DOES FINANCIAL INTEGRATION SPUR ECONOMIC GROWTH? NEW EVIDENCE FROM THE FIRST ERA OF FINANCIAL GLOBALIZATION

Transcription

1 DOES FINANCIAL INTEGRATION SPUR ECONOMIC GROWTH? NEW EVIDENCE FROM THE FIRST ERA OF FINANCIAL GLOBALIZATION MORITZ SCHULARICK THOMAS M. STEGER CESIFO WORKING PAPER NO CATEGORY 5: FISCAL POLICY, MACROECONOMICS AND GROWTH MARCH 2006 An electronc verson of the paper may be downloaded from the SSRN webste: from the CESfo webste: Twww.CESfo-group.deT

2 CESfo Workng Paper No DOES FINANCIAL INTEGRATION SPUR ECONOMIC GROWTH? NEW EVIDENCE FROM THE FIRST ERA OF FINANCIAL GLOBALIZATION Abstract Does nternatonal fnancal ntegraton boost economc growth? The queston has been dscussed controversally for a long tme. As of yet, robust evdence for a postve mpact s lackng (Edson et al., 2002). However, there s substantal narratve evdence from economc hstory that hghlghts the contrbuton European captal made to economc growth of perpheral economes before We have compled the frst comprehensve data set to test ths hypothess. The man fndng s that there was ndeed a sgnfcant and robust growth effect. Our theoretcal explanaton stresses property rghts protecton as a prerequste for the standard neoclasscal model to work properly. JEL Code: F15, F21, F30, N10, N20, O11, O16. Keywords: nternatonal fnancal ntegraton, economc growth, frst era of globalzaton. Mortz Schularck Free Unversty of Berln Pestalozzstr Berln Germany mortz.schularck@amyacaptal.com Thomas M. Steger Insttute of Economc Research ETH Zurch Zürchbergstr Zurch Swtzerland tsteger@ethz.ch Ths verson: March 2006 We would lke to thank Mchael Clemens, Nall Ferguson, André Hofmann, Angus Maddson, Chrstopher M. Messner, Maurce Obstfeld, Sevket Pamuk, Alan M. Taylor, Marc Wedenmer and Jeffrey Wllamson for generously sharng ther data. Thorsten Beck, Hal Edson, Henrk Enderlen, Carl-Ludwg Holtfrerch, Davd Leblang, Ross Levne, Davd Roodman, and Torsten Slok provded helpful comments. We thank Lll Balluch for research assstance.

3 1 1. Introducton The nexus between nternatonal fnancal ntegraton the degree to whch an economy s ntegrated wth and open to the nternatonal captal market and economc growth contnues to be one of the most debated ssues among nternatonal economsts. Do fnancally more open economes grow faster than closed ones, precsely because of ther openness to the global captal market? Are polces sensble that promote growng nternatonal fnancal ntegraton, hence fnancal globalzaton? In ths paper, we am to shed a new lght on these questons by lookng at the evdence from the frst era of fnancal globalzaton from In a perfect neoclasscal textbook world, there are good arguments for a postve growth mpact of ntegraton wth the nternatonal captal market, especally for developng countres. By tappng the pool of global savngs captal-poor countres could free themselves of a bndng constrant on economc growth lack of captal. Closer fnancal ntegraton could also strengthen domestc fnancal systems leadng to more nvestment, more effcent allocaton of captal and hgher growth (Levne, 2001). On a global level, the effcent allocaton of captal and nternatonal rsk sharng would be promoted (Obstfeld, 1994). However, arguments aganst the economc wsdom of openness to global captal flows have also been put forward. Fnancal ntegraton does not have to be welfare enhancng n the presence of other dstortons such as trade barrers and weak nsttutons, or f nformaton asymmetres affect the proper workng of the nternatonal fnancal market (Bhagwat 1998; Stgltz, 2000). Despte a rch body of contrbutons, the emprcal lterature s stll nconclusve wth regard to the fnancal ntegraton-growth nexus. Emprcal work by Grll and Mles-Ferrett (1995), Kraay (1998), Edson et al. (2002) and Fratzscher and Bussère (2004) has not confrmed a robust long-term mpact of fnancal openness on growth. Ther results have mrrored the early and well-known study by Rodrk (1998, p. 9) who concluded that captal controls are essentally uncorrelated wth long-term economc performance. On the other sde of the spectrum are studes that fnd support for a relatonshp between openness to the global captal market and economc growth such as Qunn (1997), Bekaert (2001), and Edwards (2001). More recent research has amed to throw more lght on the queston whether the postve growth mpact of fnancal ntegraton depends on thrd factors such as a sound nsttutonal framework, but the results remaned mxed at best (Edson et al., 2002; Klen, 2005). Detaled revews of the lterature on fnancal openness and growth have been gven by Echengreen (2002) and Edson et al. (2004), we thus content ourselves wth ths bref revew. A balanced summary of emprcal research on the ssue has been gven n a study by the research department of the Internatonal Monetary Fund (IMF), one of the man proponents of captal account lberalzaton n the 1990s: Theoretcal models have dentfed a number of channels through whch nternatonal fnancal ntegraton can promote economc growth n developng countres... However, there s as yet no clear and robust emprcal proof that the effect s quanttatvely sgnfcant. (Prasad et al., 2003, p.1) One reason why emprcal research on the fnancal ntegraton-growth lnk remaned nconclusve to date s that dfferent approaches and econometrc technques made t dffcult to synthesze the results. Although the maorty of papers took smlar (albet not dentcal) cross-country growth models as ther startng pont, marked dfferences remaned wth regard to the sample of countres, the perod under nvestgaton and the estmaton technques

