FOREIGN LIQUIDITY, ECONOMIC OPENNING AND GROWTH IN LATIN AMERICAN ECONOMIES

FOREIGN LIQUIDITY, ECONOMIC OPENNING AND GROWTH IN LATIN AMERICAN ECONOMIES Abstract The man concern of our emprcal study s to shed lght to the queston of whether or not and n whch drecton long-run growth has been assocated wth fnancal (lqudty) and trade openng snce early 1970s usng a panel data approach for 11 Latn Amercan countres. Prevous emprcal studes reported mxed results n terms of fndng a stable assocaton between captal account lberalzaton and growth or even for trade openng and growth. Our emprcal results suggest an mportant lnk between nternatonal lqudty and growth, but the same does not apply for trade openng and growth. Key-Words Foregn lqudty, Trade Openng; Economc Growth; Latn Amerca. JEL Classfcaton F41; F43; C3. Resumo O propósto central deste nosso estudo empírco é dscutr as relações entre o crescmento econômco de longo prazo e a lqudez externa, assm como deste com a abertura comercal, desde começo dos anos 70, usando a abordagem de panel para onze economas da Amérca Latna. Estudos anterores encontraram resultados controversos em termos de relações estáves entre a lberalzação da conta de captal e o crescmento econômco, ou anda, entre a abertura comercal e o crescmento de longo prazo do PIB. Nossos resultados empírcos sugerem um mportante vínculo entre a lqudez externa e este crescmento, mas o mesmo não pode ser observado para a relação entre abertura comercal e crescmento econômco. Palavras-Chaves Lqudez Externa; Abertura Comercal; Crescmento Econômco,; Amérca Latna. Classfcação JEL: F41; F43; C3.

I Introducton The man concern of our emprcal study s to shed lght to queston of whether or not and n whch drecton long-run growth has been assocated wth fnancal (lqudty) and trade openng snce early 1970s usng a panel data approach for 11 Latn Amerca countres. It s well known by the lterature that more recently, developng countres have faced perods of nternatonal fnancal crses wth sgnfcant outcomes generally assocated wth lower economc growth rates durng the last two decades when compared to hstorcal rates durng the sxtes and seventes. Gven the fact that most developng and emergng economes have gone through a perod of openng trade and captal accounts snce early 1990s we want to nvestgate f and how dfferent measures of lqudty (three) and trade flows can be part of the explanaton for long-run economc growth n Latn Amercan economcs. It s far to relate trade and fnancal openng n the sense that the frst one nvolves an ncrease n trade of goods whle the second n trade of captal snce foregn nvestment s a form of ntertemporal trade, and based on the argument that trade benefts growth one can argue that hgher captal moblty wll have smlar mpact on growth. The mplcatons of ths lne of thought s n the background of our research where we wll be usng a panel data for Latn Amercan economes to nvestgate possble mplcatons of changes n nternatonal lqudty and trade openng to long-run growth. One has to remnd that nternatonal lqudty s assocated wth captal account lberalzaton n the sense that wthout the latter (no captal moblty) nternatonal fnancal markets has a lmted role to be played n fosterng hgher economc growth rates. The paper s dvded n three sectons other than ths one and fnal consderatons. Secton two develops a general revew of the lterature on fnancal openng and growth, secton three deals wth some methodologcal ssues related to panel data analyss and varable descrpton, and secton four summarzes the man emprcal fndngs. We can draw a general concluson from the present work, suggestng no clear lnk between trade openness and growth, even though there s evdence that hgh nternatonal lqudty and an mprovement n long-run growth rates are somehow assocated for Latn Amercan economes. II Fnancal Openng and Growth: Theory and Emprcs The man task of ths secton s to summarze what theory says regardng the relatonshp between fnancal openng and growth, and also analyze the emprcal fndngs assocated to ths ssue. One can say that theory has no unambguous predcton of whether or not captal account lberalzaton enhances growth and the emprcal evdence can be consdered nconclusve. At a frst look, there are two channels through whch captal account lberalzaton affects growth. The frst one s assocated wth the argument that hgher captal moblty ncreases the domestc nvestment rate snce captal flows towards countres where captal s relatvely scarce and where the margnal productvty of captal s hgher, and the outcome s hgher economc growth rates. A second possble channel can be assocated wth captal flows to sectors wth hgher rates of return (portfolo dversfcaton) when fnancal markets does hot operate wth sgnfcant dstortons and the outcome of captal account

3 lberalzaton tend to generate a more effcent resource allocaton and a faster rate of economc growth. 1 The predctons offered by theores regardng the nternatonal fnancal ntegraton effects on growth can be consdered conflctng to some extent. Internatonal fnancal ntegraton facltates rsk-sharng (dversfcaton) enhancng captal allocaton and economc growth, but t can have a negatve mpact on growth f t s mplemented under economc condtons where the exstence of dstortons s the rule rather the excepton. The polcy prescrpton to extend the process of fnancal ntegraton n less-developed countres s controversal. 3 A sgnfcant number of lterature revews on captal account lberalzaton and growth have been developed n the last fve years and here we wll brefly survey the most mportant emprcal results assocated wth them. 4 The frst studes on captal account lberalzaton and growth have not fnd supportve results. Alesna, Grll and Mles-Ferrett (1995) found that fnancal open has small and nsgnfcant effects on growth. Rodrk (1998) uses a smlar approach for a larger sample and found no stable assocaton between captal account lberalzaton and growth. Kraay (1998) fnd no lnk between economc growth and the IMF restrcton measure. On the other hand, Qunn (1997) develops an emprcal analyss consderng the mpact of both captal account openness and the change n openness where the results suggest a postve assocaton between the change n captal account openness and growth. 5 Edwards (001) found a sgnfcant postve effect of captal account lberalzaton on growth, but the 1 The outcome s condtoned on how domestc and nternatonal fnancal markets operate, where fnancal nstablty can be harmful to acheve hgher economc growth rates. Obstfeld (1994) s a poneer work lnkng fnancal openness and growth n a model wth portfolo dversfcaton. Barro, Mankw and Sala--Martn (1995) develops an open economy verson of a neoclasscal growth model where the constrant that domestc savngs s the only source to fnance domestc nvestment, and n ths stuaton, captal account lberalzaton (ncrease n the access to foregn savngs) wll ncrease captal accumulaton and the economy can acheve hgher economc growth rates. Other possble benefcal effects of nternatonal fnancal ntegraton on growth are assocated wth the mprovement of the domestc fnancal system (hgher competton and new fnancal servces mported). Trade dstortons may result n a process of captal account lberalzaton where captal nflows to sectors where the economy does not have a comparatve advantage wth unfavorable economc growth outcomes. 3 See Rodrk (1998) for the argument and emprcal evdence that captal account lberalzaton cannot be postvely assocated to hgher economc growth rates. 4 See Table 1A of the Appendx for an overvew of the emprcal studes on fnancal openng and growth where we compare each one of them n terms of number of countres studed, perod of nvestgaton, varables used, estmaton technque and man results. 5 The emprcal study developed by Qunn (1997) suggest that the change n captal account lberalzaton has a strongly sgnfcant effect on the growth n real GDP per capta but he does not nclude a regresson wth both of these ndcators (fnancal and trade openness). One has to remnd that changes n fnancal openness s correlated wth changes n trade openness, where the fndng of a sgnfcant effect of the change n captal account lberalzaton on growth may reflect the correlaton of changes n restrctons on the captal account and the current account.

