Impact of International Monetary

Transcription

1 Impac of Inernaional Moneary Policy in Uruguay: A favar Approach Absrac Elizabeh Bucacos This sudy analyzes he Uruguayan economy s vulnerabiliy o foreign moneary policy in he las 20 years. The usual way of assessing moneary policy ransmission effecs such as panel daa analysis, correlaion analysis and even case sudies have no offered much saisically significan evidence for Uruguayan economic growh. However, being a small open dollarized economy wih a relaively less sophisicaed asse marke, i seems plausible ha Uruguay may suffer from inernaional moneary policy shocks. The challenge, hen, is o unveil he channels hrough which hose moneary shocks finally affec relevan Uruguayan variables. In his paper, facor augmened vecor auoregressive (favar) models are used in wo sages. In he firs sage, he impac of foreign moneary policy is assessed on commodiy prices, foreign oupu, and regional oupu. In he second one, he effecs on real exchange rae, domesic asses (as housing prices) and on domesic oupu are analyzed. Keywords: apering, emerging economies, housing prices, Uruguay. JEL classificaion: E42, R31, E62. E. Bucacos <ebucacos@bcu.gub.uy>, Economic Research Area, Banco Cenral del Uruguay. The auhor hanks for commens and suggesions by Gerardo Licandro, Fernando Borraz, Serafín Frache, Gonzalo Zunino, Ángel Esrada, Albero Oriz, Kolver Hernández, Hécor Valle, Caludia Ramírez, Francisco Ramírez, and paricipans of Join Research Moneary Policy Spillovers boh a he workshop in Mexico Ciy and a he final presenaion a XX Annual Meeing of he Cenral Bank Researchers Nework in Sano Domingo (2015). Also, I hank commens on ypographic and semanic syle made by Winer inerns from Harvard s Program in Mexico ha visied cemla, Devonae Freeland and Julie Zhu. The opinions and views in he aricle are exclusively he responsibiliy of he auhor and do no necessarily reflec he insiuional posiion of Banco Cenral del Uruguay. 321

2 1. INTRODUCTION On May 22h, 2013, in his esimony o Congress, he chairman of he Federal Reserve announced he possibiliy of a decrease in securiy purchases from 85 billion dollar a monh o a lower amoun. This apering alk had significan consequences for economic and financial condiions in emerging markes (em), refleced in he movemens in em exchange raes and sock prices following he announcemens (Figure 1). As many commenaors and analyss poin ou, no only was he impac sharp bu i was surprisingly large (Eichengreen and Gupa, 2013). The 2014 Regional Economic Oulook (reo) repors: Overall, he resuls presened so far sugges ha a gradual and orderly normalizaion of us moneary condiions should affec emerging marke bond markes in a relaively moderae fashion. Local yields have hisorically ended o respond o us moneary shocks, bu less han one for one. Oher news shocks, which include posiive us growh surprises, appear o have even more limied (and possibly benign) effecs on emerging marke bond yields. I poins ou ha here may be effecs, hough, in he flow of capial o em. 1 There are similariies and differences among em. In paricular, Uruguay is a small open economy sill highly dollarized wih a relaively poorly developed asse marke. I is basically a commodiy producer (mainly beef, wool, and mos recenly soybean) Brazil, Argenina, China, he us, and oher eu developed counries being is main produc desinaions; on he oher hand, Uruguay is a ne oil imporer. 2 Anoher imporan feaure of Uruguayan economy is is service secor which provides 56% of oal income boh from foreign (especially regional ourism) and inernal demand. 1 According o he simulaions repored by he imf, gross inflows decline markedly, falling by almos wo percen of gdp over six quarers in response o a 100-basis-poin increase in he real Treasury rae. When conrolling for oupu growh in he us he couner face of he normalizaion of us moneary policy, hey found ha ne capial flows o emerging markes respond posiively o an increase in us gdp growh despie he associaed rise in us ineres raes. 2 ancap (Adminisración Nacional de Cemeno, Alcohol y Porland) is he public enerprise ha monopolisically impors and refines oil. 322 E. Bucacos

3 Figure 1 US INTEREST RATES, AND EXCHANGE RATES AND STOCK PRICES FOR SELECTED EMERGING MARKETS 0.08 US INTEREST RATES (in percenage) FF rae T10 real rae FF real rae REAL EFFECTIVE EXCHANGE RATES (2010 =1) Mexico India I Brazil II III IV I II III IV I II III IV I II III IV I II III IV VOLATILITY OF STOCK PRICES (2010 =1) Mexico Brazil India Source: FRED and own calculaions for US ineres raes. FRED for exchange rae and sock prices. Impac of he Moneary Policy in Uruguay 323

4 Figure 2 URUGUAY: DOLLARIZATION BANK CREDITS BY THE PRIVATE BANKING SYSTEM (as percenage of oal bank credi) Foreign currency Domesic currency Source: BCU and auhor s calculaions. Figure 3 URUGUAY: DOLLARIZATION BANK DEPOSITS IN THE PRIVATE BANKING SYSTEM (as percenage of oal deposis) Foreign currency Domesic currency Source: BCU and auhor s calculaions. 324 E. Bucacos

5 Figure 4 URUGUAY: PUBLIC SECTOR DOLLARIZATION (as percenage of oal bank credi) Foreign currency denominaed GPS deb over GDP Foreign currency denominaed deb over oal global public secor deb Source: BCU and auhor s calculaions. A sylized fac of Uruguay is dollarizaion. There have been imporan aemps o alleviae his problem, bu Uruguayan economy sill remains highly dollarized: almos 80% of oal deposis and more han 50% of oal credis in he banking sysem are foreign currencydenominaed. The main problem, hough, is currency mismaches. According o recen sudies, 87% of Uruguayan firms repor o have liabiliies denominaed in currencies (mainly us dollars) differen from hose of heir incomes (mainly Uruguayan pesos). 3 In addiion, he public secor (33% of oal gdp) is mainly endebed in foreign currency. An imporan change in he Uruguayan economy in he las decade is he decrease in he dollarizaion of he public deb 4 and he increase in he average ime for mauriy. We expec ha hese changes reduce he Uruguayan economy s vulnerabiliy o global shocks. 3 See Licandro e al. (2014). 4 During he 2002 crises, more han 80% of oal public deb was denominaed in foreign currency; in 2002Q2-2002Q3, he nominal exchange rae jumped 16% and public deb denominaed in foreign currency over gdp rose from 70% o more han 150%, bu dropped o around 30% en years laer. I was 37% in 2014Q4. Impac of he Moneary Policy in Uruguay 325