4 2 employed. 1 Not least, the measurement of the key varable, namely the degree of fnancal openness, was controversal. Whle some authors looked at the presence or absence of legal restrctons on fnancal transactons, others have opted to measure fnancal ntegraton va the actual extent of captal flows. It s for the sake of comparablty that we are ntentonally conservatve throughout ths paper wth regard to changng the underlyng growth model, ntroducng new estmaton technques or a new measure for fnancal ntegraton. On the contrary, we ntentonally rely on models and technques employed before and am to ensure comparablty of our results wth prevous studes. Ths s because the most mportant nnovaton of ths paper s n a dfferent feld: we am to benchmark the present to the past. Economc hstorans have often underscored the contrbuton that nternatonal captal flows made to economc growth n developng countres durng the frst era of globalzaton the years of the classcal gold standard from Yet t has never been tested econometrcally for a broad crosssecton of countres whether the frst era of fnancal globalzaton does ndeed provde evdence that fnancal ntegraton can spur growth. The man queston ths paper tackles reads as follows: Has nternatonal fnancal openness promoted growth n the frst era of global fnance? We have put consderable effort nto assemblng the largest possble dataset for the years It covers 24 developng and developed countres. We use captal nflows from the UK (n relaton to GDP) as a proxy for the degree of fnancal ntegraton of ndvdual countres. These detaled captal flow data are avalable from a recently publshed analyss of the geographcal patterns of stock and bond ssues at the London Stock Exchange (Stone, 1999). The new dataset allows us to show that n stark contrast to the ambguous fndngs n contemporary research nternatonal fnancal ntegraton had a statstcally sgnfcant effect on growth n the frst era of global fnance. In addton, we can exclude that ths fndng s purely drven by dfferent estmaton technques or model specfcaton, because we frst run our regressons on a contemporary dataset wth the am of reproducng the results of what we consder the most comprehensve contemporary study (Edson et al., 2002). In a second step we run the dentcal model wth the same econometrc methodology on our newly collected hstorcal dataset. It s thus the data, not dfferent model specfcatons or econometrc technques, that lead us to conclude that the frst era of fnancal globalzaton proves a postve relatonshp between nternatonal fnancal ntegraton and economc growth. Our core fndngs support all these economsts who beleve n the vrtues of nternatonal captal moblty and, ncdentally, the professon of economc hstorans who have for a long tme ponted to the mportant role of foregn captal for growth n the perphery before WW1. In other words, the late 19 th and early 20 th centures experence demonstrates that nternatonal fnancal ntegraton can contrbute to hgher growth. Ths opens up a potentally frutful roadmap for future research whch we brefly dscuss n the last part of the paper: Why dd fnancal openness promote growth back then but no longer today? Our study s related to two strands n recent research n nternatonal economcs. Frst, we are methodologcally ndebted to the rch contemporary lterature that poneered emprcal nvestgaton of the fnancal ntegraton-growth nexus. In partcular, we rely on the work of Edson et al. (2002). Our study also relates to Echengreen and Leblang (2003) who analyzed the relatonshp between fnancal openness and growth over the entre 20 th century. Yet whle Echengreen and Leblang focus on the long-run effects of fnancal openness over dfferent nternatonal polcy regmes, we contrast two perods of hgh nternatonal fnancal 1 See the dscusson n Edson et al. (2002).

5 3 ntegraton. Second, we take nspraton from recent research on globalzaton n hstorcal perspectve (Bordo et al., 2003; Obstfeld and Taylor, 2004; Ferguson and Schularck, 2005), a recent strand of research whose man dea s to explore the frst epsode of hgh nternatonal captal moblty wth an eye on polcy lessons for today. The structure of ths paper s as follows. In the followng secton, we brefly present the emprcal strategy and ntroduce our new dataset. The thrd and fourth part present and dscuss the emprcal results for the contemporary and hstorcal perods. In the ffth secton we present a prelmnary explanaton for the dverse mpact of captal market ntegraton n hstorcal and modern tmes. The last secton summarzes the man fndngs and concludes. 2. Emprcal Strategy, Conceptual Issues and Data The overall emprcal strategy of ths paper runs as follows: We use a newly collected hstorcal dataset to test whether there s emprcal evdence that fnancal ntegraton translated nto hgher growth n the frst era of fnancal globalzaton. To arrve at fully comparable results wth contemporary studes on the fnancal ntegraton-growth nexus, we use the same models and econometrc technques as the recent lterature. We run dentcal growth regressons both on a contemporary ( ) and a newly assembled dataset for the frst era of fnancal globalzaton ( ). We also algn our emprcal analyss to the most comprehensve contemporary study (Edson et al., 2002). In addton, we test the senstvty of our results across two dfferent growth models found n the recent lterature. There s substantal narratve evdence from economc hstory that hghlghts the contrbuton European captal made to economc growth of perpheral economes before 1914 (Fes, 1965; Woodruff, 1966). The degree of nternatonal fnancal ntegraton reached before 1914 was truly mpressve. In the decades before WW1, gross foregn nvestments n relaton to gross domestc product (GDP) n 1913 stood at about 200 percent n Argentna, Chle and South Afrca, and at or above 100 percent n countres such as Brazl, Mexco, Egypt, and Malaysa actually about twce as hgh as the correspondng fgures at the end of the 1990s (Twomey, 2000). Not only North and South Amerca were well ntegrated nto the nternatonal captal market. Southern and Eastern Europe, Afrca and Asa all attracted consderable amounts of captal (Stone, 1999). European nvestors fnanced Amercan ralroads, Argentnean farms, sewerage systems n the Mddle East, ports n Asa and telegraph networks n Afrca. From the hstorcal narratve, t would appear that ntegraton nto the global captal market could have been an mportant growth drver. But does ths narratve stand up to detaled econometrc nvestgaton of a broad cross-country sample? Contemporary research on the growth effects of nternatonal fnancal ntegraton has typcally regressed the growth rate of real per capta GDP growth on a measure for the degree of nternatonal fnancal ntegraton plus a vector of control varables whch proxy fundamental growth drvers such as human captal and ntal GDP per capta as a neoclasscal convergence term. However, the models employed n prevous studes dffered n two mportant respects: On the one hand wth regard to the measures for the degree of fnancal ntegraton, on the other wth regard to the vector of control varables. Both ssues need to be addressed brefly. Fnancal ntegraton or fnancal openness as t has also been called n analogy to openness to trade n goods has on the one hand been measured by the extent to whch legal barrers mpede the free flow of captal (Qunn, 1997; Rodrk, 1998). On the other hand, along the lnes of the emprcal lterature on trade openness and growth n whch trade openness s typcally measured by the value of traded goods and servces over GDP one can argue that fnancal openness should be measured quanttatvely. Kraay (1998) and Edson et al. (2004)

