen srucural relaionships beween he respecive variables as follows: (6) 1 a a a a a a a a a a a a a a a a a a a v v v v v a67 v 1 v YR PI HP M SR ER SP z z z z z z z YR PI HP M SR ER SP wih v as he vecor of errors in he reduced-form equaions and z as he idiosyncraic srucural disurbances. Our resricions on a naional level largely resemble he assumpions we made for global variables. For example, GDP (YR) and house prices (HP) are no influenced conemporaneously by oher naional variables. The inflaion figure (PI) equals consumer prices and is affeced only by he real effecive exchange rae (ER) a he same ime. The laer variable and share prices (SP) are assumed o respond conemporaneously o all oher variables. Before presening he srucural impulse response funcions, some cauionary remarks in inerpreing he resuls seem o be appropriae. The idenificaion paern in equaion (6) is used for all counries included in he sample, alhough hey differ markedly in economic size. In addiion, various insiuional seings in naional moneary

24 -21- and fiscal policy exis. Second, deriving he impac of global shocks on naional variables is raher cosly in erms of degrees of freedom. Since our SFAVAR conains seven facors and seven naional variables wih wo lags each, 3 parameers (including a consan and a deerminisic componen) have o be esimaed per equaion. Bagliano and Morana (29) esimae a oal of even 43 coefficiens by using a sample (only) ranging from Q1 198 o Q2 25. In he following, we focus on he effecs of global money supply (or wha we label global liquidiy shocks ) on naional variables. As has been he case before, we boosrap he confidence bands on he basis of a residual boosrap wih 5 draws. Accordingly, global liquidiy disurbances have a significan impac on broad naional moneary aggregaes in he US, he EMU and Canada. However, he effecs differ, hereby indicaing ha various naional ransmission mechanisms are a work. In he UK and Japan no significan impac of global liquidiy shocks on naional money supplies can be found. Global liquidiy shocks also rigger significan responses of oher major macroeconomic variables in some counries, like real GDP, consumer prices, house prices and shor erm ineres raes. A deailed overview on a counry level is included in Table A2 in he Appendix. For example, real GDP in he euro area reacs posiively afer a ime lag of seven quarers. In order o ge a yardsick, we derive he response of EMU GDP afer a srucural idiosyncraic money supply shock as well. The impulse response funcion displays a raher similar dynamic bu is somewha more pronounced in is impac on real GDP. House prices in he US are srongly affeced by global liquidiy shocks, whereas naional money supply disurbances play a comparaively small role. Hence, we suspec ha he bubble in he US residenial propery marke in recen years can no only be explained by excepionally low shor erm and long erm ineres raes bu by excessive global liquidiy as well. Ineresingly, global liquidiy shocks do no seem o be a major driver for he housing marke in he euro area. In conras, idiosyncraic disurbances o he money supply lead o srongly rising residenial propery prices in he EMU.

25 -22- Char 3 - Response of naional GDP and house prices afer global and idiosyncraic liquidiy shocks SFAVAR: Global liquidiy -> EMU GDP SFAVAR: EMU liquidiy -> EMU GDP SFAVAR: Global liquidiy -> US GDP SFAVAR: US liquidiy -> US GDP SFAVAR: Global liquidiy -> EMU house prices 1.4 SFAVAR: EMU liquidiy -> EMU house prices

26 Addiional global shocks Our SFAVAR focuses on global forces, like moneary liquidiy, inflaion, share prices, ec. However, here could be oher variables which also play a role in he global economy. Hence, we look for he influence of wo oher forces which may influence our SFAVAR approach: echnology and long-erm ineres raes. The imporance of echnology shocks is sressed in he real business cycle heory (RBC) and in he endogenous growh lieraure. There may be inernaional knowledge spillover effecs, like he impor of goods ha embody new echnologies, FDI flows, join venures and he migraion of key personnel (Klenow, Rodriguez-Clare 24). Especially rade-relaed new-good exernaliies could be cenral in ransmiing new echnologies from one counry o anoher. New goods of higher qualiy are inroduced and hen imiaed by oher companies worldwide. The second global force which could be ineresing is he behavior of long-erm ineres raes. Insead of global excess liquidiy, he shorage of financial asses could have played a cenral role in shaping he global economy in recen years (Caballero, 26). Emerging markes' FX reserves have been surging under he so-called Breon Woods II sysem. Given increasing global demand for financial asses and limied supply by indusrialized counries, long-erm ineres raes fell o hisorically low levels. In addiion, he phenomenon of perodollar recycling from commodiy-exporing counries exered furher downward pressure on real ineres raes. This in urn could have conribued o he srong rise in house prices in many counries. In order o derive echnology shocks, we use he idenificaion paern proposed by Galí (Galí 1999). This procedure has been discussed inensively in he las few years in he RBC lieraure. Accordingly, here are echnology and non-echnology shocks which are orhogonal o each oher. Galí s basic idenifying assumpion is ha echnology innovaions are he only shocks which have an effec on he long-run level of labor produciviy. Assuming ha boh variaions in he log produciviy ( x ) and log hours ( n ) are inegraed of order one, one ges he following expression: z x C ( L) C ( L) (7) m n C ( L) C ( L)

27 -24- where z and m equal he echnology and he non-echnology innovaions, respecively. Since Galí (1999) assumes ha he uni roo in produciviy sems exclusively from echnology shocks, he marix of long-run mulipliers is lower riangular ( C 12 (1) ). In order o esimae his approach on a global level, he following five-sep procedure is pursued. Firs, labor produciviy figures for he G-7 plus euro zone are calculaed by subracing he log of oal employee hours from he log of GDP for each counry or region. 6 Second, he obained figures for labor produciviy and oal employee hours are sandardized on he basis of firs differences wih a mean of zero and a variance of one. Third, global forces for labor produciviy and hours worked are esimaed by deriving he firs principal componen (PC1) each. Accordingly, for labor produciviy he proporion of he oal variance aribuable o PC1 is 26.1% compared o 22.5% for PC2. For oal hours worked, PC1 s share is 38.3% (PC2: 22.7%). Fourh, we re-consruc he wo firs principal componens in levels by seing each global facor zero in Q and calculaing he cumulaive sum. Fifh, we apply a baery of uni roo ess (Dickey- Fuller; Phillips-Perron; Kwiakowski, e. al., Ellio e. al.; Ng and Perron) for global produciviy and global hours worked. Accordingly, he majoriy of uni roo ess indicaes ha global produciviy is inegraed of order one. Wih respec o global hours worked, he empirical evidence is more mixed, wih some ess indicaing inegraion of order one or being saionary. Hence, concerning hours, equaion (7) is esimaed boh in levels and firs differences. All lag crieria poin lag o a lengh of one. Afer deriving he global echnology innovaion, he following equaion is esimaed: (8) z z z z z u c c1 1 c2 2 c3 3 c4 4 wih u as global srucural shocks derived from our SFAVAR (cf. equaion 5) z and as global echnology shock. Wald F-ess are done wih he null hypohesis ha he echnology innovaions do no have an impac on he global srucural shocks wihin he firs four quarers, i.e. c, c 1, c 2, c 3 and c 4 are joinly equal o zero. A consan in 6 A series for oal hours worked in he EMU is no available o our knowledge. We herefore consruc he EMU figure as he sum of hours worked in he following counries: Germany, France, Ialy, Spain, he Neherlands, Belgium, Ireland and Finland. On basis of PPP weighs provided by he IMF, our measure covers 89% of oal EMU GDP in 27.