6 Under hose circumsances, a igher moneary policy decided by he Federal Reserve sounds like bad news for a dollar-indebed counry ha does no prin dollars. Firs, a rise in he federal funds rae leads o a rise in marke raes hrough arbirage, increasing Uruguay s deb burden and worsening is exernal deb condiions. 5 Twelve-year susained economic growh ha began in 2003 may be pu o a hold. Second, a rise in he federal funds rae appreciaes he dollar agains oher currencies, in paricular he Uruguayan peso. This local currency depreciaion may fuel domesic inflaion, which is already ou of he arge zone, because many prices of he consumpion baske are updaed according o he depreciaion rae. 6 Third, higher inflaion may reduce invesmen projecs, which are needed for growh. The concern ha rising us ineres raes could slow or reverse he flow of capial o emerging markes is somehow miigaed for he case of Uruguay by he shallowness of is financial marke. For insance, real asses are he bigges par of a household s ne wealh, and no only are hey inensive in using cash (70%) bu also here is a low and sable use of credi (22%) and debi cards (8%). 7 As a resul, an observer migh wonder he rue dimension of he effecs of a new foreign moneary scenario. The challenge, hen, is o unveil he channels hrough which hose foreign (us) moneary shocks migh finally affec Uruguayan relevan variables. The sraegy ress on using informaion on pas performances o ry o figure ou he mos probable pah. There has been a lo of research on he effecs of regional facors on Uruguayan performance. 8 Favaro and Sapelli (1989) use var models o quanify he regional linkages of he Uruguayan economy for he period and hey find a large impac of regional variables especially bilaeral real exchange raes. Talvi (1995) calibraes he imporance of Argenina during wo exchange-rae-based sabilizaion programs aemped in Uruguay (Ocober 1978 and December 5 Alhough fixed-rae foreign public deb accouns for almos 90% of oal foreign public deb, i is denominaed in us dollars and, in ha way, varies according o he exchange rae evoluion. 6 A one-ime adjusmen in relaive prices does no necessary lead o inflaion, bu i may pu inflaionary pressures ino acion because oher relevan economic variables are cpi-indexed. 7 See Lluberas and Odriozola (2014) and Lluberas and Saldain (2015). 8 Sosa (2010) presens a deailed review of he relaed lieraure. 326 E. Bucacos

7 1990, respecively) hrough an ineremporal opimizaion model wih boh radable and regional goods. Bergara e al. (1994) develop a model semming from he ones wih Duch disease and a booming secor and incorporae a regional radable secor in order o analize he effecs of a regional demand shock and a shock o exernal capial inflows on Uruguayan performance. Masoller (1998) uses a near-var model o sudy he mechanisms of ransmissions of regional shocks in Uruguay. Bevilaqua, Caena and Talvi (2001) concenrae on rade linkages, formalize he concep of regional goods and analyze he vulnerabiliy of Argenina, Paraguay and Uruguay o real devaluaions in Brazil. Kamil and Lorenzo (1998) sudy he correlaion beween he Uruguayan business cycle and he cyclical componen of some key regional macroeconomic variables, finding ha he Uruguayan business cycle is srongly influenced by regional facors. Voekler (2004) sudies how regional shocks affec secoral Uruguayan oupu, finding ha he mos imporan causes of flucuaions a he secoral level are shocks o oupu and relaive prices in he region wih shocks from Argenina having he larges impac. In he same line, Eble (2006) finds ha Uruguay s exposure o regional shocks has adversely affeced growh in recen decades. Sosa (2010) examines he role played by regional facors in Uruguay, idenifies he sources and ransmission mechanisms of shocks semming from he region and assesses how vulnerable Uruguay is o a poenial crisis in he region. He uses a var model wih block exogeneiy resricions and finds ha shocks from Argenina which accoun for abou 20 % of Uruguayan oupu flucuaions have large and rapid effecs. Sosa poins ou ha his is mainly due o he exisence of idiosyncraic real and financial linkages beween Uruguay and Argenina, which also explain he very high correlaion beween heir business cycles. More recenly, he imf (2014) repor on Uruguay esablishes: The response of Uruguay s local currency bond yields o he change in us yields was 1.7, in line wih he la5 average bu lower han he beas of Colombian, Brazilian, and Peruvian local currency bonds (which were closer o 2.5). Similarly, he bea of Uruguay s long-erm foreign currency bond yields o us yields was 1.4, in line wih Colombia and Mexico, bu lower han he beas of Brazil, Chile and Peru. Thus, as in oher ems, Uruguayan yields moved more han one-for-one wih us bond yields in he afermah of he apering announcemen, alhough he increase in Uruguayan yields was a he moderae end of la6 reacions. Impac of he Moneary Policy in Uruguay 327