6 4 have looked at varous measures of gross captal flows and stocks over GDP as quanttatve ndcators for the degree of nternatonal fnancal ntegraton. Echengreen (2001) and Edson et al. (2004) dscuss the advantages of both approaches. Clearly, the choce of the ndcator s not only a queston of convenence and data avalablty. For example, a country may operate captal controls, but they could be leaky or selectve so that despte formal legal barrers, the actual degree of nternatonal fnancal ntegraton could be qute substantal. Usng a quanttatve measure for the degree of ntegraton would n ths case seem to mprove greatly as t would show a hgh degree of openness to nternatonal captal. However, n ther comprehensve study Edson et al. (2002) test vrtually all avalable ndcators, rule-based as well as quanttatve, but fnd no robust evdence for a postve growth effect of ether one for the perod 1980 to In ths study we use a measure that was tested (wthout clear results for the perod) by Edson et al. (2002) gross nflows of drect foregn nvestment and portfolo flows over GDP. It would seem that ths ndcator s very lkely to capture the potental growth effects of captal nflows. But our choce s also data-drven. For the frst perod of globalzaton gross nflows of captal are the only varable for whch we possess relable data, and formal captal controls were unheard of n ths perod. The second man ssue n whch contemporary studes have dffered relates to the specfcaton of the emprcal model. Some authors have argued that short-term polcy varables lke the budget defct and the nflaton rate need to be ncluded (Edson et al., 2002). Others have opted to control for a smaller set of long-run determnants of economc growth mrrorng the results of Barro and Sala-I-Martn (1992) and the robustness analyses by Levne and Renelt (1992). 2 In partcular, the ncluson of the nvestment rato has proved problematc. 3 Klen and Olve (2001), Edson et al. (2002), McLean and Shrestsha (2002), Echengreen and Leblang (2003), Obstfeld and Taylor (2003), Bekaert et al. (2004) dd not nclude t, on grounds of potental endogenety. Other authors such as Rodrk (1998), Arteta et al. (2001), Edwards (2001) and Klen (2005) explctly control for dfferent nvestment ratos at the begnnng of the observaton perod. We follow an ntermedate strategy and employ three dfferent emprcal models. The frst excludes the nvestment rato, whereas the second and the thrd nclude t, controllng for ts potental endogenety. Snce the nvestment rato s one of the most mportant explanatory determnants of long-run growth, we hold that t should be ncluded, at least n some specfcatons, to test the senstvty of the results. To make our fndngs ndependent from potentally parsmonous specfcatons, we specfy three dfferent models: Model (I) s an exact reproducton of the benchmark regresson of Edson et al. (2002),.e. we regress real per capta growth on ntal ncome, average years of schoolng (proxyng human captal), average consumer prce nflaton and budget defcts, plus the perod average of captal nflows to GDP as a measure of nternatonal fnancal ntegraton. Model (II) s dentcal to (I), but adds the nvestment rato. Model (III) drops the two polcy varables (nflaton and budget defct) from (II), but adds populaton growth. It thus relates most closely to the fundamental growth regressons of Barro and Sala-I-Martn 2 It can be argued that addtonal polcy varables are approprate f nternatonal fnancal ntegraton s measured on a quanttatve ( de facto ) bass as the polcy performance s lkely to be a central factor determnng the attractveness of a country to foregn nvestors. 3 There s both an endogenety and a collnearty problem. Economc theory suggests that growth should depend on the average nvestment rato of the perod under study. On the other hand, t s plausble to argue that the nvestment rato depends on expected growth. Hence, the nvestment rato must be consdered as beng potentally endogenous wth respect to the growth rate. As a consequence, some authors have used the ntal nvestment rato drectly or used t to nstrument the average nvestment rato. Moreover, t s plausble to thnk that nternatonal fnancal ntegraton manly affects economc growth by boostng nvestments. In ths case, nternatonal fnancal ntegraton and the nvestment rato should be hghly correlated.

7 5 (1992) and the robustness analyss by Levne and Renelt (1992). Recall that we run all three models on contemporary and hstorcal data. We also follow the lterature n carryng out the emprcal analyss va two dfferent econometrc approaches. Frst, startng at the lower end of econometrc sophstcaton, we run a smple cross-sectonal regresson on the perods under nvestgaton. Ths s to say, we use only one observaton per country. Second, we move to the hgher end of econometrc technques by runnng a generalzed methods of moment (GMM) dynamc panel estmaton. 4 Ths two-step approach allows to combne the advantages of both estmators (Edson et al., 2004; Echengreen and Leblang, 2003; Fratzscher and Bussère, 2004). Whle the results of the cross-secton are easy to nterpret and communcate, ths relatvely smple method could bas the fndngs due to the omsson of country fxed effects, a low number of observatons, and possble endogenety of explanatory varables. It should be noted that, n the underlyng context, the dynamc panel estmaton s partcularly helpful because t enables us to explctly address the potental endogenety of the captal flow varable. Ths s mportant because an economy that exhbts hgh growth wll lkely become much more attractve for foregn nvestors. The cross-sectonal regresson, whch s estmated wth heteroskedastcty robust standard errors, takes the followng form: y = α + β IFI + γ ' X + ε, (1) where y, the dependent varable, s the logarthmc growth of real GDP per capta, IFI denotes the average captal nflow to GDP rato over the perod under study, X a vector of control varables, ε represents an..d. stochastc term, and subscrpt ndcates the countres, respectvely. The vector of control varables always ncludes the logarthm of schoolng and GDP per capta. To those we add the logarthm of perod averages of nflaton and the budget defct n model (I); the ntal nvestment rato, the nflaton rate and the budget defct n model (II); the ntal nvestment rato and average populaton growth n model (III), whle droppng the two polcy varables. The GMM dynamc panel estmaton mproves over the pure cross-secton regresson for several reasons. It uses both the cross-sectonal and the tme dmenson of the data, ncreases the number of observatons, controls for country-fxed effects and allows us to take the potental endogenety of the regressors nto account. Fve year averages reduce the cyclcalty of the data. The startng pont for the dynamc panel estmaton s the followng growth regresson: y y = ( α 1) y + β IFI + γ'x + η + ε, t, t, 1 t, 1 t, t, t, (2) where y t, s the logarthm of per capta ncome, X t, represents a set of weakly exogenous and predetermned control varables (as above), η s a (tme-nvarant) countryspecfc effect, and subscrpt t ndcates the tme perods under consderaton. We also 4 Also an ntermedate step, usng two-stage nstrumental varables regresson, seems plausble (Edson et al., 2002). We expermented wth a two-stage nstrumental varable regresson. Yet ust lke other authors, we found t hard to defne sutable nstruments as nether geographcal dstance nor legal orgn seem sutable.

