External Wealth, the Trade Balance, and the Real Exchange Rate

Transcription

1 Revised Version Exernal Wealh, he Trade Balance, and he Real Exchange Rae Philip R. Lane Insiue for Inernaional Inegraion Sudies, Triniy College Dublin and CEPR Gian Maria Milesi-Ferrei Inernaional Moneary Fund and CEPR This draf: December 20, 2001 Absrac We examine he link beween he ne foreign asse posiion, he rade balance and he real exchange rae. In paricular, we decompose he impac of a counry s ne foreign asse posiion ( exernal wealh ) on is long-run real exchange rae ino wo mechanisms: he relaion beween exernal wealh and he rade balance; and, holding fixed oher deerminans, a negaive relaion beween he rade balance and he real exchange rae. We also provide addiional evidence ha he relaive price of nonradables is an imporan channel linking he rade balance and he real exchange rae. JEL Classificaion: F21, F31, F41. Keywords: Ne foreign asses, rade balance, real exchange rae. * plane@cd.ie; gmilesiferrei@imf.org. This paper was prepared for he 2001 EEA/NBER Inernaional Seminar on Macroeconomics held in Dublin, June 8-9. A slighly shorer version is forhcoming in he European Economic Review in June We hank our discussans Barry Eichengreen and Richard Pores, and wo anonymous referees for useful commens. We are also graeful o Parick Honohan, Sam Ouliaris, Eswar Prasad, Andy Rose, Kees Koedijk, Maaren van Rooij, Neil Rankin, and seminar paricipans a ISOM, he European Cenral Bank, he Cenre for Cenral Banking Sudies, he Inernaional Moneary Fund, he Duch Cenral Bank, he Bank of Spain, Princeon, Columbia and he Federal Reserve Bank of New York. Mahias Hoffman provided efficien research assisance. Par of his work was carried ou while Lane was a visiing scholar in he research deparmens of he IMF and he Federal Reserve Bank of New York. Lane's work on his paper is par of a research nework on The Analysis of Inernaional Capial Markes: Undersanding Europe s Role in he Global Economy, funded by he European Commission under he Research Training Nework Programme (Conrac No. HPRN CT ). Lane also graefully acknowledges he suppor of a TCD Berkeley Fellowship and he TCD Provos s Academic Developmen Fund.

2 1. Inroducion The rapid growh in he ne exernal liabiliies of he Unied Saes and is implicaions for a possible reversal in he curren srengh of he dollar are a dominan heme of discussion in economic policy circles. This heme is cerainly no new: he debae on he relaion beween inernaional paymens and real exchange raes has a long and disinguished inellecual hisory, and was a he forefron in he lae 1920s, wih he debae beween Keynes and Ohlin on he impac of German war reparaions, in he 1970s, wih he debae on he implicaions of oil price shocks, in he early 1980s in he afermah of he deb crisis, and in he mid- and lae 1980s, wih he debae on causes and consequences of he large swings in he value of he dollar. 1 In his paper, we revisi he relaion beween inernaional ransacions, ne exernal asse posiions, and he real exchange rae, making use of a new daa se on exernal asses and liabiliies ha we recenly consruced (Lane and Milesi-Ferrei, 2001a). Our approach is empirical, and focuses on long-run relaions beween hese variables. In simple erms, he sandard argumen linking ne foreign asses, he rade balance and he real exchange rae runs as follows. A posiive seady-sae ne exernal asse posiion enables a counry o run persisen rade deficis. In urn, all else equal, he capabiliy o susain a negaive ne expor balance in equilibrium is associaed wih an appreciaed real exchange rae. Conversely, a debor counry ha mus run rade surpluses o service is exernal liabiliies may require a more depreciaed real exchange rae. In previous empirical work, Faruqee (1995), Gagnon (1996), Broner, Loayza and López (1997), Alberola, Cervero, López and Ubide (1999) and Lane and Milesi-Ferrei (2000) have provided esimaes of a posiive long-run relaion beween ne foreign asses and he real exchange rae, using a variey of daa, mehods and specificaions. 2 In his paper, we insead decompose his relaion ino wo pars: (i) he relaion beween he ne foreign asse posiion and he rade balance; and (ii) he relaion beween he rade balance and he real exchange rae. We argue ha explicily allowing for his 1 See also Krugman (1987, 1991) and Obsfeld and Rogoff (2001a). 2 Lane and Milesi-Ferrei (2000) emphasize he relaive price of nonradables as he primary endogenous componen of he real exchange rae. In conras, he oher papers rule ou any effec of exernal paymens on he relaive price of nonradables by including proxies for ha relaive price as regression conrols.

3 - 2 - decomposiion is imporan boh heoreically and empirically, and provide economeric evidence o back his argumen. Indeed, he size of he rade surplus ha a debor counry has o run o service is exernal liabiliies will depend on he rae of reurn i has o pay on hese liabiliies, as well as on is oupu growh rae. For insance, going back o he US example, a debor counry ha grows quickly and manages o earn reurns on is foreign asses ha are higher han he payous on is foreign liabiliies requires a much smaller rade surplus o sabilize is ne foreign asse posiion han a counry wih poor growh performance and unfavorable ne invesmen income flows. By exension, he magniude of any real exchange rae depreciaion will be smaller in he former case. In he empirical analysis, we provide direc evidence on how he relaion beween he rade balance and ne foreign asses depends on invesmen reurns, oupu growh and exchange rae movemens. We also highligh ha he link beween rade balance and real exchange rae depends on oher facors, such as relaive oupu per capia, relaive produciviy levels, and he erms of rade, and we provide evidence on he economic and saisical significance of long-run co-movemens beween hese variables. The empirical analysis focuses on a sample of OECD economies for he period By selecing his group of counries for which higher-qualiy daa are available, we are able o refine our empirical analysis--for insance, by direcly conrolling for produciviy variables in esimaing he long-run relaion beween he rade balance and he relaive price of nonradables. In addiion o he rade balance, our empirical findings confirm he imporance of relaive produciviy as a key deerminan of he relaive price of nonraded goods and he real exchange rae. The srucure of he res of he paper is as follows. Secion 2 discusses some analyical issues ha are imporan in inerpreing he subsequen empirical work. The relaion beween he ne foreign asse posiion and rade balance is explored in secion 3. Secion 4 esimaes long-run equaions for he real exchange rae and he relaive price of nonradables as a funcion of he rade balance, plus some conrol variables. Some concluding remarks are offered in secion 5.