8 Neverheless, he impac on real aciviy of a sronger us recovery accompanied by an increase in em risk premiums would moderaely dampen growh in Uruguay hrough financial channels, according o he imf. In his paper, facor-augmened vecor auoregressive (favar) models are used for he firs ime wih Uruguayan daa in wo sages. In he firs sage, he impac of foreign moneary policy is assessed on commodiy prices, foreign oupu and regional oupu. In he second, he effecs on real exchange rae, domesic asses (as housing prices) and domesic oupu are analyzed. An ineresing alernaive o he favar approach is he global var (gvar) model inroduced by Dees e al. (2007) and recenly applied o Uruguayan daa by Noya e al. (2015). The gvar incorporaes an explici model for each counry which are linked by a se of observed and unobserved inernaional facors. In his way, he gvar is paricularly convenien when shocks come from very specific foreign counries insead of he res of he world. As argued by Mumaz and Surico (2008), he favar approach is paricularly convenien when one of he main goals is o analyze he response of a large number of home variables. The res of he paper is organized as follows. Secion 2 develops he prior research. Secion 3 describes he daa se and explains he way i is used. Secion 4 presens he resuls. Secion 5 performs some robusness ess and, finally, Secion 6 concludes. 2. PRIOR RESEARCH There is a vas empirical lieraure on he inernaional ransmission of moneary and nonmoneary shocks using small-scale srucural var. The main purpose of srucural var (svar) esimaion is o obain non-recursive orhogonalizaion of he error erms for impulse-response analysis. This alernaive o he recursive Choleski orhogonalizaion requires he user o impose enough resricions o idenify he orhogonal (srucural) componens of he error erms. Several researchers have proposed alernaive idenificaion srucures including, among ohers, he recursive schemes in Grilli and Roubini (1995), Eichenbaum and Evans (1995), and Faus and Rogers (2003); he nonrecursive schemes in Cushman and Zha (1997), Kim and Roubini (2000), and Kim (2001); and he sign resricions in 328 E. Bucacos

9 Canova (2005) and Scholl and Uhlig (2005). All of hem employ a relaively small number of variables (a var wih 14 variables) and have difficul o solve long-lasing puzzles in inernaional macroeconomics, 9 simulaneously. Mumaz and Surico (2009) use a wider informaion se in order o achieve a beer undersanding of inernaional ransmission of shocks and o ge new evidence o solve hose longlasing puzzles. This secion proposes a facor-augmened vecor auoregresssive (favar) model o assess he impac of a foreign moneary shock on relevan Uruguayan economic variables. The model resembles Bernanke, Boivin and Eliasz (2005), Mumaz and Surico (2009) and Fukawa (2012). 2.1 The favar Model Srucural facor models res on he idea ha a large number of observable economic variables can be described by a relaively small number of unobserved facors. These facors, in urn, can be affeced by a few shocks which can be undersood as macroeconomic disurbances. Consider n observed saionary variables. Le us assume ha each saionary variable of our macroeconomic daa se x i is composed of wo muually orhogonal unobservable componens, he common componen χ i and he idiosyncraic componen ξ i : 1 х i = χ i + ξ i. The idiosyncraic componens arise from shocks ha affec a specific variable or a small group of variables and may reflec secor specific variaions, variaions o foreign counries or measuremen errors. These componens can be weakly correlaed across variables bu common and idiosyncraic componens are orhogonal for each variable. The common componens are he ones responsible for mos of he co-movemens beween macroeconomic variables and are represened by a linear combinaion of a relaively small number ( r << n ) of unobserved facors (hese are also called saic facors in he lieraure): 9 Delayed exchange-rae overshooing and forward discoun puzzles. Impac of he Moneary Policy in Uruguay 329

10 2 X = a f + a f a f = Af. i 1i 1 2i 2 ri r The opimal number of facors can be deermined by several saisical ess, such as Bai and Ng (2002) and Onaski (2010) or Velicer s (1976). 10 Alhough facors do no need o have an economic meaning and heir main purpose is o summarize he informaion conen of he observed variables, someimes i is possible o find an economic inerpreaion for he firs few facors. When allowing a var model for vecor f componens, dynamic relaions among macroeconomic variables arise: 3 f = D f + D f + + D f p p ε, 4 ε = Ru, ( ) where R is an r q marix and u = u1 u2 uq is a q-dimensional vecor of orhonormal whie noises, wih q r. Such whie noises are he common or primiive shocks or dynamic facors (whereas he enries of f are he saic facors). Observe ha, if q < r, he residuals of he above var relaion have a singular variance covariance marix. From Equaions 1 o 3 i is seen ha he variables hemselves can be wrien in he dynamic form xi = bi( L ) u + ξ i, where p bi( L) = ai( I DL DL p ) 1 1 R. The dynamic facors u and b i (L) are assumed o be srucural macroeconomic shocks and impulse-response funcions, respecively. 11 Vecor auoregressive (var) models are very useful in handling muliequaion ime-series models because he economerician does no always know if he ime pah of a series designaed o be he independen variable has been unaffeced by he ime pah of he dependen variables. The mos basic form of a var reas all variables symmerically wihou analyzing he issue of independence. 10 The firs wo ess are used when principal componens analysis (pca) are applied o esimae he facors while he laer is used when facors analysis (fa) is applied. In pca, i is assumed ha all variabiliy in an iem should be used in he analysis while in fa only he variabiliy ha he iem has in common wih he oher iems is used. pca is preferred as a mehod for daa reducion while fa is ofen preferred when he goal is o deec srucure. See discussion secion. 11 They are called dynamic facor models. 330 E. Bucacos

11 5 p O = i i + i = 1 O AO u Neverheless, here are some ools such as Granger causaliy, impulse-response analysis and variance decomposiion ha can shed some ligh on he undersanding of heir relaion and guidance ino he formulaion of more srucured models. Facor-augmened var (favar) models combine facor models and var models a he same ime:. 6 F L L F = φ O L L 11( ) φ12( ) φ φ 21( ) 22( ) O 1 1 F u + O u, ( ) where O is he M 1 vecor of observable variables and F is he ( k 1 ) vecor of unobserved facors ha capures addiional economic informaion relevan o model he dynamics of O. Unobserved facors are exraced from he informaional ime series included in he daa se. The number of he informaional ime series is large and mus be greaer han he number of facors (r) and observed variables in he favar sysem. Le us assume ha he informaional ime series X are relaed o he unobservable facors F by he following observaion equaion: f O 7 X =Λ F +Λ O + e, ( ) ( ) ma- where F is a k 1 vecor of common facors, 12 f Λ is a N k rix of facor loadings, Λ O is ( N M), and e are mean zero and normal, and assumes a small cross-correlaion, which vanishes as N goes o infiniy. 2.2 The Empirical Model The favar approach developed by Bernanke e al. (2005) was exended o he open economy by Mumaz and Surico (2009) in order o model he ineracion beween he uk economy and he res of he world, which hey call he foreign block. They occupy a large panel of daa covering 17 indusrialized counries and around 600 price, 12 Unobservable facors in favar do no have exac meanings. The Forni and Gambei (2010) model is differen from favar in ha hey ried o give he facors hemselves a srucural inerpreaion. Impac of he Moneary Policy in Uruguay 331