8 6 nclude strctly exogenous tme-dummes whch are not reported to save space. Mnor reformulaton of equaton (2) leads to a dynamc panel regresson model of frst order: y = α y + β IFI + γ'x + η + ε. t, t, 1 t, t, t, (3) To elmnate the country specfc effects η, the precedng equaton s formulated n frst dfferences: y y = α( y y ) + β ( IFI IFI ) + γ' ( X X ) + ( ε ε ). t, t, 1 t, 1 t, 2 t, t, 1 t, t, 1 t, t, 1 (4) The system GMM estmator, ntroduced by Arellano and Bover (1995) and Blundell and Bond (1997), combnes the standard set of equatons n frst dfferences wth sutably lagged levels as nstruments wth an addtonal set of equatons n levels wth sutable lagged frst dfferences as nstruments (Bond et al., 2001). 5 Agan, the applcaton of the system GMM estmator for emprcal growth analyss s n lne wth prevous studes. We examned also the valdty of the nternal nstruments and tested for seral correlaton of the error term. 6 A detaled econometrc dscusson can be found n Bond et al. (2001), Edson et al. (2002), and Fratzscher and Bussère (2004). Our data for the contemporary perod come from commonly used sources such as the World Development Indcator database (World Bank, 2004). Inflows of portfolo and equty captal over GDP are taken from the Internatonal Fnancal Statstcs (IMF, 2005). Data on educatonal attanment (average years of schoolng) are taken from the updated Barro-Leedataset (Barro and Lee, 2000). In total, we count observatons for 54 countres for the contemporary perod ( ) coverng a large number of developng and developed countres. More demandng was the constructon of the hstorcal dataset. Ths effort would not have been possble wthout the support of numerous scholars. To a substantal extent, our dataset bulds on three recently compled datasets for the frst era of fnancal globalzaton, namely those of Obstfeld and Taylor (2003), Clemens and Wllamson (2004) and Ferguson and Schularck (2005). From these datasets come all data for schoolng (prmary school enrollment), the government balance and populaton growth. The data were collected by the authors both from prmary and secondary sources. All real GDP data come from the semnal work of Angus Maddson (Maddson 1995, 2001). Captal flow data are taken from the work by Stone (1999). It s mportant to note that the data from Stone cover only captal nflows from Great Brtan whch s the only country for whch a detaled by-country breakdown of captal outflow data exsts. However, Brtan was by far the most mportant captal exporter of the tme, traled by a large dstance by France and Germany. Brtsh data are thus very lkely to be a relable proxy for ntegraton nto the nternatonal captal market and are hghly correlated wth the overall stocks of nternatonal nvestment n 1914, whch are avalable from dfferent sources (Fes, 1965; Woodruff, 1966). The most challengng part conssted n collectng nvestment data for pre- WW1 perod whch are needed for our models (II) and (III). For many countres, we could 5 We use the Stata xtabond2 routne mplemented by Roodman (2005) wth the one-step robust estmator. Two-step estmaton yelded analogous results. 6 Test statstcs are avalable from the authors on request.

9 7 rely on Taylor (2002) and Jones and Obstfeld (1997). To these data we added nformaton from Hofmann (2001) and partly reled on unpublshed worksheets by Maddson (1992). A detaled data appendx s avalable from the authors on request. In total, we have assembled data for 24 countres over coverng more than 80 percent of global GDP n The hstorcal dataset comprses of European countres (Austra-Hungary, Denmark, France, Germany, Greece, Italy, Norway, Portugal, Russa, Span), North Amercan and Australasan settler economes (Canada, USA, Australa, New Zealand) as well as South Amercan (Argentna, Brazl, Chle, Mexco, Uruguay), Asan (Ceylon, Inda, Japan) and Mddle Eastern (Egypt, Turkey/Ottoman Empre) economes. As usual wth hstorcal data not all seres are avalable for all countres across the dfferent specfcatons n what consttutes an unbalanced panel. The summary statstcs of both the contemporary and the hstorcal dataset can be read from Table 1. Table 1: Summary statstcs Varable Mean Std. Dev. Mn Max Growth (1) Intal ncome Intal schoolng Inflaton rate Government balance Investment rato Populaton growth Fnancal ntegraton Growth (1) Intal ncome Intal schoolng Inflaton rate Government balance Investment rato Populaton growth Fnancal ntegraton (1) Mean of the growth rates of real GDP per capta over the 5-year perods n the sample. 3. Results for the Contemporary Perod As a frst step we turn to the cross-sectonal analyss of the perod We am to reproduce the fndngs of Edson et al. (2002). As explaned above, model (I) s an exact reproducton of ther benchmark regresson, albet wth data for two addtonal years (2001 and 2002). Model (II) adds the nvestment rato and model (III) proxes a standard neoclasscal growth model. 7 Among the maor economes, the only large country mssng n the hstorcal sample s Chna for whch GDP seres do not exst.

10 8 Table 2: Cross-secton estmaton, Dependent varable: growth rate of real GDP per capta Regresson (1) (2) (3) Emprcal specfcaton Model (I) Model (II) Model (III) Fnancal ntegraton 0.02** (2.10) (1.01) (1.51) Intal ncome -0.24** -0.21** (2.36) (2.37) (1.00) Intal schoolng 0.28** 0.25** 0.19** (2.30) (2.38) (2.10) Inflaton -0.02** -0.02* (2.04) (1.95) Government balance (0.41) (0.20) Investment rato 0.02* 0.02** (1.88) (2.65) Populaton growth -0.23*** (3.90) Constant 2.45*** 1.79** 2.87*** (2.71) (2.31) (3.72) # of obs R (ad.) Note: OLS wth heteroskedastcty robust standard errors; t-values n brackets; * denotes statstcal sgnfcance at the 10%, ** at the 5%, *** at the 1%-level. Table 2 dsplays the results for the 54 countres n the modern sample. Regresson (1) neatly reproduces the fndng n Edson et al. (2002), because n the cross secton there appears to be a sgnfcantly postve growth mpact of nternatonal fnancal ntegraton, measured va gross captal nflows to GDP. Also the other regressors are well behaved. There s evdence of condtonal convergence, ndcated by the negatve sgn of the coeffcent attached to ntal ncome, and also schoolng as well as nflaton carry the expected sgns. In regressons (2) and (3), usng the dfferent models dscussed above, the control varables keep the rght sgns. However, the IFI-coeffcent sees both ts statstcal sgnfcance and ts mpact on the growth rate greatly reduced. It no longer exerts sgnfcant nfluence on the per capta growth rate. Gven the apparent senstvty of the IFI-varable to changes n the specfcaton, t would seem hardly ustfed to speak of a robust relatonshp between the two.