4 results were restrcted to hgh ncome countres. 6 Klen and Olve (000) fnd a postve effect of captal account lberalzaton on growth for ndustral countres, but not for lessdeveloped countres. Arteta, Echengreen, and Wyplosz (001) when ntroducng proxes for the degree of macroeconomc stablty fnd some support for dfferences n the effect of captal account lberalzaton across countres. 7 Among the studes on captal account lberalzaton and growth that ncludes some measure of trade openness as an addtonal (control) varable we can menton Echengreen and Leblang (00) and they fnd a postve and sgnfcant coeffcent for trade openng and ts mpact on growth. 8 Another emprcal research on nternatonal fnancal lberalzaton and growth that ncludes trade openness as a control varable s McLean and Shrestha (00)) where the coeffcent shows up as postve and sgnfcant regardless f the sample ncludes both developed and developng countres or only the latter. Arteta, Echengreen and Wyplosz (001) use the Sachs-Warner trade openness measure and n the pooled regresson the coeffcent s postve and sgnfcant. 9 Klen and Olve (000) examnes the mpact of fnancal development on growth ncludng a vector of control varables that are potentally related wth a country s economc growth, where one of them s the 1986 rato of exports plus mports to GDP and the results seems have not changed by the ncluson of such varable (open captal account ncreases fnancal depth and hgher economc growth rates). Levne, Loayza and Beck (1999) uses a dynamc panel estmaton wth two sets of condtonal nformaton where n one of them openness to trade (log) s used and the results ndcate a postve and sgnfcant effect for fnancal ntermedaton and growth. Snce we are ncludng openness to trade defned as the rato of the sum of exports mports relatve to GDP as one of our varables to capture possble mpacts to long-run growth n Latn Amerca, a bref word on how the lterature and emprcal research has seen ths relatonshp s necessary. The lterature on openness to trade and growth has been characterzed by many controverses n terms of assocatng openness wth hgher growth rates. Rodrguez and Rodrk (001) s one of the emprcal works that does not fnd such postve assocaton n the sense that lberal trade polces does not guarantee faster growth rates. On the other hand, dfferent emprcal studes fnd that lower trade barrers together wth a stable exchange rate system, sound monetary and fscal polces help promotng economc growth. 10 6 Dfferent results were found by Edson, Klen, Rcc and Slok (00) where the assocaton of captal account lberalzaton wth growth s stronger n less developed countres, whle Arteta, Echengreen and Wyplosz (001) fnd evdence that nether for developed or developng countres captal account lberalzaton affects growth. 7 Sachs and Warner (1995) ntroduced openness ndex and the exchange black market premum. The results ndcate that countres that open ther captal accounts grow faster only f they elmnate the black market premum. 8 The results apply for 47 countres durng the perod of 1975-95. 9 The same result has not been derved when ncludng a dfferent measure of trade openness based on the nteracton between the Barro-Lee trade measure and Qunn s measure of captal account lberalzaton. The coeffcents are postve but not sgnfcant. 10 See Baldwn (003) for a survey of the lterature on openness and growth.

5 III Varable Descrpton and Econometrc Model The man model to be estmated and the varables descrpton are the followng: yˆ t = α + β 0Opent + β1fl01t + β FL0t + β 3FL03t + β 4 yt + ε t (1) where: ŷt s the real growth rate of GDP; Opent s the trade openng; FLt s the foregn lqudty measured accordng to FL01 (the rato between the foregn reserves and the mports), FL0 (the rato between the external debts and real GDP), FL03 (the rato between the external debt and exports), and s the real GDP. We are expectng a postve y t coeffcent for β1and a negatve coeffcent for β and β 3. A postve coeffcent should be assocated wth the dea that more lqudty mproves growth, whle for a negatve one ndcates that a hgher lqudty ndcator deterorates growth. The data were collected usng the IMF (CD-Rom, Aprl 003), Internatonal Fnancal Statstcs, and the World Bank (World Development Indcators). The estmaton of equaton (1) has been mplemented usng the orgnal sample wth annual data from 197 to 000, averagng the data for each fve years, except for the frst observaton (197-1975). We can see, n table 3A the correlaton matrx, where we obtaned smlar mean samples and dsperson measures, and the same apples for the correlaton between the varables used n emprcal research. We used the transformed mean sample to estmate model (1) usng a panel data analyss. Estmaton usng panel data has several advantages over purely cross-sectonal estmaton. Frst, besdes consderng the cross-country relatonshp between fnancal development (nternatonal lqudty) and growth, we also would lke to take nto account how fnancal development over tme wthn a country may have an effect on the country s growth performance. Workng wth a panel helps ganng degrees of freedom by addng the varablty of the tme-seres dmenson to the analyss. Second, n a panel context, we are able to control for unobserved country-specfc effects and thereby reduce bases n the estmated coeffcents. Thrd, our panel estmator controls for the potental endogenety of all explanatory varables, whle the cross-sectonal estmator presented by prevous studes only controls for the endogenety of fnancal development. The way our panel estmator controls for endogenety s by usng nternal nstruments, that s, nstruments based on lagged values of the explanatory varables. Ths method does not allow us to control for full endogenety but for a weak type 11. The panel approach allows for two basc models: fxed and random effect models, both of them acceptng statc and dynamc specfcatons. The fxed effect model, also known as least square dummy varable (LSDV), s a generalzaton of an ntercept-slope-constant model for panel analyss, ntroducng a dummy varable to capture the effects of omtted varables, that are constant over tme. 11 To be precse, Levne, Loayza and Beck, (1999) assume that the explanatory varables are only weakly exogenous, whch means that they can be affected by current and past realzatons of the growth rate but must be uncorrelated wth future realzatons of the error term. Thus, the weak exogenety assumpton mples that future nnovatons of the growth rate do not affect current fnancal development. Ths assumpton s not partcularly strngent conceptually and we can examne f t has statstcal valdty. Weak exogenety does not mean that economc agents do not take nto account expected future growth n ther decson to develop the fnancal system; t just means that future (unantcpated) shocks to growth do not nfluence current fnancal development. It s the nnovaton n growth that must not affect fnancal development.