4 Analyical Issues The seady sae of a sandard ineremporal open-economy model ypically involves he equilibrium equaions 3 b * = r b (1) rer = φb + λ X (2) where b is he rade balance o GDP raio, r * he rae of reurn on exernal asses and liabiliies, b is he sock of ne foreign asses as a raio o GDP, rer is he (log) CPIbased real exchange rae, and X are oher facors affecing he real exchange rae. 4 Equaion (1) jus saes ha a counry can run a seady-sae rade defici equal o he invesmen income on is ne foreign asse posiion. Equaion (2) says ha, for given values of oher facors X, he real exchange rae will be more depreciaed, he bigger he seady-sae rade surplus. Equaions (1)-(2) can be solved o yield * rer φr b λx αb λx = + + (3) Tha is, all else equal, he real exchange rae is increasing in he ne foreign asse posiion of he counry. This is he ype of equaion ypically esimaed in he empirical lieraure on he long-run relaion beween ne foreign asses and real exchange raes. However, his approach is poenially resricive for wo reasons. Firs, raes of reurn vary across counries, over ime and beween differen caegories of asses and liabiliies. 5 Second, in a non-zero growh environmen, he inrinsic dynamics of he ne foreign asse posiion depends on he oupu growh rae as well as raes of reurn. These qualificaions sugges ha we can obain exra informaion on he mechanics of he ransfer effec by separaely considering he relaions beween ne foreign asses and he rade balance and beween he rade balance and he real exchange rae. 2.1 Exernal Wealh and he Trade Balance 3 For example, see Mussa (1984), Obsfeld and Rogoff (1996, Ch. 10) and Lane and Milesi-Ferrei (2000). This sysem assumes a zero-growh seady sae. We inroduce he implicaions of non-zero growh laer in his secion. 4 An increase in rer corresponds o a real appreciaion. 5 In a sochasic environmen, ex-pos reurns will vary even if ex-ane reurns were equalized, depending on which sae of naure is realized in a given period.

5 - 4 - Accordingly, we rewrie he rade balance equaion in a more general form. If we assume for simpliciy ha he raes of reurn on exernal asses and exernal liabiliies are equal, he long-run condiion for rade balance can be wrien as (1 + r) (1 + g)(1 + e! ) b b 1 + ε Ψ + ε (4) (1 + g)(1 + e! ) whereb and b 1 are he raios of he rade balance and ne foreign asses o GDP respecively, r is he real rae of reurn on foreign asses and liabiliies in US dollars, comprising yields and capial gains, g is he real GDP growh rae of he economy, and e! is he rae of real exchange rae appreciaion vis-à-vis he US dollar. The erm muliplying b 1 can also be wrien as ( i γ) /(1 + γ), where i is he nominal rae of reurn and γ he nominal GDP growh rae, boh in US dollars. The disurbance erm ε capures emporary deviaions from his long-run value, reflecing cyclical disurbances and shifs in he desired ne foreign asse posiion. Appendix B clarifies how equaion (4) is alered when he raes of reurn on gross asses and gross liabiliies differ. We label Ψ he adjused reurns variable: i deermines he size of he rade imbalance--as a funcion of ousanding exernal wealh, invesmen reurns, oupu growh and exchange rae movemens--ha is consisen wih a unchanging raio of ne foreign asses o GDP. In he long run, we should observe an inverse relaion beween he ne foreign asse posiion and he rade balance if he rae of reurn exceeds he growh rae ( i > γ ). 2.2 The Trade Balance and he Real Exchange Rae We measure he real exchange rae as he raio of consumer prices in he home counry relaive is rading parners, expressed in a common currency. The reduced-form specificaion for he real exchange rae equaion is as follows rer = q( b, yd, ) + µ q < 0, q > 0, q > 0 (8) b yd where an increase in he (log) real exchange rae index rer corresponds o a real appreciaion, yd is he log of relaive GDP per capia, is log of he erms of rade and µ is a disurbance erm.

6 - 5 - We will also empirically examine he (log) relaive price of nonradables p p = p( b, yc, rprod, ) + ν p < 0, p > 0, p > 0, p > 0 (9) N T i i i i i b yc rprod where yc is log GDP per capia, rprod is he log of he raio of secoral labor produciviy in he radable secor versus he nonradable secor and ν is a disurbance erm. A variey of open-economy models can generae specificaions such as equaions (8)-(9). Especially for a small open economy, he primary endogenous componen of he real exchange rae is he relaive price of nonraded goods. 6 A weak real exchange rae in a counry ha runs a persisen rade surplus may be aribued o several facors. 7 Firs, he negaive wealh effec of mainaining absorpion below producion lowers demand for nonradables. 8 Second, his negaive wealh effec also poenially raises labor supply, reducing coss in he nonraded secor. Third, a decline in he relaive price of nonradables may also be useful in providing incenives for mobile facors o shif from he domesic o he expor secor. These forces all poin o an inverse relaion beween he rade balance and he relaive price of nonradables. 9 I should be noed ha here is also a ransiional relaion beween hese variables in porfolio balance models: a counry ha mus run surpluses o converge o is desired long-run ne foreign asse posiion will experience real depreciaion along he ransiion pah (see Branson and Henderson 1985 and Alberola e al 1999). 10 More generally, various business cycle shocks will generae shor-erm comovemens beween he rade 6 As emphasized by Obsfeld and Rogoff (2001b), here is a srong parallel beween wo-secor models wih raded and nonraded goods and models in which all goods are radable bu ranspor coss are significan and here is limied subsiuabiliy beween home and foreign goods. The reader may prefer his alernaive framework, according o ase. 7 Lane and Milesi-Ferrei (2000) describe a simple dynamic general equilibrium model ha generaes a negaive seady-sae effec of he rade balance on he relaive price of nonradables. The model also idenifies he erms of rade and relaive oupu per capia as addiional deerminans of he long-run real exchange rae. We reurn o hese variables laer in his secion. 8 The wealh effec of a given ne foreign asse posiion is capured by he addiional permanen income i generaes. In urn, i is he level of hese invesmen reurns (inclusive of ne capial gains) ha deermines he long-run rade imbalance ha a counry can mainain. 9 I is well undersood ha here exis a range of condiions under which he relaive price of nonradables depends only on relaive secoral produciviy (see Obsfeld and Rogoff 1996, chaper 4). Some deparure from hese condiions is required for he rade balance effec o be operaive. 10 Alberola e al. only esimae he long-run relaion beween he ne foreign asse posiion and he real exchange rae and do no empirically address he ransiional feaures of he porfolio balance model.