11 9 Table 3: System GMM dynamc panel estmaton, Dependent varable: growth rate of real GDP per capta Regresson (4) (5) (6) Emprcal specfcaton Model (I) Model (II) Model (III) Fnancal ntegraton (0.30) (0.06) (0.37) Intal ncome (1.18) (1.42) (0.55) Intal schoolng ** (0.77) (1.32) (2.09) Government balance ** (1.53) (2.11) Inflaton (1.12) (0.74) Investment rato ** (1.52) (2.26) Populaton growth 0.07*** (2.92) Constant 0.23* * (1.76) (1.12) (1.76) # of obs. (# of countres) 225 (54) 225 (54) 265 (54) Note: Arellano-Bond dynamc panel estmaton, robust one-step system GMM results; t-values n brackets; * denotes statstcal sgnfcance at the 10%, ** at the 5%, *** at the 1%-level. Yet, as dscussed above, a cross-sectonal analyss usng OLS regressons could be based f captal nflows were endogenous,.e. determned by the growth rate of an economy. We therefore look at the dynamc panel estmaton. The results are presented n Table 3. They are consstent wth much of the recent lterature. Regresson (4) reproduces the results by Edson et al. (2002) as t lends lttle support to the dea of an effect of fnancal openness on growth. Moreover, addng the nvestment rato as an addtonal control varable n regresson (5) the sgn of the fnancal ntegraton varable turns wrong. Fnancal ntegraton no longer enters the equaton postvely. Regresson (6) eventually confrms that n a standard growth regresson t seems mpossble to dentfy a sgnfcantly postve nfluence of nternatonal fnancal ntegraton on economc growth durng the years In the past two decades fnancally more ntegrated countres dd not, on average, grow faster than closed economes. 4. Results for the Frst Era of Fnancal Globalzaton Were thngs dfferent n the frst era of global fnance? Table 4 presents the results of the cross-sectonal regressons for the hstorcal perod. It suggests that the answer s affrmatve. The results can be compared to those n Table 2, as we run dentcal OLS regressons on both datasets. We focus n our cross-sectonal analyss of the hstorcal perod to the years for whch we dspose of the broadest dataset, but results for earler sub-perods were also supportve. In contrast to the mxed evdence n the contemporary perod, the cross-sectonal analyss of the hstorcal perod yelds a less unambguous relatonshp between fnancal openness and growth. Captal nflows to GDP appear as a sgnfcant growth drver n all three cross-sectonal regressons, albet the sgnfcance level s somewhat lower n regresson (9). Gven the small number of observatons we are nclned not to read too much evdence nto

12 10 the cross-sectonal results, but they certanly lend some support to the dea of a more robust relatonshp between nternatonal fnancal ntegraton and growth n the frst era of globalzaton. However, we are alerted that the cross-secton may be msleadng and thus look to the dynamc panel estmaton for more conclusve evdence. Table 4: Cross-secton analyss, Dependent varable: growth rate of real GDP per capta Regresson (7) (8) (9) Emprcal specfcaton Model (I) Model (II) Model (III) Fnancal ntegraton 0.28*** 0.26*** 0.23* (3.55) (3.60) (1.77) Intal ncome (0.05) (1.46) (1.39) Intal schoolng 0.06*** (2.98) (0.10) (0.41) Inflaton (1.42) (1.36) Government balance (0.32) (0.96) Investment rato (1.47) (1.12) Populaton growth 0.01 (0.19) Constant -0.37* -0.34* (1.79) (2.05) (1.45) # of obs R (ad.) Note: OLS wth heteroskedastcty robust standard errors; t-values n brackets; * denotes statstcal sgnfcance at the 10%, ** at the 5%, *** at the 1%-level. The results of the dynamc panel estmaton are presented n Table 5. We start agan by runnng the base model from Edson et al. (2002). Regresson (10) shows that also the dynamc panel model yelds a hghly sgnfcant effect of nternatonal fnancal ntegraton on growth. Recall that the dentcal regresson wth data for the years faled to exhbt a robust lnk. The control varables reman well-behaved and enter wth the rght sgn. There s evdence of condtonal convergence and also the human captal proxy s strongly sgnfcant. The degree of nternatonal fnancal ntegraton also remans a sgnfcant growth drver n regressons (11) and (12) whch mplement the two other models presented above. The statstcal sgnfcance declnes somewhat f the nvestment rato and populaton growth are added as regressors, but stays above the 5% and 10%-levels. It s the ncluson of populaton growth, whch enters the hstorcal regresson wth a postve sgn, that drags down the growth effect of fnancal ntegraton. Summng up, we have run dentcal regressons on the contemporary and hstorcal datasets to examne the fnancal ntegraton-growth nexus. Our regressons ndcate that before WW1 nternatonal fnancal ntegraton spurred economc growth, whereas there s lttle evdence of a comparable effect n recent decades. In addton, ths fndng appears robust over a number of dfferent specfcatons. How can we thus explan that fnancal