6 In ths specfcaton, the ndvdual-effects can be freely correlated wth the regressors. Ther estmaton s, n fact, the own estmaton of the model of multple regressons wth bnary varables for each one of the n unts of the analyss, n such a way that the ntroducton of them wll cause the ntercept of the regresson to be dfferent for each one of these varables and pck up the heterogenety among them. The ordnary least square (OLS) estmator wll be a consstent and effcent estmator and t s know as LSDV. The random-effect model specfcaton treats the ndvdual-specfc effects as random varables. Ths model assumes no correlaton between the ndvdual effects and the other random varables, where the estmaton was pursued usng the Generalzed Least Square (GLS). One crucal queston s to know whch s the most approprate model? Accordng to Frees (003) t depends on the avalable nformaton and the estmaton objectves. If, for example, the man concern of the analyss wll be to test the effect of the varables where the ndvduals are classfed n groups, then the random effect specfcaton s more approprate. In Hsao (1999: 4): The fxed-effects model s vewed as one n whch nvestgators make nferences condtonal on the effects that are n the sample. The random-effects model s consdered as the one n whch one can make uncondtonal or margnal nferences wth respect to the populaton of all effect. A statc panel-data model can be wrtten as: ' yt = x t β + λt + η + ε t t = 1,..., T = 1,..., N () where: λt and η are tme and ndvdual specfc effects respectvely, x t s a vector of explanatory varables, N s the number of cross-secton observatons and NT s the total number of observatons. The man goal s to obtan a consstent estmator of β wth the desred effcency propretes. The choce of the estmaton technque to be used depends on the hypothess assumed for the relatonshp between the error-term ( ε t ) and the regressors (x t ) n terms of random error and the fxed effect η. In the more restrctve case, one can assume that E( η,x t ) = 0 (the orthogonalty between the fxed-effect and the regressors) and E( ε t,x t-s ) = 0 for any lag s. One can use OLS (Ordnary Least Square) or LSDV (Least Square Dummy Varable) snce both provde consstent estmators, but the second s the more effcent. If we do not consder the hypothess of orthogonalty between the fxed effect and the regressors, that s, f we assume E( α,x t ) 0, t s not possble to assume consstence for the OLS estmaton, and LSDV should be the estmaton choce snce t s the only one that s consstent. Another consstent estmator s OLS usng the frst dfference (FD-OLS) 1, but some cauton s necessary snce t presents effcency problems. One can also assume that E( α,x t ) = 0 and E( α,x t ) 0. In ths case, none of the above estmators (OLS, LSDV or FD-OLS) are consstent, and to obtan consstent 1 Takng the frst dfference n (1), the fxed effect s elmnated.

7 estmators of β we need to use Instrumental Varables or GMM (Generalzed Methods of Moments). Consderng the model represented by equaton () f α s not assumed to have fxed parameters, we are dealng wth a Random Effect Model, and n ths case we assume thatα are dentcally and ndependently dstrbuted wth zero mean and varance and { α } are ndependent varables wth random errors { } Comparng the estmated slopes for the fxed effect and the random effect models, one can say that: 1) assumng that the formulaton of the fxed effects s rght, so β EF s consstent and asymptotcally effcent, and β RE s nconsstent. ) assumng that the formulaton n terms of random effect s rght, so β RE s consstent and asymptotcally effcent, and β EF s consstent, as well. Accordng to Hsao (1999:36), the GLS estmator (Generalzed Least Square) s the weghted average between-groups and wthn-groups. The GLS estmator can converge to OLS or to LSDV. In the LSDV procedure (fxed effect model) the source of varaton s not taken nto account and OLS and LSDV can be consdered as an example of all or nothng n terms of varaton between groups. The procedure that consders α as random allows for an ntermedate soluton and does not have to treat everyone as dfferent or smlar, accordng to GLS estmators. In models (1) and (), there are no lagged varables, nether regressors or explanatory varables. Incorporatng such elements, we propose the followng model: ε. t σ α y t = α + γy βx + t 1 + t + λt ε t (3) for = 1,..., N and t = 1,..., T where Eε = 0, Eε ε = σ for = j and t = s, and Eε ε = 0, for all the other cases. t t j u If we assume E( α,x t-s ) = E(ε t,x t-s ) =0, to s = 0,1, then the β parameters can be estmated n a consstent way usng any methods suggested so far. However, t s not possble to estmate a consstent parameter ρ, and the dea s to use nstrumental varables to get 13 consstency. One possblty s to use the varables yt-j and y t-j, where the followng property wll be fulflled: E [( y t -ρ Y t-1 -β 1 x t -β x t-1 )y t-j ] = E [( y t -ρ y t-1 -β 1 x t -β x t-1 ) y t-j ] = 0 for (j =,..., t-1; t =,..., T). If E(x t-s ε,) 0 and E( α,x t ) 0 to s = 0,1, OLS and LSDV do not provde consstent estmatons of β. We have to use the regressors n frst dfference and nstruments to xt e x t-1, where a good example wll be x t- or x t-, followng the suggeston from Hsao and Anderson. t j 13 Anderson & Hsao suggested these estmators. See Arellano & Bond (1990:.78).