7 - 6 - balance and he real exchange rae. However, our focus in his paper is on he long-run relaion beween hese variables. For counries wih marke power in inernaional markes, rade imbalances may also affec he srucure of inernaional relaive prices. For insance, a rade defici may be associaed wih a srenghening of he exernal erms of rade, since an increase in he price of expors relaive o impors could accompany a conracion in ne expors. Alhough here is no necessarily a mechanical connecion beween he erms of rade and he CPI-based real exchange rae, a erms of rade improvemen will lead o real appreciaion if here is a home bias in he radables consumpion baske and/or he associaed posiive wealh effec raises demand for, and reduces he supply of, nonradables. Due o he laer mechanism, erms of rade movemens are a poenial source of shifs in he real exchange rae. Alhough he erms of rade may in par be endogenously deermined for larger counries, exogenous erms of rade movemens are likely o predominae for smaller counries. Accordingly, we direcly conrol for he erms of rade in our empirical work, so ha any relaion beween he rade balance and he real exchange rae in our specificaion will no be operaing via he erms of rade channel. In addiion o he erms of rade, we also conrol for he impac of relaive oupu per capia. I is well undersood ha an improvemen in produciviy in he raded secor relaive o he nonraded secor can generae an increase in he relaive price of nonradables by driving up economy-wide real wage levels: his is he Balassa- Samuelson effec. We will direcly conrol for produciviy for he subsample of counries for which secoral produciviy daa are available. For he broader panel of counries, we employ relaive oupu per capia as a conrol variable. In par, his may proxy for he relaive produciviy effec since oupu per capia and relaive raded-secor produciviy are likely o be posiively correlaed. However, relaive oupu per capia may also exer addiional influences on he real exchange rae. By he same mechanisms described above for he wealh effec of a posiive ne foreign asse posiion (via he capabiliy o run persisen rade deficis), a rise in oupu per capia may be associaed wih an increase in demand for nonradables

8 - 7 - and a conracion in labor supply. 11 In addiion, if ases are non-homoheic and he income elasiciy of demand for nonradables is greaer han one, his may furher apply upward pressure on he relaive price of nonradables (Bergsrand 1991). I is also worh noing ha conrolling for relaive oupu per capia serves a useful addiional purpose in seeking o capure he long-run relaion beween he rade balance and he real exchange rae, since relaive oupu per capia conrols for any effecs of hisorical curren accoun imbalances on he level of oupu via pas effecs on he pah for domesic invesmen. 3. Exernal Wealh and he Trade Balance In his secion, we examine he relaion beween he balance on goods and services, he ne exernal posiion, is composiion, and he adjused reurns erm Ψ. Our sample spans he period and includes 20 OECD counries. 12 The daa on he rade balance come from he IMF s Balance of Paymens Saisics and refer o he balance of goods, services and ransfers. 13 We use wo sources of daa for ne foreign asses: he esimaes we consruced in Lane and Milesi-Ferrei (2001a) ( adjused cumulaive curren accoun ), which are available for he period , and he Inernaional Invesmen Posiion daa (IIP) repored in he IMF s Balance of Paymens Saisics, which are generally available for a shorer period. Appendix A provides more deails on daa sources and definiions for all variables. The raio of nominal invesmen reurns o GDP is calculaed as he sum of ne invesmen income and ne capial gains on ousanding exernal asses and liabiliies 11 Noe ha, using relaive oupu per capia should be similar o using relaive income per capia, since we are holding fixed he rade balance (which we argue in he long-run is negaively relaed o ne invesmen income flows, he primary difference beween income per capia and oupu per capia). 12 The sample here excludes Belgium, because he balance of paymens daa refer o Belgium and Luxembourg bu he ne foreign asse daa o Belgium only, and Ireland, a counry for which measuremen of he rade balance and ne invesmen income in he 1990s is srongly affeced by ransfer pricing issues. Daa availabiliy for he balance of goods and services sars, for some counries, in he mid-1970s. 13 We employ here he rade balance inclusive of curren ransfers since a rade surplus and an inward ransfer are equivalen ways o finance a given level of deb service paymens. As a pracical maer, he inclusion of ransfers maers mosly for Greece. In he real exchange rae secion, we jus use he balance on goods and services since i is he difference beween domesic producion and absorpion ha should maer for he real exchange rae.

