FOREIGN LIQUIDITY, ECONOMIC OPENNING AND GROWTH IN LATIN AMERICAN ECONOMIES

The main concern of our empirical study is to shed light to the question of whether or not and in which direction long-run growth has been associated with financial (liquidity) and trade opening since early 1970s using a panel data approach for 11 Latin American countries. Previous empirical studies reported mixed results in terms of finding a stable association between capital account liberalization and growth or even for trade opening and growth. Our empirical results suggest an important link between international liquidity and growth, but the same does not apply for trade opening and growth. Key-Words Foreign liquidity, Trade Opening; Economic Growth; Latin America. JEL Classification F41; F43; C23. Resumo O propósito central deste nosso estudo empírico é discutir as relações entre o crescimento econômico de longo prazo e a liquidez externa, assim como deste com a abertura comercial, desde começo dos anos 70, usando a abordagem de painel para onze economias da América Latina. Estudos anteriores encontraram resultados controversos em termos de relações estáveis entre a liberalização da conta de capital e o crescimento econômico, ou ainda, entre a abertura comercial e o crescimento de longo prazo do PIB. Nossos resultados empíricos sugerem um importante vínculo entre a liquidez externa e este crescimento, mas o mesmo não pode ser observado para a relação entre abertura comercial e crescimento econômico. Palavras-Chaves Liquidez Externa; Abertura Comercial; Crescimento Econômico,; América Latina. Classificação JEL: F41; F43; C23.


I -Introduction
The main concern of our empirical study is to shed light to question of whether or not and in which direction long-run growth has been associated with financial (liquidity) and trade opening since early 1970s using a panel data approach for 11 Latin America countries.It is well known by the literature that more recently, developing countries have faced periods of international financial crises with significant outcomes generally associated with lower economic growth rates during the last two decades when compared to historical rates during the sixties and seventies.Given the fact that most developing and emerging economies have gone through a period of opening trade and capital accounts since early 1990s we want to investigate if and how different measures of liquidity (three) and trade flows can be part of the explanation for long-run economic growth in Latin American economics.
It is fair to relate trade and financial opening in the sense that the first one involves an increase in trade of goods while the second in trade of capital since foreign investment is a form of intertemporal trade, and based on the argument that trade benefits growth one can argue that higher capital mobility will have similar impact on growth.The implications of this line of thought is in the background of our research where we will be using a panel data for Latin American economies to investigate possible implications of changes in international liquidity and trade opening to long-run growth.One has to remind that international liquidity is associated with capital account liberalization in the sense that without the latter (no capital mobility) international financial markets has a limited role to be played in fostering higher economic growth rates.
The paper is divided in three sections other than this one and final considerations.Section two develops a general review of the literature on financial opening and growth, section three deals with some methodological issues related to panel data analysis and variable description, and section four summarizes the main empirical findings.We can draw a general conclusion from the present work, suggesting no clear link between trade openness and growth, even though there is evidence that high international liquidity and an improvement in long-run growth rates are somehow associated for Latin American economies.