8 Arellano & Bond (1990) suggests an alternatve approach usng GMM (Generalzed Moments Method) based on equaton (3): y t = ρ y + β x + β x + ε ε (4) t 1 1 t t 1 t t 1 There are two basc dfferences between (3) and (4): a) the fxed effect, α I, presented n (3) was elmnated n (4) by dfferentaton and; b) frst order autocorrelaton was ntroduced n (4). Even though the estmator HD (Anderson & Hsao proposton) allows one to obtan consstent estmators, t does not have the desred effcency property. Effcency s present due to automatc autocorrelaton n the dsturbance terms and the eventual presence of heteroskedastcty also would result n effcency problems. 14 IV Emprcal Fndngs At a frst glance, t s mportant to hghlght smlar features observed among Latn Amercan economes. Fluctuaton n economc growth over tme snce 1970 s has been assocated to an ncrease n the degree of trade openng and n foregn lqudty, regardless of the ndcator examned. But we need to know whether experences of hgh economc growth are followed by hgh foregn lqudty and a hgher degree of trade openness. Although we are studyng a large number of emergng countres, t should be mentoned that there are many dfferences among them over the perod consdered. We can see that Chle has grown at ncreasng rates, has hgh trade openng and faced ncreasng nternatonal lqudty. On the other hand, we have economes lke Brazl where for each decade the growth rate has been lower than the secular one, the degree of trade openness s low, but faces an ncreasng nternatonal lqudty throughout the past decades. Mexco can be consdered as an ntermedate case showng a sustanable long run economc growth, a strong process of trade openng and hgh foregn lqudty. Regardng the other economes, we can say that there are unclear sgns n terms of trade openng or even when we try to take nto account the role of lqudty to economc growth. As a general rule, we can say that each Latn Amercan country has experenced an ncrease n nternatonal lqudty durng the last decade when comparng to hstorcal levels, even though economc growth rates have not followed the same path for most of them. 15 A second feature to be hghlghted s that the degree of trade openng averages around 6%, wth a hgh dsperson wthn the regon, wth a coeffcent of varaton near 80%. On the other hand, the nternatonal lqudty ndcators have ncreased throughout the last decades but wth a hgh dsparty among countres. Based on ths, one ssue that comes to 14 Arellano and Bond (1990) suggest Hausman and Sargan tests to analyze whether or not equaton (3) specfcaton s the rght choce. Sargan (1958, 1988) proposed a test of overestmaton where the dea s to verfy f the nstruments used are orthogonal to estmated resduals. The Hausman test on coeffcents of lagged varables can be mplemented n a sequental way. In ths case, frst lag s not a vald nstrument snce t wll generate correlaton between the varable and the resdual, such as the estmaton usng GMM where only n ths condton the estmaton wll be nconsstent. When the null hypothess s rejected, t s an evdence of frst order autocorrelaton. Then, for the statstc of Hausman test, our null hypothess s that the fxed effect model s the rght one and the alternatve hypothess s that the random effect model s the rght. The statstc β EF - β RE tends to zero under the null and to some dfferent value from zero under the alternatve. More specfcally, under Ho, ' 1 the Hausman statstc s: HS = ( β ) ( ( )) ( ) wth X dstrbuton wth K degrees of FE β RE Var β FE β RE β FE β RE freedom. 15 Brazl, Ecuador, Colomba and Paraguay are good examples. See table A.

9 mnd and should be ponted out s the exstence of a structural heterogenety n many dmensons of the analyss, as we can see n table 3A, wth dfferences n terms of economy sze, real GDP varaton coeffcent (143%), nternatonal lqudty (Bolva, Ecuador and Paraguay for low levels), trade openng (Chle, Bolva, Ecuador, Venezuela and Mexco wth hgher than average ndex). It s dffcult to say that economes wth lower GDP levels have lower degree of trade openng, specal when consderng Brazl, whch s the largest economy of the regon, but t s far to say that economes wth hgher GDP levels faces a hgher degree of foregn lqudty, wth some varaton over tme. As we know, estmaton usng panel data (pooled cross-secton and tme-seres data) allows us to explot the tme-seres nature of the relatonshp between lqudty and trade openng wth respect to growth. It s mportant to menton that n a pure cross-country nstrumental varable (IV) regresson, as n most ntal emprcal studes, any unobserved country-specfc effect becomes part of the error term, brngng up a problem of bas n the coeffcent estmates. On the other hand, the dynamc panel approach offers some advantages when compared to OLS estmaton, where the emprcal results has shown some mprovement on prevous efforts to examne the lnk between nternatonal fnancal ntegraton and growth. We estmate the model (1) usng dfferent methods of analyss, ncludng the fxed effect and random effects models, both for a statc and a dynamc approaches. In a panel data settng we have tme-seres observatons for multple economes, and the tme-seres observatons cover the same perod, what s called a balanced panel. Our statc panel data estmators were obtaned by OLS n levels, by GLS (OLS resduals) and by Maxmum Lkelhood (ML), where the dynamc panel data estmators were obtaned usng ML one step GMM (Generalzed Methods of Moments) estmaton. The fxed effect model was estmated by LSDV (Least Square Dummy Varable). The frst Wald test for the sgnfcance on all varables except the dummy (whch s the constant term), s the χ equvalent to the overall F-test. The next Wald test reports the sgnfcance of the constant term, and s just the square of the t-value. The AR(1) test s for frst order seral correlaton, wtch s sgnfcant when one varable s consdered. And, more generally speakng, the Sargan test deals wth the over dentfyng restrctons. The fgure 1A shows actual and ftted values, cross-plot between both of them and resduals for only one estmaton that we consder reasonable, that s usng ML one-step estmaton. The econometrc results from our panel estmaton are summarzed on table 5A. There are three mportant lessons to be hghlghted. Frst, observng only the parameter estmaton for trade openng t s dffcult to conclude that a hgh degree of trade openng can explan hgh economc growth for most specfcatons, except for the OLS (pooled regresson) where we found a negatve and sgnfcant coeffcent. All remanng specfcatons have shown dfferent coeffcents (negatve and postve) but all of them are not statstcally sgnfcant. One should remnd that the OLS estmaton (pooled regresson) gnores the panel aspect of the data, n other words, the country-specfc effect s not captured by the model. In the specfcaton OLS-Dff, we take frst dfferences removng countryspecfc effects and the ntercept. The least-squares dummy varable (LSDV) estmaton reports smlar results, except that the coeffcents on the country dummes are reported. By usng ths specfcaton t s dffcult to accept the dea that a hgher degree of trade openng explans a low real GDP growth rate.