9 - 8 - measured in US dollars, divided by GDP in US dollars ( ib / 1 Y). 14 To calculae real reurns, we subrac he impac of US inflaion on he ousanding socks of ne foreign asses. The adjused reurns erm is calculaed as he difference beween he raio of nominal reurns o GDP ib / 1 Y and he impac of GDP growh on he raio of ousanding ne foreign asses o GDP γ B / 1 Y. Equaion (4) in he paper posulaes he exisence of a long-run relaion beween he rade balance and he adjused reurns erm Ψ along he ime series dimension. Table 1 shows ha ha boh variables are nonsaionary bu ha he residual from equaion (4) is saionary. 15 Tha is, a coinegraion es wih (1, -1) coefficiens imposed (in line wih he heory) is easily acceped. Pu differenly, we find ha he wo componens of changes in he raio of ne foreign asses o GDP (he rade balance and adjused reurns ) are individually non-saionary bu he change in he ne foreign asse posiion is saionary. 16 We urn o he cross-secional dimension in Table 2. Our dependen variable is he rade balance averaged over he period (columns (1), (3), (5)) and (columns (2), (4) and (6)). To check for he robusness of our resuls, we use IIP daa as our measure of ne foreign asses for he period whenever possible. 17 Our explanaory variables are he sock of ne foreign asses a he beginning of he sample period, he average adjused reurns erm and is componens. From columns (1) and (2), i is clear ha here is no cross-secional relaion beween he iniial ne foreign asse posiion and he subsequen average rade balance. However, he relaion beween he 14 When we use he adjused cumulaive curren accoun measure of ne foreign asses we can only esimae capial gains and losses on FDI and porfolio equiy asses and liabiliies, bu no on deb insrumens. The Appendix describes how such capial gains and losses are esimaed. When we use he IIP measure of ne foreign asses we can calculae capial gains and losses on all componens of ne foreign asses (as he difference beween he change in he sock and he underlying flow). 15 All panel uni roo and coinegraion ess in he paper were performed using he NPT 1.2 package for GAUSS ha has been developed by Kao and Chiang (2001). 16 Our goal here is o highligh he long-run relaion beween he rade balance and adjused reurns. In urn, adjused reurns depend on he long-run ne foreign asse posiion (plus raes of reurn and growh raes). In Lane and Milesi-Ferrei (2001b), we have explored he deerminans of he ne foreign asse posiion (including relaive oupu levels, fiscal posiions and demographic variables), plus heir conribuion o he dynamic shor-run adjusmen of he rade balance owards is long-run value. 17 We use he IIP daa for Ausria, Canada, Finland, Germany, Ialy, Japan, Neherlands, Spain, Sweden, Swizerland, Unied Kingdom, Unied Saes. For he remaining counries, IIP daa is available only for a shorer period or no available a all. Resuls are analogous if we use our own measure of ne foreign asses for all counries for he same period.

10 - 9 - average rade balance and he adjused reurns variable is close o one-o-one: counries wih posiive adjused reurns run rade deficis, while counries wih negaive adjused reurns run rade surpluses. To our knowledge, he only oher sudy ha has looked a wheher flow measures respond o he iniial exernal sock posiion is Chinn and Prasad (2000) who conduc a cross-counry sudy of he deerminans of medium-erm curren accoun balances. Their findings for indusrial counries sugges ha he iniial sock posiion is posiively correlaed wih subsequen curren accoun balances along he cross-secional dimension, while here is no relaion wihin counries beween iniial ne foreign asses and subsequen curren accoun balances. Since he curren accoun is he sum of he rade balance and ne facor income, our cross-secional finding of no relaion beween he iniial ne foreign asse posiion and he rade balance can be reconciled wih a posiive relaion wih he curren accoun, since ne invesmen income is obviously posiively relaed o ne exernal wealh. In he ime-series analysis in Table 1, we found he firs difference of he ne foreign asse posiion o be saionary and coinegraion o exis beween he rade balance and adjused reurns. Once again, his resul is consisen wih he lack of a ime-series relaion beween iniial ne foreign asses and he curren accoun repored by Chinn and Prasad. The cross-secion relaion beween he average adjused reurns and he average rade balance is also illusraed graphically in Figure 1. Over he period, counries ha enjoyed posiive adjused reurns (such as he US, UK and Greece) ran average rade deficis; conversely, hose counries ha on ne were paying ou adjused reurns (such as he Neherlands, Canada, Denmark and Finland) ran average rade surpluses. Figure 2 shows insead he relaion (or lack hereof) beween he average rade balance over he period and he sock of ne foreign asses a he end of 1982: differences in raes of reurn and growh raes means ha he cross-secion relaion beween ne foreign asses and he rade balance is weaker han he relaion beween adjused reurns and he rade balance. When we break down he adjused reurns elemen ino is underlying componens in columns (4) and (5) of Table 2, we find ha boh he real reurn and he growh componen are highly significan and have a coefficien ha is saisically no differen

11 from minus one. Tha he ne foreign asse posiion operaes only via he adjused reurns erm is perhaps no surprising in he cross-secion dimension: for insance, counryspecific mean reversion in ne foreign asses is more plausible han reversion across counries o a common mean. The nex logical sep in he analysis is an invesigaion of differences in raes of reurn beween differen counries, differen periods and differen foreign asse classes. However, as Appendix B makes clear, aemping o infer he ne raes of reurn i or r direcly (raher han muliplied by he raio of lagged ne foreign asses o GDP) is an exercise fraugh wih problems, which can be especially severe for counries ha have ne foreign asse posiions no far from zero. The reason is ha ne invesmen income is given by he difference beween reurns on exernal asses and on exernal liabiliies, and such reurns will no in general be equal. 18 As a resul, i is possible for a counry o have nonzero ne invesmen income or reurns even when is ne foreign asse posiion is zero. Simply dividing ne invesmen income by he lagged ne foreign asses when he denominaor approaches zero gives clearly biased resuls. The problem can be addressed if we can esimae wih confidence he gross socks of exernal asses and liabiliies, and herefore he raes of reurn on exernal asses and exernal liabiliies separaely. Wih he ne foreign asse daa we consruced we have esimaes of he gross posiions only for porfolio equiy and FDI, bu for he deb componen we can only provide esimaes of he ne posiion. However, he IIP daa provide measures of all gross asses and liabiliies, albei for a smaller se of counries and a shorer period of ime. In Table 3 we use his daa for he period o cas furher ligh on our regression resuls. We repor he iniial ne foreign asse posiion, is period average, he rade balance, adjused reurns, real reurns (all as raios o GDP), median real raes of reurn on exernal asses and exernal liabiliies for hose counries ha have IIP daa available for a sufficienly long period of ime (a leas 10 years), and finally he average rae of growh and real appreciaion. We presen hese findings wih a noe of cauion: measuremen error issues are pervasive and differen mehodologies in 18 The Unied Saes had ne exernal liabiliies bu posiive ne invesmen income during mos of he 1990s.