12 aciviy, and money indicaors. They have only one observable variable, hough, he uk shor-erm ineres rae. In our model, however, here are six domesic observable variables because our main goal is o invesigae domesic ransmission channels of a foreign shock, in paricular, us moneary shock. The model presened here consiss of hree blocks: The foreign * observable variables, O ; he informaion abou he indusrialized world, he relevan region and he Uruguayan economy, which is summarized in k unobserved facors, F ; and he domesic observable variables, O. As a resul, he dynamic sysem moves according o he following ransiion equaion: 8 O F O O = B( L) F O * * u, where B(L) is a comformable lag polynomial of finie order p, 1/2 and u = Ω e wih he srucural disurbances e N ( 01, ) and Ω= A A 0( 0). The unobserved facors are esimaed by maximum likelihood and he opimum number of facors is deermined using Velicer s minimum average parcial (map) mehod, and saring values for he communualiies 13 are aken from he squared muliple correlaions (smc). Oher auhors consisenly esimae he unobserved facors by he firs r principal componens of X (Sock and Wason, 2002). For his resul o hold, i is imporan ha he esimaed number of facors, k, is larger han or equal o he rue number, r. Because N is sufficienly large, he facors are esimaed precisely enough o be reaed as daa in subsequen regressions. 14 The esimaed loadings and facors are no unique; ha is o say, here may be ohers ha idenically fi he observed covariance srucure. This observaion lies behind he noion of facor roaion, in which ransformaion marices are applied o he original facors and loadings in he hope of obaining a simpler and easier-o-inerpre 13 Communualiies are he common porion of he variance of he variable. See EViews 9 Reference Manual. 14 See Fukawa (2010). 332 E. Bucacos

13 facor srucure. I apply an orhogonal roaion implying ha he roaed facors are orhogonal. In he second sep, I esimae he favar equaion, replacing F wih Fˆ. As a resul, he response of any observable variable o a shock in he ransiion Equaion 8 can be raced ou applying he facor loadings and Equaion Discussion Several criicisms of he var approach o policy shock idenificaion focus on he small amoun of informaion used by low-dimensional var. To conserve degrees of freedom, sandard var rarely employ more han 10 variables, even hough his small number of variables is unlikely o span he informaion ses acually used by he policymaker. Using low-dimensional var means ha he measuremen of policy innovaion is likely o be conaminaed. Facor-augmened var (favar) models iniiaed by Bernanke e al. (2005) are a mixure of a facor model and a var model. The facors can provide an exhausive summary of he informaion in large daases, and in his sense hey are precious o alleviae omied variable problems in empirical analysis using radiional small-scale models (see Bernanke and Boivin, 2001). In fac, Bernanke and Boivin (2001) and Bernanke e al. (2005) proposed exploiing facors in he esimaion of var o generae a more general specificaion. Chudik and Pesaran (2007, 2011) illusrae how a var augmened by facor could help in keeping he number of parameers o be esimaed under conrol wihou loosing relevan informaion. Facor models impose a considerable amoun of srucure on he daa, implying resriced var relaions among variables (see Sock and Wason, 2005, for a comprehensive analysis). In his sense, facor models are less general han var models. On he oher hand, facor models, being more parsimonious, can model a larger amoun of informaion. The abiliy o model a large number of variables wihou requiring a huge number of heory-based idenifying resricions is a remarkable feaure of srucural facor models. If economic agens base heir decisions on all of he available macroeconomic informaion, srucural shocks should be innovaions wih respec o a large informaion se, which can hardly be included in a var model. The esimaion of favar models is usually done following a wo-sep procedure in which he facors are found firs and hen he co-movemens among he observed variables and he facors are analyzed. Impac of he Moneary Policy in Uruguay 333

14 Some auhors sugges exracing facors by he firs of principal componens of he series involved, such as Bernanke e al. (2005) and Boivin and Giannoni (2008), among ohers. There are oher researchers ha prefer o apply a maximum-likelihood mehod in he firs sep. Resuls given by principal componens analysis (pca) and facor analysis (fa) are very similar in mos siuaions, bu his is no always he case, and here are some problems where he resuls are significanly differen. Boh pca and fa creae variables ha are linear combinaions of he original variables. Bu differen from pca, fa is a correlaion-focused approach seeking o reproduce he iner-correlaions among variables, in which he facors represen he common variance of variables, excluding unique variance. In erms of he correlaion marix, his corresponds wih focusing on explaining he off-diagonal erms (i.e., shared covariance), while pca focuses on explaining he erms ha are on he diagonal. However, as a side resul, when rying o reproduce he on-diagonal erms, pca also ends o fi relaively well he off-diagonal correlaions. pca resuls in principal componens ha accoun for a maximal amoun of variance for observed variables; fa accouns for common variance in he daa. Tha is one of he reasons why fa is generally used when he research purpose is o deec daa srucure (i.e., laen consrucs or facors) or causal modeling while pca is generally preferred for purposes of daa reducion (i.e., ranslaing variable space ino opimal facor space) bu no when he goal is o deec he laen facors. An imporan drawback of fa, however, refers o is heurisic analysis of facors, because more han one inerpreaion can be made from examining he same daa facored in he same way. 3. DATA 3.1 Policy Rae The effecive federal funds rae has been he measure for he Federal Reserve s moneary policy sance in he economic lieraure and has been used as he link beween moneary policy and he economy. Bu since he end of 2008, he effecive federal funds rae has been a he zero lower bound (zlb), damping is hisorical correlaion wih economic variables like real gross domesic produc 334 E. Bucacos