10 Second, t s far to conclude that foregn lqudty measured ether by the rato of foregn reserves to mports (R / M) or by external debt to exports (D / X), can explan mprovements n real growth rates n Latn Amercan economes, ether usng the ML onestep estmaton or the LSDV estmators for the frst proxy of nternatonal lqudty. It s mportant to emphasze that the coeffcents have the expected sgn (postve for β 1 and negatve for β and β 3 ). Fnally, we have an mportant concluson regardng country sze and the effect on long-run real economc growth. As we have already sad, although we are analyzng emergng economes wthn the same regon, there are many structural dfferences among them, especally n terms of country sze. On one sde we have small economes wth GDP of approxmately US$7,5 bllons and US$9,5 bllons (Bolva and Paraguay, respectvely), whle on the other we have Brazl wth a GDP of US$700 bllons (Brazl), not to menton economes wth ntermedate sze lke Argentna (GDP around US$90 bllons and Venezuela wth US$80 bllons). Because of these dspartes we ntroduced the real GDP n the equaton for economc growth rate to capture possble country sze effects on long-run growth. All coeffcents estmated wth dfferent econometrc technques have expected sgns (negatve) but are not statstcally sgnfcant except for the LSDV cases. Ths s an ndcaton country sze matters and that large economes tend to face lower economc growth rates over the long run. Once we have analyzed our emprcal results for Latn Amercan economes snce early 1970s and after comparng them wth the emprcal evdence reported by the lterature and summarzed n the frst secton of the paper, we can say that t s dffcult to fnd a stable assocaton captal account lberalzaton and growth and for trade openng and growth. Our emprcal results follow the same trend from the lterature, but at the same tme we could fnd some evdence lnkng nternatonal lqudty and growth, but not for trade openng and growth. V Fnal Consderatons One of the conclusons we can draw from the emprcal fndngs s the dffculty to fnd a stable relatonshp assocatng nternatonal lqudty and growth, whch s condtoned on the proxy used for foregn lqudty and to the estmaton method used for panel data analyss. On the other hand, there s no emprcal evdence for a lnk between trade openness and growth n Latn Amercan countres snce 197. Stronger evdence of a sgnfcant assocaton between foregn lqudty and growth was found when we use the concepts of external debt relatve to exports, followed by the case when usng foregn reserves relatve to mports, but the result does not hold when we measure lqudty as the rato of external debt relatve to GDP. Comparng these results to the other ones n the lterature, we beleve that t s dffcult to reject the dea that ncreasng n foregn lqudty does not have a sgnfcant mpact on long-run economc growth, although we can accept the dea that captal account lberalzaton n developng countres (Latn Amerca ncluded) plays a decsve role n real GDP growth. On the other hand, we have to consder the presence of heterogenety across countres, expressed by dfferent country sze, degree of foregn lqudty, and degree of trade openness. As we surveyed n the second secton of ths paper, dfferent studes have found a

11 consderable lnk between nternatonal lqudty (fnancal openng) and growth as suggested by Edwards (001) among others, or no lnk at all as ponted out by Rodrguez & Rodrk, (001). A fnal word based on the analyss of Latn Amerca economc performance over the last decades should emphasze that economes wth a hgher degree of fnancal and trade openng are not necessarly the ones wth hgher growth rates, but the ones who face a hgher nternatonal lqudty may be more sutable to sustan a hgher economc growth rate over tme. Bblography Anderson, T.W. e Hsao, C. (198). Formulaton and Estmaton of Dynamc Models usng Panel Data. In: Journal of Econometrcs, 18, 47-8. Arellano, M. e Bond, S. (1991). Some Tests of Specfcaton for Panel Data: Monte Carlo Evdence and an applcaton to employment equatons. In: The Revew of Economc Studes, vol 58(), nº 194, Aprl. Arteta, Carlos, Barry Echengreen and Charles Wyplosz, 001, On the Growth Effects of Captal Account Lberalzaton, NBER Workng Papers, No. 8414, August, 001. Baldwn, Robert E. 003. Openness and Growth: What s the Emprcal Relatonshp? NBER Workng Papers, No. 9578, March, 003. Barro RJ, NG Mankw and X Sala--Martn 1995 Captal Moblty n Neoclasscal Models of Growth, Amercan Economc Revew, 85(1), pp 103 115. Bekaert, Geert, Campbell Harvey, and Chrstan Lundblad. 001 Does Fnancal Lberalzaton Spur Growth? NBER Workng Paper No. 845. Blackburn, K and Vctor T Y Hung 1998 A Theory of Growth, Fnancal Development and Trade. Economca, 1998, vol. 65, Número 57, pages 107-4. DeGregoro, Jose, Sebastan Edwards, and Rodrgo O. Valdes. 000 Controls on Captal Inflows: Do They Work? NBER Workng Paper No. 7645 (Aprl) (Cambrdge, Massachusetts: Natonal Bureau of Economc Research). Edson, H., Ross Levne, Luca Rcc and Torsten Slok.00 Internatonal Fnancal Integraton and Economc Growth, No 9164 n NBER Workng Papers, Natonal Bureau of Economc Research, 00. Edson, Hal J., and Carmen Renhart. 001 Stoppng Hot Money, Journal of Development Economcs, Vol. 66, pp 533.53. Edson, Hal J., and Francs E. Warnock, 001 A Smple Measure of the Intensty of Captal Controls, IMF Workng Paper 01/180. Edwards, Sebastan 001 Captal Moblty and Economc Performance: Are Emergng Economes Dfferent? NBER Workng Paper No. 8076. Echengreen, Barry, 001 Captal Account Lberalzaton: What Do Cross-Country Studes Tell Us?' Mmeo, Unversty of Calforna, Berkeley. Echengreen B. and Davd Leblang. 00. Captal Account Lberalzaton and Growth: Was Mr. Mahathr Rght? NBER Workng Papers 947, December 00. Feldsten, Martn, and Charles Horoka, 1980 Domestc Savng and Internatonal Captal Flows, The Economc Journal, Vol. 90, No. 358, pp. 314.9. Frees, E. W (003). Panel Data Analyss. (http://nstructon.bus.wsc.edu/jfrees/gb806.htm). Greene, W. H. (1999). Econometrc Analyss, Fourth edton, Prentce Hall. Grll, Vttoro, and Gan Mara Mles-Ferrett, 1995 Economc Effects and Structural Determnants of Captal Controls, IMF Staff Papers, Vol. 4, No. 3, pp. 517.51.