12 calculaing he value of exernal asses and liabiliies in IIP daa complicae cross-counry comparisons. One of he mos sriking sylized facs emerging from he Table is he high median real rae of reurn on exernal asses and liabiliies (nominal dollar reurns minus US inflaion) for mos counries (mean and median are above 6 percen). The main facor behind his resul is he impac of capial gains on FDI and equiy holdings--indeed, he mean and median values of real yields (no repored) are around 4 percen. We discuss furher he impac on reurns of equiy capial gains and losses below. 19 Debor counries such as Canada, Finland, New Zealand, Spain and Sweden have negaive average adjused reurns (column (4)) and negaive average ne foreign asses (column (1)), suggesing a posiive i γ. Conversely, he adjused reurns erm is posiive for debor counries such as Greece, Porugal and he Unied Saes and negaive for credior counries such as Germany and he Neherlands, suggesing ha i γ is negaive. However, if we look a columns (6)-(9) i is clear ha he raes of reurn on exernal asses and liabiliies are higher han he combined growh/real appreciaion effec, mos clearly in Japan, he Neherlands and he Unied Saes. Wha is a work here is a rae of reurn differenial beween exernal asses and liabiliies (posiive for he US, negaive for Japan and he Neherlands) which makes he average raio of real reurns o GDP (and he adjused reurns erm) posiive for a debor counry like he US and negaive for credior counries such as Germany, Japan and he Neherlands. 20 Bu wha facors accoun for he high measured raes of reurn on US asses and liabiliies? Capial gains on FDI and equiy holdings are he main facor. Unlike mos oher counries, he US calculaes direc invesmen asses and liabiliies (and no only porfolio equiy holdings) a marke value, raher han book value. This implies ha he rapid increase in sock marke values worldwide during he 1990s had a paricularly srong impac on raes of reurn. Indeed, if we ne ou capial gains and losses on FDI and 19 Unforunaely, wih he excepion of a few counries such as he US and he UK, we do no have a breakdown on invesmen income flows by asse caegory, and canno herefore calculae raes of reurn on differen asse classes. 20 See Appendix B for he expression corresponding o he raio of real reurns and adjused reurns o GDP when here are raes of reurn differenials beween asses and liabiliies.

13 equiy insrumens from reurns, he reurn on US asses drops o 5.3 percen and he reurn on US liabiliies o 3.4 percen. If we repea he same exercise for he oher counries in our sample, raes of reurn are also reduced, albei less drasically han in he US: for example, he reurns on UK asses and liabiliies drop o 4.5 and 4 percen, respecively. As shown in column (5), only wo counries have posiive real reurns on average during he sample period. In an accouning sense, his is explained by he fac ha only four counries had posiive ne foreign asses hroughou he period (Germany, Japan, Neherlands, and Swizerland) and all of hem had higher raes of reurn on liabiliies han on asses. More generally, measuremen error problems are more severe for reurns on asses han on liabiliies, because of underreporing of invesmen income earnings on asses. 4. The Real Exchange Rae, he Relaive Price of Nonradables and he Trade Balance: Empirics Having esablished a link beween he ne foreign asse posiion and he rade balance, he objecive of he second par of our empirical exercise is o capure he long-run relaion beween he rade balance and he real exchange. We begin by examining he ime series properies of he daa in Table 4. We consruc a mulivariae CPI-based real exchange rae and he same rade weighs are employed o consruc relaive oupu per capia as he difference beween domesic GDP per capia and he weighed average of he GDP per capia of each counry s main rading parners (see Appendix A for more deails on he definiions and sources of variables). The panel uni roo ess all indicae ha a null of saionariy can be rejeced for all he variables we consider in his secion. However, for boh he real exchange rae and he relaive price of nonradables specificaions (equaions (8) and (9) above), coinegraion ess indicae ha a saionary long-run relaion exiss among he variables Furher deails on he uni roo and coinegraion ess are available from he auhors upon reques. As is well known from he PPP lieraure, ess of he null of non-saionariy offer more mixed resuls bu he null of saionariy is more relevan for our purposes.

14 Having esablished ha a coinegraion framework is appropriae, esimaion is by dynamic ordinary leas squares (DOLS(-1,1)). 22 Accordingly, he general form of he panel regression for he real exchange rae is rer = α + φ + β * b + β *log( yd ) + β *log( ) + ρ b i i 1 i 2 i 3 i k= 1 1k i k k= 1 k= 1 ρ 1 2k log( ydi k) ρ 1 3k log( i k) µ k= k= i (7) k= 1 where α i and φ are counry and ime fixed effecs. 23 Including he leads and lags of he firs differences of he regressors improves efficiency in esimaing he long-run coefficiens β1, β2, β 3. The β coefficiens capure he relaion beween he long-run values of he regressors and he long-run real exchange rae. 4.1 The Real Exchange Rae The resuls for he real exchange rae equaion are shown in Table 5. In columns (1)-(2), he full panel is employed. The sample is hen spli beween he non-g3 and G3 counries in columns (3)-(4) and (5)-(6) respecively. I is naural o expec a difference in he sensiiviy of he real exchange rae o various fundamenals beween large and small counries: for insance, he relaive size of he nonraded secor ypically varies direcly wih he size of he counry. In all cases, counry fixed effecs are employed: hese are necessary since he real exchange rae daa are index measures and herefore no comparable across counries. We repor resuls boh wih and wihou ime fixed effecs. The rade balance eners significanly in all specificaions in Table Taking he specificaion ha includes ime dummies, he rade balance coefficien for he full panel is However, he spli beween he non-g3 and G3 subsamples reveals a large difference in magniude. For he non-g3 counries, a 3 percenage poin increase in he rade surplus as a raio o GDP is associaed wih only a 1 percen real depreciaion. For 22 Sock and Wason (1993) originally developed he DOLS esimaor. See Mark and Sul (1999) and Kao and Chiang (1999) on is performance as an esimaor of long-run relaions in panel daa. In urns ou ha an ARDL specificaion generaes very similar resuls for our sample. 23 A similar specificaion is employed for he relaive price of nonradables. 24 Noe ha our variable is he nominal rade balance as a raio o GDP. Since we conrol for he erms of rade in he regression, we capure he effec of he real rade balance.