13 11 ntegraton apparently contrbuted to growth n the frst era of fnancal globalzaton, but no longer today? The followng secton presents a potental explanaton for ths observaton. Table 5: System GMM dynamc panel estmaton, Dependent varable: growth rate of real GDP per capta Regresson (10) (11) (12) Emprcal specfcaton Model (I) Model (II) Model (III) Fnancal ntegraton 0.04*** 0.06** 0.04* (2.75) (2.11) (1.97) Intal ncome -0.03* * (1.88) (0.70) (2.06) Intal schoolng 0.03*** ** (4.81) (1.28) (2.29) Government balance (1.53) (1.10) Inflaton (1.06) (0.27) Investment rato (0.40) (1.60) Populaton growth 0.01 (0.73) Constant (0.57) (0.65) (1.54) # of obs. (# of countres) 141 (24) 93 (17) 107 (19) Note: Arellano-Bond dynamc panel estmaton, robust one-step system GMM results; t-values n brackets; * denotes statstcal sgnfcance at the 10%, ** at the 5%, *** at the 1%-level. 5. Dscusson of the Man Results 5.1. Prelmnary Thoughts The man fndng of the precedng secton was that nternatonal fnancal ntegraton promoted economc growth n the frst era of globalzaton but fals to do so n modern tmes. How can we explan ths phenomenon? We am to sketch a prelmnary answer to the queston. In a nutshell, we argue that the standard neoclasscal textbook model provdes a better descrpton of the real world for the hstorcal era than t does for the modern perod. In other words, the neoclasscal model works better for the hstorcal perod, because the very prerequstes for t to work properly were n place back then, but are no longer today. Ths concluson rests on four (nterdependent) observatons: (1) It s very lkely that the qualty of nsttutons, especally the assgnment and enforcement of property rghts, was hgher n a number of countres (belongng to net captal mporters) than t s today. For nstance, Lucas (1990, pp. 94/95) argues: 8 Untl around 1945, much of the Thrd World was subect to European-mposed legal and economc arrangements, and had been so for decades or even centures. A European lendng to a borrower n Inda or the Dutch East Indes could expect 8 For a smlar lne of reasonng see Ferguson (2003).

14 12 hs contract to be enforced wth exactly the same effectveness and by exactly the same means as a contract wth domestc borrowers. The mportance of ths pont for the favored reasonng s renforced by the comprehensve emprcal study of Alfaro et al. (2003), who am to resolve the Lucas paradox n modern tmes. Employng cross-country data for 50 countres over they fnd that nsttutonal qualty s the most mportant varable n explanng the paucty of rch-poor captal flows. (2) The hstorcal perod under study spans the tme of the nternatonal gold standard. Ths monetary system produced a comparably stable nternal (nflaton rate) and external (exchange rate) monetary envronment. Despte the fact that the gold standard dd not gve rse to strct prce stablty, as argued by Cooper (1982), t brought a number of countres comparably stable monetary condtons, especally when contrasted wth the experences of most developng countres n the post-ww2 perod. 9 In addton, the frequency of exchange rate crses was much smaller durng that tme (Echengreen, 2002). The fulfllment of the two precedng premses s an mplct but necessary condton for the textbook neoclasscal model to work properly. As a consequence, the patterns of captal moblty n the hstorcal perod were markedly dfferent from what we observe n the contemporary world economy: (3) Net captal movements were hgher n the hstorcal era than they are today. A standard way to examne de facto captal moblty s the Feldsten-Horoka (1980) test. Numerous applcatons of ths test produced ample emprcal evdence that captal moblty was ndeed much hgher n the perod compared to the post-ww2 perod (Bayoum, 1990; Echengreen, 1990; Taylor, 1996; Obstfeld and Jones, 1997). (4) In the frst era of fnancal globalzaton, gross captal flows to perpheral countres (about 2 % of GDP on average, as documented n Table 1) gave rse to substantal net captal movements. The modern perod s also characterzed by massve gross captal flows, but they do not translate nto substantal net captal flows from rch to poor regons. In the pre-1914 boom, there was lttle dfference between net flows and gross flows because most flows were un-drectonal from the rch core to the poor perphery (Obstfeld and Taylor, 2004). Post 1970, however, gross flows (both nflows and outflows) ncreased tremendously. But net flows (nflows mnus outflows) held constant at relatvely low levels for the last thrty years. Most nternatonal captal flows nowadays take place between rch countres (north-north as opposed to north-south). Obstfeld and Taylor (2004) conclude that contemporary fnancal globalzaton s more characterzed by dversfcaton fnance not by development fnance. As a result, the Lucas paradox of mssng rch-poor captal flows was less pronounced n the frst era of globalzaton. Whle Clemens and Wllamson (2004) fnd that the Lucas paradox was as strong before 1914, other studes have questoned ths result and ponted to a much less pronounced wealth bas n nternatonal captal flows before WW1 (Obstfeld and Taylor, 2004; Schularck, 2005). Ths mrrors the earler fndngs by Edelsten (1982), who showed that the hstorcal perod was assocated wth substantal nvestment flows from contnental Europe to overseas areas. In summary, one can argue that nternatonal captal market ntegraton sgnfcantly fostered economc growth n the hstorcal epsode because of the combnaton of two key factors: Frst, there were tremendous nvestment opportuntes n regons where fnancal 9 Table 1 shows an average nflaton rate whch s nearly twce as large for the modern perod compared to the hstorcal perod. It s true that t s more the varablty f the nflaton rate rather than ts level whch represents a problem. However, the varablty and the level of nflaton are typcally hghly correlated.

15 13 funds were scarce. Second, a favorable nsttutonal settng and a comparably stable monetary envronment were secured, so that the prerequstes for the textbook model to work properly were gven. Taken together, these two factors gave rse to substantal net captal flows from the rch core to the poor perphery that contrbuted to the observed growth effect of nternatonal fnancal ntegraton n the frst era of global fnance A Smple Model wth Imperfect Property Rghts That fnancal globalzaton can have had markedly dfferent effects on economc growth n the two eras of nternatonal captal market ntegraton can be llustrated through a farly smple neoclasscal model that ntegrates mperfect property rghts. Consder a perfectly compettve economy populated by a contnuum of mass one of dentcal frms and households. Frms produce a homogenous good accordng to: ( ) ( ) ( ) 1 α β α β Y = K N L, (5) where {1, 2}, Y s fnal output, taken as the numerare, produced by frm [0,1] n country K, N and L denote captal, labor and land employed by frm n country, 0 < α, β < 1 and 0 < α + β < 1. In ths economy ndvdual output α β α β gven by Y ( K ) ( N ) ( L ) 1 Y equals average output =, whch also equals aggregate output. Moreover, we assume that the supply of labor and land s fxed n every country and that labor s mmoble nternatonally. To smplfy matters and wthout loss of generalty, we can set ( N ) β ( ) 1 L α β equal to unty for both countres. Turnng to property rghts protecton, we assume that output produced by an ndvdual frm s nsecure n that frm s ' clam to hs own output can be contested by other frms, and n turn may contest the clams of others to ther producton. For smplcty ths process s modeled by magnng that frms compete aganst the economy-wde average and that ndvdual output s reallocated accordng to an exogenous sharng rule ndcated by 0< µ < The gross revenue of the representatve frm accordngly reads: ( ) α (1 ) ( ) α Y = µ K + µ K. (6) Note that the term ( K ) α represents ndvdual output of frm n country, whereas ( K ) α stands for the average output of economy. Proft maxmzaton mples: r = µαy / K, w / = µβy N, and p (1 ) / = µ α β Y L, where r, w, and p denote the rental prce of captal, the wage rate and rental prce of land, respectvely. Captal goods (as well as land) are owned by households and rented to frms. In addton, households are the owners of the frms. Income of the typcal household I reads: 10 Ths s smlar to Gonzàlez (2004, 2005), where µ s endogenously determned by nvestments n defensve and offensve captal.