Hansen, L. P. (198). Large Sample Propertes of Generalzed Methods of Moments. Econometrca, 50, 109-1054. Hsao, C. (1986). Analyss of Panel Data. Cambrdge: UP. (Econometrc Socety Monographs, n. 11). Internatonal Monetary Fund. Annual Report on Exchange Arrangements and Exchange Restrctons, Varous Issues. Klen, M and Govann Olve 1999 Captal account lberalzaton, fnancal depth, and economc growth. No 7384, NBER Workng Papers, October, 1999. Kraay, Aart, 1998, In Search of the Macroeconomc Effects of Captal Account Lberalzaton, Mmeo, The World Bank. Levne, R. and Sara Zervos 1998 Stock Markets, Banks, and Economc Growth. Amercan Economc Revew, 1998, vol. 88, ssue 3, pages 537-58 Levne, Ross, and Davd Renelt, 199, A Senstvty Analyss of Cross Country Growth Regressons, Amercan Economc Revew, pp. 94.63. McLean, B. and S. Shrestha. 00 Internatonal Fnancal Lberalzaton and Economc Growth, Research Dscusson Papers, Reserve Bank of Australa, January, 00. Obstfeld M. 1994 Rsk-takng, Global Dversfcaton and Growth, Amercan Economc Revew, 84(5), pp 1310 139. Qunn, Denns 1997 The Correlates of Change n Internatonal Fnancal Regulaton, Amercan Poltcal Scence Revew, Vol. 91, No. 3, pp. 531.51. Rodrguez, Francsco and Dan Rodrk.001 Trade Polcy and Economc Growth: A Skeptc s Gude to the Cross-Natonal Evdence, n Ben Bernanke and Kenneth S. Rogoff, eds., NBER Macroeconomcs Annual 000, Cambrdge, MA, MIT Press. Rodrk, Dan. 1998 Who Needs Captal-Account Convertblty?, Harvard Unversty, Mmeo, February, 1998. Sachs, Jeffrey, and Andrew Warner.1995 Economc Reform and the Process of Global Integraton, Brookngs Papers on Economc Actvty, Vol. 1, pp. 1.118. 1

Table 1A: Overvew of Emprcal Studes on Fnancal Openng and Growth Author (Year) Klen (003) Echengreen and Leblang (00) Edson, Levne, Rcc and Slok (00) Edson, Klen, Rcc and Sloek (00) McLean and Shrestha (00) Edwards (001) Arteta, Echengreen. & Wyplosz (001) Countres (Perod) 85 (1976-1995) 1 (1880-1997) 47 (1975-1995) Measures of Captal Account Lberalzaton Share Average Qunn Share 57 (1976-000) Share 89 (1976-1995) Share, Qunn, and Dates of Stock Market Lberalzaton 40 (1976-1995) Captal Flows 6 (1980-1989) 59 (1980-1989) Qunn Share Qunn Sachs-Warner Non fnancal Openness Dependent Varable Other Varables Estmaton Technque (Methodology) Change n natural log of real per capta ncome Growth of real percapta GDP Real per capta GDP growth Real per capta GDP growth Growth rate n real GDP per capta Average real GDP growth Average rate of Total Factor Productvty growth durng 1980s Rate of growth of real GDP per capta Log of real per capta ncome n 1976, secondary school enrollment rate, average rate of I / GDP (1974-78), populaton growth rate (1976-95), dummy for Afrca. Log of ncome per capta relatve to the US, prmary and secondary school enrollment, captal control, government consumpton, nflaton, trade openness, domestc and nternatonal crses. Intal ncome (1980), average years of schoolng, fscal balance / GDP, nflaton rate (CPI), prvate credt / GDP, stock market total value traded / GDP, corrupton n government, FDI and Portfolo nflows and outflows / GDP. Log Real per capta ncome; log secondary school enrollment rate; average I / GDP rato; populaton growth rate; dummy Afrca. Instruments: Government consumpton / GDP, Imports / GDP, Dummy for Latn Amerca and East Asa. OLS and IV (Instrumental Varables) (Cross Secton) Dynamc Panel OLS (cross secton) SLS IV (cross secton) GMM dynamc panel OLS and IV Stock of human captal (years of secondary educaton), level of real per capta GDP, openness to trade (X + IM) / GDP, government consumpton, Panel (average over fve nonoverlappng years) black market exchange rate premum, FDI, Portfolo and Bank Flows, Bankng sector sze, ndex of law and contract enforcement, ndex of accountng standards, crss dummy, US nterest rate, terms of trade. Investment / GDP, number of years of schoolng, and log of real GDP per capta n 1965 Real nvestment / GDP, average years of schoolng, log of GDP per capta, fnancal depth, law and order ndex, trade openness dummy, black market premum, dstance to the equator, OECD membershp dummy, languave varables, landlocked naton dummy, sland naton dummy. WLS, WSLS, SURE and W3SLS and IV (Cross Secton) OLS (cross secton) WLS and ULS wth IV (cross secton) Movng Average Panel Data Man Results Inverted U-shaped relatonshp between captal account (KA) openness and ncome per capta. Mddle ncome countres benefts from KA openness whle rch or poor countres do not face postve effects. Net effect of captal controls on growth s postve n perods of fnancal nstablty but negatve when n the absence of crses. Captal account lberalzaton s not a panacea and ts benefts domnate the costs when the economes have a robust fnancal system Internatonal fnancal ntegraton does not accelerate economc growth Mxed evdence that that captal account lberalzaton promotes long-run economc growth. Fnd some support for a postve effect of KA lberalzaton on growth, especally for developng countres (East Asa) FDI and Portfolo nflows mproves economc growth, whle bank nflows have a negatve effect. KA lberalzaton sgnfcantly rases GDP growth. Evdence suggestng that an open captal account has a postve effect on long-run growth only after certan degree of economc development (support the dea of an optmal sequencng for KA lberalzaton). Fnd ndcatons of a postve assocaton between KA lberalzaton and growth, but the effect changes wth tme and how KA lberalzaton s measured. More evdence of a correlaton between KA lberalzaton and growth when ncludng other dmensons of openness. Share s proporton of years that IMF s AREAR shows open captal accounts (bnary measure of restrctons on captal transactons) Qunn s Qunn s 0 4 measure of captal account ntensty Qunn s change n value of Qunn 0 4 Volume s measure of volume of captal flows Sachs-Warner openness dummy, defned as a bnary varable equal to one f none of the fve followng crtera holds: the country had average tarff rates hgher than 40 per cent, ts nontarff barrers covered on average more than 40 per cent of mports, t had a socalst economc system, the state had a monopoly of major exports, and ts black market premum exceeded 0 % 13