15 he G3 counries, by conras, he same improvemen in he rade balance is associaed wih a 19.3 percen real depreciaion. A similar sory applies for he role played by relaive oupu per capia: in all specificaions, is relaion wih he real exchange rae is significanly posiive bu he poin coefficien is en imes larger for he G3 han for he non-g3 counries --- a 10 percen increase in relaive oupu per capia is associaed wih less han a 2 percen real appreciaion for he non-g3 counries bu a 19 percen real appreciaion for he G3 counries. Finally, he erms of rade is significanly posiive for he full panel and he non- G3 counries bu is insignifican for he G3 counries. Counry-by-counry esimaion reveals ha he laer resul is aribuable o he Japanese daa: he counry resuls for he US and Germany indicae a uniary coefficien for he erms of rade in he real exchange rae equaion. For he non-g3 counries, he erms of rade coefficien is 0.52: a 10 percen improvemen in he erms of rade is associaed wih a 5.2 percen real appreciaion. By holding fixed he erms of rade, we emphasize ha he observed relaion beween he rade balance and he real exchange rae mus be operaing hrough some oher channel, such as he relaive price of nonradables. Overall, he resuls in Table 5 provide broad suppor for an inverse long-run relaion beween he rade balance and he real exchange rae, holding fixed relaive oupu per capia and he erms of rade. As a coinegraion equaion, hese esimaes are consisen even if here is correlaion beween he regressors and he error erm. 25 We can go furher and inerpre he equaion as quasi-srucural in he sense of Davidson (1998), if he long-run real exchange rae does no play a role in deermining he long-run rade balance. Alhough here is considerable shor-erm feedback beween he real exchange rae and he rade balance, he long-erm rade balance should depend on hose facors ha drive he long-run ne foreign asse posiion. 26 The lieraure on he deerminans of ne foreign asses poins o facors such as demography, fiscal posiions and relaive incomes: we know of no heory ha generaes a direc dependence of he long-run ne foreign asse posiion on he long-run value of he real exchange rae (see also Lane and 25 See Hayashi (2000) for a review of he properies of coinegraion esimaors. 26 Of course, he deerminans of he long-run growh rae and raes of reurn also maer.

16 Milesi-Ferrei 2001). Raher he real exchange rae adjuss in he long run as an equilibrium response o he level of he rade balance ha is associaed wih he desired long-run ne foreign asse posiion. We noe ha our esimaed long-run equaion for he real exchange rae could form he basis for calculaing fundamenal equilibrium real exchange raes. In paricular, we would argue ha he rade balance is a key variable ha should be included in such an exercise. However, his sep would also require an assessmen of he appropriae equilibrium long-run values for he explanaory variables in he equaion, which is beyond he scope of his paper (see Driver and Wesaway, 2001 for a recen review). In resuls available from he auhors, we also ran counry-by-counry regressions. The poin esimae on he rade balance is negaive in 19 ou of 21 cases: all else equal, real exchange raes end o be more depreciaed, he larger is he rade balance surplus. 27 As is illusraed in Figure 3, a negaive relaion beween counry size and he magniude of he rade balance coefficien is clearly eviden in he daa (he correlaion is 0.46). 28 This is o be expeced: in sandard open-economy models, he deerioraion in he real exchange rae ha is associaed wih a given improvemen in he rade balance is direcly relaed o he relaive size of he nonraded secor in he economy. 4.2 The Relaive Price of Nonradables We urn now o examining wheher we can direcly observe a relaion beween he rade balance and he relaive price of nonradables, even when we direcly conrol for he impac of differenial relaive secoral produciviy growh. We build proxies for secoral produciviy and secoral prices from he OECD s Inersecoral Daabase (ISDB): his provides sufficien daa for hireen of he counries in our sample. 29 The manufacuring secor is aken o represen he radable secor; he nonraded secor is proxied by an 27 Wih respec o he oher real exchange rae deerminans, here is a posiive relaion beween relaive oupu and he real exchange rae for 14 counries (9 significan): counries ha experience a relaive improvemen in oupu per capia end o experience real appreciaion. Finally, he erms of rade eners wih a posiive sign in 18 ou of 21 cases (11 significan). 28 Counry size is measured by oal GDP in 1990, in consan US dollars. 29 These are he Unied Saes, Japan, Germany, France, Ialy, he Unied Kingdom, Canada, Denmark, Finland, Norway, Ausralia, Belgium and Sweden. The Neherlands is also in he daabase bu he daa coverage is very poor.