16 14 α I = rk + wn + pl+ K = Y = Y. (7) (1 µ ) ( ) Both economes are perfectly dentcal except for two aspects: Frst, the ntal captal endowment, relatve to labor and land, mght dverge. Second, the extent of property rghts protecton, gven by µ, mght also be dfferent. Fnally, t s mportant to note that equaton (6) can also be nterpreted as a shortcut formulaton whch captures the redstrbutve effects of unexpected nflaton or unexpected exchange rate changes. Ths s to say that by redstrbutng the revenue of economc actvtes from one ndvdual to another, weak property rghts and an unstable monetary envronment are very smlar. Ths completes the descrpton of the model. We now compare world output under autarky to world output under captal market ntegraton. World output under autarky Y s gven by: w a 1 α 2 ( ) ( ) w α a a a Y = K + K, (8) 1 where K a and contrast, world output under nternatonal captal moblty 2 K a denote the captal stock n economy 1 and 2 under autarky. In 1 α 2 ( ) ( ) w α g g g w Y g reads: Y = K + K, (9) 1 2 where K g and K g denote the amount of captal allocated to economy 1 and 2 under w captal market ntegraton. The nternatonal allocaton of the global captal stock K s 1 2 determned by the equlbrum condton r = r, whch mples: µα Y µα Y =. (10) Kg Kg Takng w K : = K + K nto account, the soluton for 1 2 g g 1 K g and 2 K g reads as follows: K 1 g = 1+ K w α 1 2 ( µ α ) ( µ ), (11) K 2 g = 1+ K w α ( µ α ) ( µ ). (12).e. For nstance, f both economes exhbt the same extent of property rghts protecton, µ = µ, then the nternatonal allocaton of captal s K = K = 0.5K w. g g

17 15 We are now n the poston to answer the followng queston: How does the world output change n response to captal market ntegraton? It can be shown that the rato of world output under ntegraton and world output under autarky can be expressed as follows: Y Y w g w a = 1 α 1 α 1 α α 1 1+ γ + 1+ γ ( 1+ λ ) α 1+ λ α, (13) where γ : = µ / µ and λ : = Ka / Ka. The precedng expresson shows that, n the smple model under study, relatve world output depends on two key parameters, namely the relatve qualty of property rghts γ and the relatve ntal captal endowment λ. By assumng that the captal share s α = 0.45 (whch s consstent wth a broad concept of captal, ncludng partly human captal), the consequences of captal market ntegraton for world output can be llustrated by Fgure 1. 1 λ yw 1 g yw a γ Fgure 1: Relatve world output as a functon of γ and λ (assumng α = 0.45 ). Provded that both economes are ntally endowed wth the same stock of captal (relatve to labor and land),.e. λ = 1, and assumng that the extent of property rghts protecton s dentcal n both economes,.e. γ = 1, captal market ntegraton does not exert a w w postve output effect,.e. Yg / Y a = Now f ntal endowments dffer strongly,.e. λ s very small, and property rghts protecton s farly smlar n both economes,.e. γ s near unty, captal market ntegraton has a strong postve mpact on world output. In the numercal example llustrated n Fgure 1 ths ncrease amounts to about 40%. Ths constellaton captures the basc characterstcs of the hstorcal perod. On the other hand, the modern perod s also characterzed by a substantal dfference n ntal captal endowments,.e. λ s low. However, t appears evdent that there s a large dfferental n property rghts protecton,.e. γ s low. For ths constellaton, the model under study predcts ether a neglgble or even a negatve output effect. 11 It s unsurprsng that the ntegraton of two dentcal neoclasscal economes does not have an effect on growth. Ths would be dfferent wth endogenous growth models exhbtng scale effects due to knowledge spllovers (Bretschger and Steger, 2004).

18 16 Ths last observaton s, of course, a typcal second-best result. In the presence of a strong dstorton (.e. a large dfferental n property rghts protecton), the removal of a second market mperfecton (.e. the removal of barrers to captal moblty) may even worsen the stuaton n the sense of reducng world output. Fnally, t should be noted that the smplstc model under study mples an mmedate adustment of the nternatonal captal allocaton and therefore an nstantaneous change n world output. In realty, ths process would be dstrbuted over tme due to the presence of captal adustments costs. Hence, the model s capable of capturng the basc effects of captal market ntegraton for transtonal output growth. 6. Summary and Concluson Consderable emprcal effort has been devoted to nvestgate whether nternatonal fnancal ntegraton boosts economc growth. The overall result of these studes, whch prmarly focused on the post-ww2 perod, was that there are lttle ndcatons of a robust growth effect of nternatonal fnancal ntegraton. On the other hand, there s substantal narratve evdence from economc hstory that hghlghts the contrbuton European captal made to economc growth of perpheral economes before In ths paper, we have nvestgated emprcally the nexus between nternatonal fnancal ntegraton and economc growth for the pre-ww1 perod. We have used the same emprcal models and the same econometrc technques as recent studes allowng for a hgh degree of comparablty wth prevous work for the contemporary perod. Our results suggest that nternatonal captal market ntegraton fostered economc growth sgnfcantly n the hstorcal perod, but no longer does so today. Our prelmnary explanaton of these dverse experences s very smple. We thnk that the neoclasscal model provdes a vald descrpton of the hstorcal perod, but appears unsutable to explan the contemporary world economy. In partcular, we argue that the hstorcal perod, n stark contrast to the post-ww2 perod, was characterzed by: () low dfferentals n property rghts protecton between countres; () a (comparably) stable monetary envronment; () a hgher degree of net captal moblty, as measured by the Feldsten-Horoka crteron; and (v) substantal (net) captal movements from rch (core) to poor (perpheral) regons. Our results renforce the concluson that those economes whch open themselves to the world economy need at frst abolsh domestc dstortons to reap the benefts of globalzaton. More specfcally, t seems especally mportant to establsh good property rghts n all economes partcpatng n the world economy. Ths lne of reasonng s a straghtforward applcaton and renforces the mportance of the well-known second-best theorem.