14 Table 1A: Contnued Author (Year) Levne, Loayza and Beck (1999) Bekaert, Harvey & Lundblad (001) Klen & Olve (000) Levne, Loayza and Beck (1999) Countres (Perod) 74 (1960-1995) 95 (1980-1997) Measures of Captal Account Lberalzaton Lqud labltes of the fnancal system / GDP Assets of depost money banks / assets of depost money banks + central bank assets Credt by depost money banks and other fnancal nsttutons to the prvate sector / GDP Offcal Dates of Stock Market Lberalzaton 67 (1976-1995) Share 74 (1960-1995) Lqud labltes of the fnancal system / GDP Assets of depost money banks / assets of depost money banks + central bank assets Credt by depost money banks and other fnancal nsttutons to the prvate sector / GDP Dependent Varable Other Varables Estmaton Technque (Methdology) Real per capta GDP Growth rates n per capta ncome Growth n ncome per capta; Change n Fnancal Depth (FD) as a functon of Share; Per capta ncome growth as a functon of FD (and ntal FD). Real per capta GDP Government sze, log of ntal ncome per capta, openness to trade, nflaton, average years of secondary schoolng, black market premum, lqud labltes, prvate credt, dummy, growth rate of terms to trade, legal orgn I / GDP, secondary school enrollment, sze of government sector, C / GDP, Trade / GDP, populaton growth, lfe expectancy Msmanagement (nflaton and 1 + black market exchange premum), X + IM / GDP, lqud labltes / GDP, domestc credt / GDP, prvate bank rato, nflaton (CPI), black market premum, I / GDP, G / GDP Government sze, log of ntal ncome per capta, openness to trade, nflaton, average years of secondary schoolng, black market premum, lqud labltes, prvate credt, dummy, growth rate of terms to trade, legal orgn IV (cross secton) and Panel (GMM) Panel IV (Cross Secton) IV (cross secton) and Panel (GMM) Man Results Fnancal development s postvely assocated wth economc growth. Stock market lberalzaton sgnfcantly contrbutes to growth, wth largest effects shortly after lberalzaton. Equty market lberalzaton, on average, lead to a one per cent ncrease n annual real economc growth over a fve year perod. I / GDP ncreases after the lberalzaton due to foregn captal nflows and a worsenng n the trade balance. Sgnfcant effect of Share on fnancal deepness (FD),though results seem to be drven by OECD countres n sample. Sgnfcant effect of nstrumented values of FD and FD on growth. Fnancal development s postvely assocated wth economc growth. Rodrk (1998) 100 (1973-1996) Share Growth n per capta GDP Kraay (1998) 117 (1985-1997) Share; Qunn; Volume Qunn (1997) 58 (1960-1989) Qunn Grll & Mles- Ferrett (1995) 61 (1971-1994) Share Growth n per capta GDP Growth n per capta ncome Growth n per capta ncome for fve-years non-overlappng perods I / GDP, Inflaton, Intal per capta GDP, ntal secondary enrollment rate, ndex of qualty of government nsttutons, regonal dummes. OLS (Cross Secton) No evdence of a sgnfcant effect of Share on growth of ncome per capta. Polcy choces wth respect to the captal account are endogenous (depend on economc performance) Investment (I), nflaton OLS and IV (Cross Secton) No effect of Share or Qunn on Growth. Coeffcent on Volume sgnfcant and postve. Intal GDP per capta, I / GDP, KA lberalzaton sgnfcantly rases growth, though no populaton growth and secondary OLS (Cross Secton) regresson s presented wth both captal controls and school enrollment rates. openness. Instrumental Varables (Cross Secton) No Evdence of a sgnfcant effect of Share on growth of ncome per capta.