15 (gdp), he unemploymen rae, and inflaion. To provide a furher boos o he economy, he Federal Open Marke Commiee (fomc) has embarked on unconvenional forms of moneary policy (a mix of forward guidance and large-scale asse purchases) since hen. 15 Aemps o summarize curren policy have led some researchers o creae a virual federal funds rae. Specifically, Wu and Xia (2014) consruc a new policy rae by splicing ogeher he effecive federal funds rae before 2009 and he esimaed (by hem) shadow rae since This combinaion makes he bes use of boh series (p. 11). On he oher hand, Bauer and Rudebusch (2015) wrie: The sensiiviy of esimaed shadow shor raes raises a warning flag abou heir use as a measure of moneary policy, as in Ichiue and Ueno (2013) and Wu and Xia (2014). Our findings show ha such esimaes are no robus and srongly sugges ha heir use as indicaors of moneary policy a he zlb is problemaic. More promising approaches have recenly been suggesed by Lombardi and Zhu (2014), who infer a shadow shor rae ha is consisen wih oher observed indicaors of moneary policy and financial condiions, and Krippner (2015), who considers he area beween shadow raes and heir long-erm level. Alhough here is sill no consensus regarding which variable o use for moneary policy analysis, i is clear ha he effecive federal funds rae does no seem very appealing for i was no an accurae reflecion of he moneary policy decisions aken by he Federal Reserve during he zlb period when he effecive federal funds rae did no move. Bu as shadow ineres raes are unobserved, here is no absolue cerainy abou heir esimaed values and hey differ grealy among differen researchers. As a resul, in his sudy I perfom a sensiiviy analysis and I alernaively use he effecive federal funds rae (ffr) and he Wu-Xia virual funds rae (ffr_im), boh in real erms. 3.2 Descripion of he Daa X consiss of 36 quarerly macroeconomic ime series. 16 All of hem are expressed in real erms and in log levels (excep raios and ineres 15 For a deailed lis see Engen e al. (2015). 16 Alhough he lieraure advises handling a larger number of ime series, daa availabiliy was binding in his sudy. Impac of he Moneary Policy in Uruguay 335

16 raes) and whenever necessary, series are ransformed in order o leave hem saionary. 17 The daa span he period from 1995Q2 o 2014Q4. 18 Federal funds rae (ffr); 10-year bond rae (T10); real exchange rae (rer); domesic passive ineres rae (i_p); Uruguayan counry-risk (UBI); domesic oupu (y); and housing prices (p_h) are he observable variables O. The informaional variables also include several commodiy prices (whea, soybean, food, oil); foreign oupu (from Argenina, Brazil, usa, China, uk, Ialy, Spain, Germany, Mexico); us deb-o-gdp raio, domesic invesmen raio (oal, public and privae), rade (expors and impors), real domesic wages, unemploymen, public deb-o-gdp raio (oal, foreign, domesic, in foreign currency, in domesic currency), public asses-o-gdp raio, oal public secor income, and oal public secor expendiures including ineress. 3.3 Model Specificaion I firs esimae a baseline var model on eigh variables of ineres: Federal funds rae in real erms (FFR ); 10-year bond rae in real erms (T10); real exchange rae (rer); domesic passive ineres rae (i_p) in real erms; Uruguayan counry-risk raio (ubi); real domesic oupu (y); housing prices (p_h) 19 in real erms, and he public secor balance (pb) in real erms. In order o assess he impac of foreign moneary policy changes, I propose he following ransmission mechanism. If we suppose ha he Federal Reserve decides o change is rae (ffr), i will affec oher marke raes boh foreign and domesically hrough arbirage (T10 and i_p) and will deermine changes in domesic real exchange rae (rer), affecing domesic real oupu (y), domesic asse prices (p_h) and public secor balance (pb): 9 p O = i i + i = 1 O AO u where O FFR,, T10, rer, UBI, ip, ph, y, pb. The informaion crieria selec hree lags for he var model, which saisfies he sabiliy = ( ), 17 Sandard uni roo ess (augmened Dickey-Fuller and kpss) show ha all variables are saionary in firs differences, excep for he ineres raes; deseasonalizaion echniques were applied when necessary. 18 China gdp is available only since 1995Q2. 19 This will be he ordering ha will be used aferwards when performing impulse-response analysis. 336 E. Bucacos

17 condiion. The resuls show ha a conracionary foreign moneary policy (a one-ime rise of ffr) has no clear effecs on Uruguayan real oupu, nor housing prices or fiscal accouns (see Figure 5, graphs 7, 6 and 8, respecively). Then, I explore he possibiliy of he exisence of oher unobserved variables ha may influence he behavior of he observable ones. These variables may resume valuable informaion and be par of a more global ransmission mechanism ha is no very easy o describe a firs sigh. I seems plausible o ry o find a few facors ha could ac as vehicles once he foreign moneary shock akes place. Nex, I consider he exension of he baseline var model: 10 * O φ11( L) φ12( L) Φ13( L) O F = φ21( L) φ22( L) Φ23( L) F O Φ ( 31 L) Φ32( L) Φ33( L) O * * O u F + u, O u * where O = ( FFR, T10 ), O rer, UBI, ip, ph, y,and pb F F1, F2, F3 are he facors esimaed in he firs par by maximum likelihood. Four lags are used, based on informaion crieria (sic) and sabiliy consideraions. = ( ) = ( ) 4. RESULTS 4.1 Esimaion I esimae he model applying a wo-sep procedure. In he firs sep, he unobserved facors and heir corresponding loadings are esimaed by maximum likelihood. In he second sep, I subsiue he esimaed facors ino a var specificaion and esimae he favar model by ols. The whole available daa se is used in order o esimae he facors. Neverheless, following measures of sampling adequacy (msa) and goodness-of-fi crieria, several ime series are dropped ou of he daa se. In effec, only ime series whose msa values are greaer or very close o Kaiser s msa 20 remain. The final daa se has a Kaiser s msa 20 msa is an index of facorial simpliciy ha lies beween 0 and 1 and indicaes he degree o which he daa are suiable for common facor analysis. Values for he msa above 0.90 are deemed marvelous; values in Impac of he Moneary Policy in Uruguay 337