19 17 7. References Alfaro, L., S. Kaleml-Ozcan and V. Volosovych (2003). Why doesn t captal flow from rch to poor countres? An emprcal nvestgaton Harvard Busness School Workng Paper, No Arnello, M and O. Bover (1995). Another Look at the Instrumental Varable Estmaton of Error-Components Models. Journal of Econometrcs 68 (1): Arteta, C., B. Echengreen, C. Wyplosz (2001). "When Does Captal Account Lberalzaton Help More Than It Hurts?" Manuscrpt, Unversty of Calforna at Berkeley. Barro, R. J. and J.-W. Lee (2000). Internatonal Data on Educatonal Attanment: Updates and Implcatons. CID Workng Paper No Barro, R. J. and X. Sala--Martn (1992), "Convergence", Journal of Poltcal Economy, 100, 21, Bayoum, T. (1990). Savng-Investment Correlatons, IMF Staff Papers, 37, Bekaert, G., R. Harvey, C. Lundblad (2001). "Does Fnancal Lberalzaton Spur Growth?" NBER Workng Paper No Bhagwat, J. (1998). "The Captal Myth." Foregn Affars 77 (3): Blundell, R. and S. Bond (1997). Intal Condtons and Moment Resrctons n Dynamc Panel Data Models. Unversty College London Dscusson Papers n Economcs 97/7. Bond, Stephen, Anke Hoeffler, and Jonathan Temple (2001), GMM estmaton of emprcal growth models, Workng Paper, Unversty of Brstol. Bordo, M., A. M. Taylor, et al., Eds. (2003). Globalzaton n Hstorcal Perspectve Natonal Bureau of Economc Research Conference Report. Chcago, Unversty of Chcago Press. Bretschger, L. and T. M. Steger, The Dynamcs of Economc Integraton: Theory and Polcy, Internatonal Economcs and Economc Polcy, 2004, 1 (2-3), Clemens, M. A. and J. Wllamson (2004). "Wealth Bas n the Frst Global Captal Market Boom." Economc Journal 114 (2): Cooper, R. (1982). The Gold Standard: Hstorcal Facts and Future Prospects. Brookng Papers on Economc Actvty, Vol. 1, Edelsten, M. (1982). Overseas Investment n the Age of Hgh Imperalsm, The Unted Kngdom Columba Unversty Press, Edson, H., R. Levne, L. Rcc, and T. Slok (2002). "Internatonal Fnancal Integraton and Economc Growth." Journal of Internatonal Money and Fnance, vol. 21(6), Edson, H., M. W. Klen, L. Rcc, and T. Slok (2004). "Captal Account Lberalzaton and

20 18 Economc Performance: Survey and Synthess." IMF Staff Papers, Vol. 51, No. 2, Edwards, S. (2001). "Captal Moblty and Economc Performance: Are Emergng Economes Dfferent?" NBER Workng Paper No (January). Echengreen, B. (2002). Internatonal Fnancal Crses: Is the Problem Growng? Jahrbuch für Wrtschaftsgeschchte 1: Echengreen, B. (2001). "Captal Account Lberalzaton: What Do the Cross-Studes Tell Us?" World Bank Economc Revew, October 2001; 15: Echengreen, B. (1990). Trends and Cycles n Foregn Lendng. In: Captal Flows n the World Economy, H. Sebert (ed.), Tübngen, Mohr-Sebeck. Echengreen, B. and D. Leblang (2003). "Captal Account Lberalzaton and Growth: Was Mr. Mahatr Rght?" Internatonal Journal of Fnance and Economcs 8: Fes, H. (1965). Europe The World's Banker WW Norton, New York. Feldsten, M. S. and C. Y. Horoka (1980). "Domestc Savng and Internatonal Captal Flows." Economc Journal 90: Ferguson, N. and M. Schularck (forthcomng). "The Empre Effect: The Determnants of Country Rsk n the Frst Era of Fnancal Globalzaton." Journal of Economc Hstory. Ferguson, N. (2003). Brtsh Imperalsm Revsed: The Costs and Benefts of Anglobalzaton Development Research Insttute Workng Paper Seres No. 2. Fratzscher, M. and M. Bussere (2004). "Fnancal Openness and Growth: Short-Run Gan, Long-Run Gan?" European Central Bank Workng Paper No González, Francsco M. and Hugh M. Neary, "Optmal Growth Polcy Under Prvately Enforced Property Rghts", Workng Paper UBC, 2004, Vancouver. González, Francsco M., Effectve Property Rghts, Conflct and Growth", Workng Paper UBC, 2005, Vancouver. Grll, Vttoro, and Gan Mara Mles-Ferrett (1995). Economc Effects and Structural Determnants of Captal Controls, Staff Papers, Internatonal Monetary Fund, 42 (September): Hofman, André, The Economc Development of Latn Amerca n the Twenteth Century. Northampton, MA: Edward Elgar Publshng, Internatonal Monetary Fund (2005), Internatonal Fnancal Statstcs. Electronc verson, accessed va Research Tme Seres Database (ReTIS 3.2) at Deutsche Bank Research. Jones, T. and M. Obstfeld, M. (1997). Savng, Investment, and Gold: A Reassessment of Hstorcal Current Account Data. NBER Workng Paper Klen, M. W. (2005). "Captal Account Lberalzaton, Insttutonal Qualty and Economc