Table A. Latn Amercan Economes: Selected Indcators (1960-000) 15 Argentna Bolva Brazl Mean dy Open FL01 FL0 FL03 dy Open FL01 FL0 FL03 dy Open FL01 FL0 FL03 1960/70 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 5,90 1,83 16,97 n.a. n.a. 1970/80,93 3,58 60,7 6,61 3, 4,3 17,44 9,65 33,67 3,66 8,47 4,67 6,0 9,53 4,04 1980/90-0,73 6,53 51,9 30,17 7,70-0,41 5,60 9,5 107,5 7,78,99 7,75 43,90 3,9 5,09 1990/00 4,54 13,79 87,50 46,15 6,53 4,06 33,46 40,3 95,3 6,50 1,8 11,81 85,53 30,95 4,95 Chle Colomba Ecuador Mean dy Open FL01 FL0 FL03 dy Open FL01 FL0 FL03 dy Open FL01 FL0 FL03 1960/70 4,36 7,6 15,57 n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. n.a. 1970/80,48 15,5 0,7 8,98 3,75 5,81 7,48 60,5 11,79 3,1 9,0 17,90 43,90,03 1,85 1980/90 4,39 8,00 64,08 7,60 5,16 3,40 13,76 71,87 6,74 4,09,37 30,35 36,50 8,1 4,75 1990/00 6,43 39,99 96,88 5,37,61,84 1,37 87,67 34,90 3,37 1,88 39,36 49,1 10,1 4,75 Paraguay Peru Urugua Mean dy Open FL01 FL0 FL03 dy Open FL01 FL0 FL03 dy Open FL01 FL0 FL03 1960/70 4,7 4,40 14,01 n.a. n.a. 5,5* n.a. 17,6 n.a. n.a. 1,30 n.a. 11,31 n.a. n.a. 1970/80 7,9 9,38 74,47 10,8,1 3,94 7,00 5,76 0,65 5,7,70 94,00 5,73 8,43,37 1980/90 4,0 14,13 114,10 33,79 6,4 0,35 1,00 58,80 36,3 5,65 0,71 60,00 31,71 30,47 3,73 1990/00,19 35,88 4,56 30,03 3,04 3,9 1,00 100,1 66,45 7,13 3,1 67,00 38,77 38,19 3,41 Venezuela Mexco Mean dy Open FL01 FL0 FL03 dy Open FL01 FL0 FL03 1960/70 4,81 11,00 9,06 n.a. n.a. 6,79 3,03 n.a. n.a. n.a. 1970/80 3,97 3,00 88,4 16,14 1,10 6,43 5,41 n.a. 16,0 7,10 1980/90-0,17 39,00 8,51 71,38 3,,9 17,93 n.a. 48,09 4,4 1990/00,30 38,00 96,4 59,88,61 3,35 43,99 n.a. 58,44,54

Table 3A. Correlaton Matrx Orgnal Sample (197-000) Fve years mean sample (197-000) 16 Open FL01 FL0 FL03 y dy Open FL01 FL0 FL03 y dy Open 1 1 FL01 0,14 1 0,1 1 Fl0 0,1 0,08 1 0,1 0,1 1 FL03-0,36-0,35 0,17 1-0,36-0,44 0,17 1 y 0,3-0,5-0,09 0,11 1-0,31-0,3-0,08 0,13 1 dy -0,07 0,003-0,15-0,15-0,01 1-0,11 0,04-0,19-0,3-0,07 1 Table 4A. Basc Statstc Orgnal Sample (197-000) Fve years mean sample (197-000) Open FL01 FL0 FL03 y(us$ m) dy(% per year) Open FL01 FL0 FL03 y(% per year) dy Mean 6,3 83,44 37,35 4,1 116.730,00 3,7 5,94 83,76 36,3 4,17 115300 3,34 SD 0,85 51,63 3,19,7 167.60,00 4,77 0,37 44,88 31,75,08 16440 3,16 CV 79, 61,88 86,18 53,9 143,9 145,87 78,53 53,58 87,63 49,88 14,45 94,61 SD = Standard devaton; CV = Coeffcent of varaton measured by rato of standard devaton relatve to the mean.

Table 5A. Results of Estmaton n Panel Analyss (197-000) yˆ = α + β Open + β FL01 + β FL0 + β FL03 + β y + ε t 0 t 1 t t 3 t 4 t t 17 Methods α β 0 β 1 β β 3 β R N AR(1) RSS Wald Wald Sargan 4 N(0,1) (Number of (jont) X (dummy) Test parameters) X X (0) OLS (pooled regresson) -0,0001 (-1,75) -0,036 (-,30) -0,005 (-0,59) -0,01 (-1,49) -0,46 (-,43) -0,0001 (-1,75) 0,1 66 1,089 [0,76] 566,0 (6) OLS-Dff 0,05 [1,34] 0,016 [1,58] -0,07 [-,36] -0,46 [-1,05] -0,0006 [-0,93] 0,3 55 676,53 (6) ML by 1-Step 8,31 (,55) -0,017 (-0,791) 0,06 (1,99) -0,345 (-0,34) -0,611 (-3,15) -0,0001 (-0,43) 66 1,84 [0,068] 691,56 (6) 1,6 [0,001] 10,65 [0,001] 17,89 [0,594] GLS (w/b) 7,59 (4,8) -0,036 (-1,80-0,008 (-0,84) -0,009 (-0,83) -0,47 (-,18) -0,0001 (-1,16) 0,1 66 58,63 (6) LSDV (Fxed 7,305 0,008 0,031-0,041-0,435-0,0001 0,38 66-0,5890 398,08 4,83 77,46 Effect) (3,43) (0,14) (,41) (-1,39) (-1,6) (-3,93) [0,556] (16) [0,000] [0,000] LSDV (Fxed 5,85-0,00 0,043-0,0006-0,3098-0,001 0,59 66-0,7356 63,9 66,11 559,9 Effect) 3 (3,53) (-0,083) (3,00) (-0,180) (-1,07) (3,15) [0,46] (1) [0,000] [0,000] Notes: (1) Standard Errors n branches; () Dummes (Indvduals) are present statstclly sgnfcants, execpt ndvduals I1, I e I6. (3) Dummes (Tme) are statstc and negatve sgnfcant to 1976-1980 and 1996-000. Graphc 1A. Estmated and Ftted Real Economc Growth and Resdual from Panel Analyss 1 5 1 0 5 0 dy F tte d 1 5 1 0 0 1 0 0 3 0 4 0 5 0 6 0 7 0 dy F t t e d 5 0-1 0 1 3 4 5 6 7 8 r :d y ( s c a le d ) 0-0 1 0 0 3 0 4 0 5 0 6 0 7 0