18 Figure 5 IMPULSE-RESPONSE FUNCTIONS, BASELINE VAR (5,000 MONTE CARLO REPLICATIONS) 1.2 RESPONSE OF FFR_REAL TO FFR_REAL RESPONSE OF D(L_RER) TO FFR_REAL RESPONSE OF D(UBI_URU) TO FFR_REAL RESPONSE OF D(LPIB_URU_SA_2005) TO FFR_REAL Source: FRED and own calculaions for US ineres raes. FRED for exchange rae and sock prices. 338 E. Bucacos

19 Figure 5 (con.) IMPULSE-RESPONSE FUNCTIONS, BASELINE VAR (5,000 MONTE CARLO REPLICATIONS) RESPONSE TO CHOLESKY ONE S.D. INNOVATIONS ± 2 S.E. RESPONSE OF TREA10_REAL TO FFR_REAL RESPONSE OF I_P_MN-PI_URU TO FFR_REAL RESPONSE OF D(LP_VIVIENDF) TO FFR_REAL RESPONSE OF D(L_ING_GC_BPS_R_2_D11-L_EG_PLUS_INTS_R_D11) TO FFR_REAL Source: FRED and own calculaions for US ineres raes. FRED for exchange rae and sock prices. Impac of he Moneary Policy in Uruguay 339

20 value of 0.79 which can be labled beween middling and meriorious for common facor analysis. I ake he decision o keep Argenine and Brazilian real oupu and whea price, even hough hey have indicaors a bi lower han 0.79 because here is a rade-off beween a labeling of almos middling and he acual imporance of hose variables in domesic dynamics. I mus be aken ino accoun ha he final daa se had o be shorened a grea deal 21 in order o have a balanced panel of ime series. Velicer s map mehod has reained hree facors, labeled F1, F2 and F3. A brief examinaion of he roaed loadings indicaes ha commodiy prices (food, whea and soybean) and real wages load on he firs facor, while foreign real oupu (from he usa, Germany, Spain, he Unied Kingdom, Ialy, and probably Mexico) and American deb load on he second facor, and oil price and a relevan regional foreign real oupu (Argenina, Brazil and China) load on he hird facor. Therefore i is reasonable o label he firs facor as a measure of commodiy prices, he second facor as an indicaor of foreign demand from developed counries and he hird facor as an aggregae variable for he regional demand Idenificaion of Srucural Shocks The dynamics of he variables in he sysem depend on he srucure imposed on he facor loadings. As such, I propose differen idenificaion schemes in order o ponder he sensiiviy of he responses when a specific unanicipaed 23 rise in he foreign ineres rae occurs: a recursive idenificaion scheme (Choleski) and a non-recursive one. In he recursive scheme, he impac marix A 0 is lower riangular, implying ha boh us moneary policy and foreign variables do no respond o Uruguayan performance measured by real oupu, for insance conemporaneously. On he oher hand, he Uruguayan economy reacs in he same period o changes occurred in he res of he world, in he relevan region and in he variables ha ac as he 0.80s are meriorious; values in he 0.70s are middling; values he 0.60s are mediocre, values in he 0.50s are miserable, and all ohers are unaccepable (Kaiser and Rice, 1974). 21 I spans from 1980Q1 o 2014Q4, originally. 22 Recall, again, ha some auhors do no give facors an economic inerpreaion, raher a saisical one. 23 us moneary policy normalizaion can be regarded as unanicipaed because is precise iming of occurrence is unknown. 340 E. Bucacos

21 341 Impac of he Moneary Policy in Uruguay linkages beween hem: u u u u u u u u u u u R T F F F rer UBI i p y pb p h = e e e e e e e e e e e R T F F F rer UBI i p y pb p h , where sands for freely esimaed parameers. In he non-recursive scheme, he resricions imposed 24 are: u u u u u u u u u u u R T F F F rer UBI i p y pb p h = e e e e e e e e e e e R T F F F rer UBI i p y pb p h , which imply differen reacions of unobserved facors o foreign ineres raes. Mumaz and Surico (2009) idenify he unobserved facors hrough he upper 3 N block of he marix Λ f, which is assumed o be block diagonal. Here, I impose zero resricions on some of 24 In fac, hey come afer an opimizaion procedure applied on he daa iself, ha is, I esed for saisical significance of he conemporanoues effecs from he Choleski facorizaion.

22 he facor loadings. In effec, commodiy prices do no seem o reac o conemporaneous movemens of he federal funds rae bu o changes in he en-year bond rae wihin he period, while foreign demands boh from he developed counries (F 2 ) and he relevan region (F 3 ) reac o unanicipaed changes in boh foreign ineres raes. There is no conemporaneous response of domesic oupu o a FFR change because real aciviy seems o reac hrough a specific paern: Those hree unobserved facors canalize he iniial change in us moneary policy insrumens, affecing domesic ineres rae direcly and hrough real exchange rae and counry-risk, and finally reaching domesic oupu. Only real exchange rae and counry risk influence each oher wihin he same period, besides us ineres rae and commodiy prices. Counry risk varies conemporanously wih 10-year bond ineres rae and he relevan region demand (F 3 ). Domesic ineres rae does no respond o ffr conemporaneously bu o oher unanicipaed innovaions coming from he en-year bond rae, commodiy prices, developed counries demand, real exchange rae and counry-risk changes. The asse prices considered here (housing prices) are percieved as anoher ype of financial invesmen, and hus hey reac conemporaneously o innovaions semming from foreign ineres raes, commodiy prices, developed-counries demand, real exchange rae, domesic ineres rae and counry risk. Finally, he domesic fiscal balance does no seem o reac o changes in any of he variables considered ha ake place in he same period. 4.3 Impulse-response Analysis Once he baseline model is expanded ino a favar model, he dynamics seem more plausible because an unambiguous response of all he observed variables is reached, especially for domesic oupu. There is a clear and saisically significan impac effec bu he following resuls are uncerain (Figure 6). Under he recursive shock idenificaion scheme, an increase of one sandard deviaion of ffr (2.3 or 230 basis poins) reduces quarerly oupu growh by 0.40% on impac bu as confidence inervals grow raher fas as ime goes by, forecass are no credible 25 (see Fig- 25 In impulse-response exercises, responses are deermined from he esimaed process parameers and are herefore also esimaes. Generally, esimaion uncerainy is visualized by ploing ogeher confidence inervals wih impulse-response coefficiens (see Luekepohl, 2011). If he confidence 342 E. Bucacos