17 aggregae of consrucion, communiy, social and personal services, and producers of governmen services. 30 We follow Canzoneri, Cumby and Diba (1999) in aking labor produciviy (value added divided by he number of employees) as he appropriae produciviy variable. Secoral prices are measured by he value added deflaors. The relaive price of nonradables is he dependen variable in Table 6. We consider a slighly alered se of regressors: in addiion o he produciviy variable, our focus on he relaive price of nonradables domesically means ha we employ here GDP per capia ( yc i ) raher han relaive GDP per capia ( yd i ). As in Table 5, columns (1)- (2) employ he full panel; columns (3)-(4) include only he non-g3 counries; and column (5)-(6) show he resuls for he G3 sub-panel. Consisen wih De Gregorio, Giovannini and Wolf (1994), Asea and Mendoza (1994) and Canzoneri e al (1999), relaive secoral produciviy eners very srongly in deermining he relaive price of nonradables. However, i urns ou ha he rade balance is also a significan variable in deermining he relaive price of nonradables. The esimaed poin coefficiens in columns (4) and (6) sugges ha a one percenage poin improvemen in he rade balance is associaed wih a decline in he relaive price of nonradables of 0.75 percen and 1.21 percen for he non-g3 and G3 counries respecively. Alhough here is a difference beween he non-g3 and G3 counries in hese poin esimaes, i is much smaller in magniude han in he real exchange rae regressions in Table 6. These findings can be reconciled by he fac ha he relaive price of nonradables is a more imporan componen in he consumer price level in larger, more closed economies: he same change in he relaive price of nonradables ranslaes ino a much bigger change in he CPI-based real exchange rae for a large counry han for a small counry. 31 Wih respec o he oher regressors, he oupu per capia variable should capure demand facors or wealh effecs on he supply of labor, since produciviy is now direcly 30 This selecion maximizes daa availabiliy, since oher services secors (such as financial services) are available only for a small number of counries over a narrow ime inerval. Moreover, as is poined ou by McDonald and Ricci (2001), he wholesale and reail rade secor ha is ofen incorporaed ino he nonraded bundle in fac more closely resembles a radable secor in is characerisics. 31 Following Obsfeld and Rogoff (2000a, 2000b), nonradables should be broadly inerpreed for his purpose o encompass all goods whose price is ulimaely deermined by domesic condiions.

18 included in he specificaion. In fac, he evidence on his variable is mixed: for he G3 counries, here is a significanly posiive relaion beween oupu per capia and he relaive price of nonradables bu his is no he case for he non-g3 counries. Finally, here is weak evidence ha he erms of rade posiively affecs he relaive price of nonradables (more so for he non-g3 counries). The findings in Table 6 refer o he deerminans of he domesic relaive price of nonradables. Wha maers for he evoluion of he real exchange rae is he relaive price of nonradables a home relaive o he relaive price of nonradables overseas. Accordingly, we repor resuls for his double relaive price in Table 7. By exension, we now use he difference in he log of relaive secoral produciviy beween he home counry and is rading parners (drprod) and relaive GDP per capia in logs (yd) as he relevan regressors, in addiion o he rade balance and he erms of rade. The resuls in Table 7 are broadly similar o hose in Table 6. Cross-counry variaion in relaive secoral produciviy are very imporan in explaining differenials in he relaive price of nonradables. As before, he rade balance is imporan in explaining he relaive price of nonradables relaive o rading parners, alhough i loses significance for he G3 counries if ime dummies are excluded from he specificaion. However, for he specificaion ha includes ime dummies (columns (4) and (6)), he poin esimae for he G3 counries is almos riple ha for he non-g3 counries. Alhough subsanial, his differenial is much smaller han in he real exchange rae equaions in Table 5, in line wih he differen weighs of nonradables in he consumpion baskes of large versus small counries. In conras o Table 6, (relaive) oupu per capia is significan in all specificaions in Table 7. The poin esimaes sugges ha his variable exers a bigger impac for he G3 counries han he non-g3 counries: he magniude of he coefficien is wice as large for he specificaion ha includes ime fixed effecs (columns (4) and (6)). Finally, he erms of rade are significan only for he non-g3 counries in Table 7. We may expec a more imporan role for he erms of rade in smaller (more open) economies, since he wealh effec of a erms of rade improvemen posiively depends on he volume of expor/impor aciviy. In summary, he empirical resuls in his secion esablish a link beween he rade

19 balance and he real exchange rae: in he long run, larger rade surpluses are associaed wih more depreciaed real exchange raes. Furhermore, he magniude of he rade balance coefficien is posiively relaed o counry size. Taken ogeher, he resuls of secions 3 and 4 illusrae a key mechanism by which he ne foreign asse posiion is an imporan fundamenal driver of he real exchange rae: condiional on he paerns of invesmen reurns and oupu growh, counries wih posiive ne exernal wealh are able o run persisen rade deficis; in urn, a paern of rade deficis is associaed wih longrun real appreciaion of he real exchange rae. 5. Conclusions This paper has explored he links beween he ne foreign asse posiion, he rade balance and he real exchange rae. We have shown ha he relaion beween exernal wealh and he rade balance wihin and across counries is relaed o he raes of reurn on exernal asses and liabiliies and he rae of oupu growh. Conrolling for oher deerminans, we have esablished a negaive long-run associaion beween he rade balance and he real exchange rae. Moreover, we find ha he magniude of he rade balance coefficien is increasing in counry size and we provide direc evidence ha he relaive price of nonradables co-moves wih he rade balance, even conrolling for relaive secoral produciviy. Our work has focused on long-run relaions, and has no explored he shor-run dynamics of he relaion beween he rade balance and he real exchange rae. Ineresing issues here include possible non-lineariies in he convergence of he exchange rae o is long-run value. Anoher challenge is he idenificaion of alernaive srucural shocks ha may generae differen shor-run co-movemens beween he rade balance and he real exchange rae. Finally, he paper has highlighed he imporan role played by differences in raes of reurn on exernal asses and liabiliies in shaping he dynamics of ne foreign asses. Undersanding he sources of hese differences in raes of reurn is an imporan opic on he research agenda.

20 Appendix A. Daa Sources and Definiions b : Ne foreign asses, calculaed as cumulaive curren accoun adjused for capial accoun ransfers and for valuaion changes in porfolio equiy asses and liabiliies, FDI asses and liabiliies, and foreign exchange reserves, as raio of GDP. For furher deails on he valuaion adjusmens, see Ne Capial Gains below and Lane and Milesi-Ferrei (2001a). Source: Lane and Milesi-Ferrei (2001a). IIP: Inernaional Invesmen Posiion ne of gold holdings. Source: IMF, Balance of Paymens Saisics. b: Balance on goods, services and ransfers (raio o GDP). Source: IMF s Balance of Paymens Saisics, supplemened by OECD sources. Ne Invesmen Income: Invesmen income credis invesmen income debis. Source: IMF, Balance of Paymens Saisics. Ne Capial Gains: For he NFA measure, capial gains are differences beween he esimaed change in he sock of exernal asses (liabiliies) and he underlying flow. For FDI asses, valuaion changes are he esimaed effecs on ousanding FDI asses of changes in relaive capial goods prices beween he counries of desinaion of FDI and he US (he uni of measuremen). The counries of desinaion of FDI are assumed o be he rading parners used in he calculaion of he real effecive exchange rae, and he relaive price of capial goods across counries is assumed o follow relaive consumer prices. For FDI liabiliies, valuaion changes are given by he impac on ousanding liabiliies of changes in relaive capial goods prices beween he home counry and he US, proxied by changes in he bilaeral CPI-based real exchange rae. For porfolio equiy invesmen asses, valuaion changes reflec year-on-year changes in he US dollar value of a represenaive world porfolio (he Morgan Sanley Capial Index). For porfolio equiy invesmen liabiliies valuaion changes reflec year-on-year changes in he US dollar value of he domesic sock marke index. For foreign exchange reserves, we direcly measure he difference beween changes in he sock and he underlying flow (primarily due o cross-counry flucuaions). Capial gains calculaed on IIP are differences beween he acual change in he sock and he underlying flow. Source: Lane and Milesi-Ferrei (2001a) and IMF, Balance of Paymens Saisics.

21 rre (real reurns): Ne invesmen income/gdp + ne capial gains/gdp - B π where $ 1 Y $ π is he rae of inflaion in he US (measured wih he US GDP deflaor). γ Ψ (adjused reurns): Ne invesmen income/gdp + ne capial gains/gdp - b 1, 1+ γ where γ is he nominal GDP growh rae in US dollars and b -1 is he raio of ne foreign asses o GDP a he end of period -1. Source: auhors calculaions. rer: log of real effecive exchange rae (CPI-based). Trade weighs based on rade paerns in 1990, calculaed using he IMF s Informaion Noice Sysem (described in Desruelle and Zanello (1997)). Source: auhors calculaions based on CPI and exchange rae daa from he Inernaional Moneary Fund. yd: log of GDP per capia relaive o rading parners. Parner counries weighs are he same as hose used in he consrucion of RERCPI. Source: auhors calculaions based on he World Bank s World Developmen Indicaors (WDI) daabase. : log of raio of expor prices (or expor uni values) o impor prices (or impor uni values), boh expressed in US dollars. Sources: OECD, Analyical Daabase, and IMF, Inernaional Financial Saisics. p N p : log of price of nonraded goods relaive o raded goods. The nonraded goods T secor included consrucion, communiy, social and personal services, and producers of governmen services. The raded goods secor is he manufacuring secor. Source: auhors calculaions based on OECD s Inernaional Secoral Daabase (2000). rprod: log of labor produciviy in he raded goods secor relaive o he nonraded goods secor. Source: auhors calculaions based on he OECD s Inernaional Secoral Daabase (2000). yc: log of GDP per capia. Source: The World Bank s World Developmen Indicaors (WDI) daabase. drprod: rprod in he home counry relaive o a weighed average of rprod in rading parners. Source: auhors calculaions based on he OECD s Inernaional Secoral Daabase (2000).

22 Appendix B. Consrucion of adjused reurns erm Changes in he ne foreign asse posiion are due o curren accoun imbalances and o capial gains and losses. Assume iniially ha exernal asses and liabiliies earn he same rae of reurn. In his case, he dynamics of ne foreign asses can be wrien as follows B B = TB + TR + ( i + kg ) B (B1) * 1 1 where TB is he balance on goods, services and ransfers, i * B 1 is invesmen income and kg B is he capial gain/loss on ousanding ne exernal asses. Dividing boh erms by 1 GDP measured in US dollars and re-arranging erms, we obain i + kg γ b = b + b * 1 1+ γ (B2) Le i = i + kg, define * r 1+ i i π = 1+ π 1+ $ 1 $ $ π $, and re-wrie 1 + γ = (1 + g )(1 + e! )(1 + π ) where g is he economy s real growh rae, e! he rae of real appreciaion vis-à-vis he US dollar and $ π is he US inflaion rae, where are he las wo variables are measured in erms of GDP deflaors. We can hen re-wrie equaion (B2) as follows (1 + r) (1 + g)(1 + e! ) b = b 1 + b (1 + g )(1 + e! ) (B3) which is equaion (2) in he ex. For he empirical analysis, we use boh real reurns, calculaed as differ. Le $ i π b 1+ γ 1 i γ, and he adjused reurns erm, given by Ψ = b γ Assume now ha he rae of reurn on gross exernal asses and liabiliies can gfa, gfl be he socks of gross foreign asses and gross foreign liabiliies (as A L A L raios of GDP), respecively, and define i, i ( r, r ) as he nominal (real) US$ rae of reurn on gross foreign asses and liabiliies, respecively. In his case our compued raios of ne real reurns and adjused reurns o GDP are equal o

23 r ( r r ) rre b gfl A A L = 1+ 1 (1 + g)(1 + e! ) (1 + g)(1 + e! ) (1 + r ) (1 + g )(1 + e! ) ( r r ) Ψ = + A A L b 1 gfl 1 (1 + g)(1 + e! ) (1 + g)(1 + e! ) (B4) Clearly in he presence of a non-zero r A r L, aemping o measure r as [ rre (1 + g )(1 + e! )]/ b gives biased resuls. The size of his bias grows, he closer is he 1 ne foreign asse posiion o balance and he larger he underlying gross sock posiions.