23 Figure 6 FAVAR: IMPULSE-RESPONSE FUNCTION FOR D(Y) (10, 000 Mone Carlo replicaions) Source: BCU and auhor s calculaions. ures 6 o 12). Under he non-recursive shock idenificaion scheme, an increase of one sandard deviaion of ffr (2.3 or 230 basis poins) reduces quarerly oupu growh by 0.31% on impac bu, again, as confidence inervals grow raher fas as ime goes by, i is no possible o have credible forecass. The responses of he variables when a nonrecursive idenificaion of srucural shocks is applied are prey similar o he ones described in Figures 6 o 12. The only difference is ha hey always have a smaller value. Tha is o say, heir dynamic pahs are he same bu he acual responses are a bi lower 26. There seems o be four channels hrough which a one-ime rise in ffr affecs real oupu in Uruguay. These are: he commodiy price channel (Figure 7); he aggregae demand channel (oecd counries and relevan region, Figures 8 and 9); and he asses channel (exchange rae and housing prices, Figures 10 and 11). They can be oulined by analyzing he following irfs. inerval crosses he horizonal axis, however, he forecas can eiher be posiive or negaive wih he same probabiliy and herefore he esimae does no add any useful informaion. Tha is why I employ he expression credible forecass. 26 The resuls are available upon reques. Impac of he Moneary Policy in Uruguay 343

24 0.04 Figure 7 FAVAR: IMPULSE-RESPONSE FUNCTION FOR F1 (10,000 Mone Carlo replicaions) Source: BCU and auhor s calculaions. Once ffr rises, arbirage makes marke ineres raes rise and some financial asses become ineresing and commodiies become less aracive as financial invesmens. Figure 7 plos he evoluion of F1 facor (labeled commodiy prices facor). Only a significan negaive impac can be seen in response of a one-ime rise in ffr in real erms. Aferwards, here is grea uncerainy and nohing can be said. Then, he demand channel appears. Developed counries oupu declines, responding o he ffr rise and he decline in commodiy prices. This can be seen in Figure 8, where facor F2 significanly drops on impac. The effec coming from he so-called region is no so clear. In essence, in Figure 9 no saisically significan response is repored. Tha may arise from he way he F3 facor is composed, ha is, relevan regional oupu (Argenina, Brazil, and China which, excep for China, have limied linkages o he Unied 27 ) and oil price. Foreign moneary policy ransmission is usually done hrough changes in asse prices and capial flows. A ighening in foreign moneary 27 There are modes rade linkages beween Uruguay and he Unied Saes (only four percen of Uruguay s expors are desined for he Unied Saes). Indirec rade linkages are also limied: Almos 30 % of oal Uruguayan expors go o Brazil and Argenina which also have limied rade linkages wih he Unied Saes. 344 E. Bucacos

25 Figure 8 FAVAR: IMPULSE-RESPONSE FUNCTION FOR F2 (10,000 Mone Carlo replicaions) Source: BCU and auhor s calculaions Figure 9 FAVAR: IMPULSE-RESPONSE FUNCTION FOR F3 Source: BCU and auhor s calculaions. (10,000 Mone Carlo replicaions) Impac of he Moneary Policy in Uruguay 345

26 Figure 10 FAVAR: IRF FOR D(RER) (10,000 Mone Carlo replicaions) Source: BCU and auhor s calculaions Source: BCU and auhor s calculaions. Figure 11 FAVAR: IRF FOR D(P_HOUSE) (10,000 Mone Carlo replicaions) E. Bucacos

27 Figure 12 FAVAR: IRF FOR D(PB) (10,000 Mone Carlo replicaions) Source: BCU and auhor s calculaions. policy usually leads o a depreciaion of local currency as a consequence of he greaer araciveness of foreign currency-denominaed asses and capial mobiliy (ineres rae pariy), which will lead o a local capial exi which in urn will affec financial asse prices (see Figures 10 and 11). Finally, he asses channel poins o a decrease in housing prices once ffr rises. As inflaion had been presen in he Uruguayan economy for a very long ime, 28 economic agens in a shallow financial marke sough hedge in oher asses such as housing invesmen. I can be seen ha an increase in ffr (in real erms) lowers housing prices (in real erms) because hey lose relaive value as an invesmen. Figure 11 shows a significan effec unil he second period. The effec of a us moneary policy change on Uruguayan fiscal accouns is ambiguous, because is primary balance could eiher be 0.76% beer or 1.05% worse on impac. This siuaion is never solved and he final oucome is inconclusive. On he one hand, a fall in domesic oupu will drag income axes down, increasing he fiscal defici; on he oher hand, domesic currency depreciaion may play a dual role. I will increase deb paymens 28 Alhough several aemps o eliminae is negaive effecs had failed, unil a successful sabilizaion plan was implemened in he 1990s. Impac of he Moneary Policy in Uruguay 347

28 Figure 13 THE HISTORICAL DECOMPOSITION OF THE URUGUAYAN OUTPUT D(y) D(y_FFR shock) Source: BCU and auhor s calculaions. and impored goods purchases, which will increase he fiscal defici and will also reduce domesic expenses in real erms hrough higher inflaion, which will reduce he fiscal defici in real erms. Thus, he final resul is ambiguous. 4.4 Variance Decomposiion Analysis While irf consiue a pracical way o idenify he dynamic responses of he Uruguayan economy o exernal moneary shocks, illusraing how growh in Uruguay has ended o reac o differen shocks, variance decomposiion, in urn, provides a quanificaion of he relaive imporance of hose variables as sources of shocks affecing oupu flucuaions in Uruguay. Thusly, around 9% of domesic oupu flucuaions in he firs period can be explained by foreign ineres raes 29 (boh ffr and T10) and 6% by commodiy prices (F 1 ). As ime passes, he relaive imporance of foreign ineres raes and regional demand are almos he same. 30 The hisorical decomposiion of he Uruguayan oupu growh 29 Recall ha he impulse came from a rise in ffr. 30 Recall ha Choleski s ordering is: ffr, T10, F1, F2, F3, rer, ubi, i_p,, p_h, y, pb. Resuls are available upon reques. 348 E. Bucacos

29 rae shows ha us moneary policy shocks have had a relaively imporan impac on Uruguayan domesic oupu performance boh during recession and during economic booms. The esimaed ime series D(y_ffr shock) plos wha would have happened if only us moneary policy shocks had driven he daa. 4.5 Robusness The previous resuls are robus o differen orderings of he shocks, beginning always by ffr. There is a sligh change in he resuls, however, when counry-specific risk (measured by ubi) is handled eiher as an exogenous or an endogenous variable. I prefer o consider i endogenous because i can be argued ha counry risk may be influenced by real oupu performance which in urn is affeced by foreign moneary policy. 31 When counry-specific risk is reaed as exogenous, an increase of one sandard deviaion of ffr (230 basis poins) reduces quarerly oupu growh by 0.49% on impac bu growing confidence inervals render fuure oucomes uncerain. Impulse-response analysis is done on he favar esimaed equaion using a simple recursive framework (Choleski decomposiion) o idenify srucural shocks. Sensiiviy analysis is performed by changing he ordering of he variables, and he main resuls remain unchanged. Then, I proceed o subsiue he effecive federal funds rae (ffr) wih he Wu-Xia virual effecive federal funds rae (ffr _im) in he favar esimaion. I perform impulse-response analysis and all he dynamics described before are found again. In he new scenario, however, here is more uncerainy. Specifically, an increase in one sandard deviaion of ffr _im (289 basis poins) could make quarerly oupu growh eiher rise 0.34% or drop 0.60%, wih a mean value of Changes in inernaional real ineres raes consiue an imporan facor driving porfolio capial inflows o Lain America, hus influencing business cycles across he region (Calvo, Leiderman, and Reinhar, 1993, and Calvo, Fernandez Arias, Reinhar, and Talvi, 2001). Low ineres raes in maure markes may lead invesors here o seek higher reurns in oher markes, increasing he demand for emerging marke asses. No only does exernal financing become more abundan for emerging markes, bu also he cos of borrowing declines as a consequence of he lower ineres raes in he usa. In fac, Fernandez Arias (1996) shows ha counry-risk premia in emerging markes is indeed affeced by inernaional ineres raes, amplifying he ineres rae cycles in maure markes (Sosa, 2012). Impac of he Moneary Policy in Uruguay 349

30 I also applied block resricions on he favar equaion 32 in order o preven feedbacks from he observed domesic variables o he foreign ineres rae and he unobserved facors blocks: * O φ11( L) 0 0 O F = φ21( L) φ22( L) 0 F O Φ31( L) Φ32( L) Φ33( L) O * u + u u where O rer, UBI, ip, ph, y, pb, F = ( F1, F2, F3 ), are he facors esimaed in he firs par. Again, he unanicipaed moneary policy shock affecs he real economy by he same channels found in previous exercises in his sudy regardless of he foreign ineres rae used (see Figures A.1 and B.1 in Annex 2). However, when ffr _im is used as he Federal Reserve s moneary policy sance, he effecs on domesic variables are relaively sharper. = ( ) * O F O, 5. CONCLUSION The aim of his sudy is o analyze he vulnerabiliy of he Uruguayan economy o changes in us moneary policy by describing is linkages wih oher relevan variables in he las 20 years. The usual way of assessing moneary policy ransmission effecs such as panel daa analysis, correlaion analysis and even case sudies have no offered much saisically significan evidence for Uruguay. However i seems plausible ha Uruguay, as a small open dollarized economy wih a relaively less sophisicaed asses marke, may suffer from inernaional moneary policy shocks. The challenge, hen, is o unveil he channels hrough which hose shocks finally affec relevan Uruguayan variables. A facor-augmened vecor auoregressive (favar) model is implemened for he firs ime on a quarerly balanced Uruguayan daa se ha span from 1996Q2 o 2014Q4. 33 This approach is preferred o a radiional var because favar models, being a mixure of facor 32 A hree-lag favar wih block resricions was esimaed as a seemingly unrelaed regression (sur). 33 Sample adjused for lagged variables. 350 E. Bucacos

31 models and var models, enable he researcher o incorporae more informaion wihou adding more variables and allow a beer idenificaion of srucural shocks. In his paper, favar models are used in wo sages. In he firs sage, he impac of foreign moneary policy is assessed on commodiy prices, foreign oupu and regional oupu. In he second sage, he effecs on real exchange rae, domesic asses (as housing prices) and domesic oupu are analyzed. While irf consiue a pracical way o idenify he dynamic responses of he Uruguayan economy o exernal moneary shocks, illusraing how growh in Uruguay has ended o reac o differen shocks, variance decomposiion, in urn, provides a quanificaion of he relaive imporance of hose variables as sources of shocks affecing oupu flucuaions in Uruguay. Hisorical decomposiion helps o assess he relaive imporance of foreign moneary policy shocks in he Uruguayan economy. According o he exercises conduced in his invesigaion, Uruguay seems o be reachable. A rise of 230 basis poins in he federal funds rae (in real erms) drops Uruguayan oupu growh rae by 0.4% a once; neverheless, wha happens aferwards is uncerain. These resuls only sugges he need o delve deep ino he ransmission mechanism of a paricular shock bearing in mind ha var analysis should be complemened wih oher approaches. No formal es for srucural breaks were perfomed despie he presence of breaks in individual ime series. Saionariy of he esimaed favar model may sugges co-breaking, hough. Finally, an imporan limiaion of his sudy is he ime span considered. Fuure research on his opic should include a broader daa se o apply a dynamic facor model, analyze possible breaks and nonlineariies. Impac of he Moneary Policy in Uruguay 351