II -Financial Opening and Growth: Theory and Empirics
The main task of this section is to summarize what theory says regarding the relationship between financial opening and growth, and also analyze the empirical findings associated to this issue.
One can say that theory has no unambiguous prediction of whether or not capital account liberalization enhances growth and the empirical evidence can be considered inconclusive.At a first look, there are two channels through which capital account liberalization affects growth.The first one is associated with the argument that higher capital mobility increases the domestic investment rate since capital flows towards countries where capital is relatively scarce and where the marginal productivity of capital is higher, and the outcome is higher economic growth rates.A second possible channel can be associated with capital flows to sectors with higher rates of return (portfolio diversification) when financial markets does hot operate with significant distortions and the outcome of capital account liberalization tend to generate a more efficient resource allocation and a faster rate of economic growth. 1he predictions offered by theories regarding the international financial integration effects on growth can be considered conflicting to some extent.International financial integration facilitates risk-sharing (diversification) enhancing capital allocation and economic growth, but it can have a negative impact on growth if it is implemented under economic conditions where the existence of distortions is the rule rather the exception. 2The policy prescription to extend the process of financial integration in less-developed countries is controversial. 3 significant number of literature reviews on capital account liberalization and growth have been developed in the last five years and here we will briefly survey the most important empirical results associated with them. 4he first studies on capital account liberalization and growth have not find supportive results.Alesina, Grilli and Milesi-Ferretti (1995) found that financial open has small and insignificant effects on growth.Rodrik (1998) uses a similar approach for a larger sample and found no stable association between capital account liberalization and growth.Kraay (1998) find no link between economic growth and the IMF restriction measure.
On the other hand, Quinn (1997) develops an empirical analysis considering the impact of both capital account openness and the change in openness where the results suggest a positive association between the change in capital account openness and growth. 5Edwards  (2001) found a significant positive effect of capital account liberalization on growth, but the results were restricted to high income countries. 6 Klein and Olivei (2000) find a positive effect of capital account liberalization on growth for industrial countries, but not for lessdeveloped countries.Arteta, Eichengreen, and Wyplosz (2001) when introducing proxies for the degree of macroeconomic stability find some support for differences in the effect of capital account liberalization across countries.7 Among the studies on capital account liberalization and growth that includes some measure of trade openness as an additional (control) variable we can mention Eichengreen and Leblang (2002) and they find a positive and significant coefficient for trade opening and its impact on growth. 8Another empirical research on international financial liberalization and growth that includes trade openness as a control variable is McLean and Shrestha (2002)) where the coefficient shows up as positive and significant regardless if the sample includes both developed and developing countries or only the latter.Arteta, Eichengreen and Wyplosz (2001) use the Sachs-Warner trade openness measure and in the pooled regression the coefficient is positive and significant.9Klein and Olivei (2000) examines the impact of financial development on growth including a vector of control variables that are potentially related with a country's economic growth, where one of them is the 1986 ratio of exports plus imports to GDP and the results seems have not changed by the inclusion of such variable (open capital account increases financial depth and higher economic growth rates).Levine, Loayza and Beck (1999) uses a dynamic panel estimation with two sets of conditional information where in one of them openness to trade (log) is used and the results indicate a positive and significant effect for financial intermediation and growth.
Since we are including openness to trade defined as the ratio of the sum of exports imports relative to GDP as one of our variables to capture possible impacts to long-run growth in Latin America, a brief word on how the literature and empirical research has seen this relationship is necessary.The literature on openness to trade and growth has been characterized by many controversies in terms of associating openness with higher growth rates.Rodriguez and Rodrik (2001) is one of the empirical works that does not find such positive association in the sense that liberal trade policies does not guarantee faster growth rates.On the other hand, different empirical studies find that lower trade barriers together with a stable exchange rate system, sound monetary and fiscal policies help promoting economic growth.10

III -Variable Description and Econometric Model
The main model to be estimated and the variables description are the following: where: is the real growth rate of GDP; Open is the trade opening; is the foreign liquidity measured according to FL01 (the ratio between the foreign reserves and the imports), FL02 (the ratio between the external debts and real GDP), FL03 (the ratio between the external debt and exports), and is the real GDP.We are expecting a positive The estimation of equation ( 1) has been implemented using the original sample with annual data from 1972 to 2000, averaging the data for each five years, except for the first observation (1972)(1973)(1974)(1975).We can see, in table 3A the correlation matrix, where we obtained similar mean samples and dispersion measures, and the same applies for the correlation between the variables used in empirical research.We used the transformed mean sample to estimate model (1) using a panel data analysis.
Estimation using panel data has several advantages over purely cross-sectional estimation.First, besides considering the cross-country relationship between financial development (international liquidity) and growth, we also would like to take into account how financial development over time within a country may have an effect on the country's growth performance.Working with a panel helps gaining degrees of freedom by adding the variability of the time-series dimension to the analysis.Second, in a panel context, we are able to control for unobserved country-specific effects and thereby reduce biases in the estimated coefficients.Third, our panel estimator controls for the potential endogeneity of all explanatory variables, while the cross-sectional estimator presented by previous studies only controls for the endogeneity of financial development.The way our panel estimator controls for endogeneity is by using "internal instruments," that is, instruments based on lagged values of the explanatory variables.This method does not allow us to control for full endogeneity but for a weak type 11 .
The panel approach allows for two basic models: fixed and random effect models, both of them accepting static and dynamic specifications.The fixed effect model, also known as least square dummy variable (LSDV), is a generalization of an intercept-slope-constant model for panel analysis, introducing a dummy variable to capture the effects of omitted variables, that are constant over time.
11 To be precise, Levine, Loayza and Beck, (1999) assume that the explanatory variables are only "weakly exogenous," which means that they can be affected by current and past realizations of the growth rate but must be uncorrelated with future realizations of the error term.Thus, the weak exogeneity assumption implies that future innovations of the growth rate do not affect current financial development.This assumption is not particularly stringent conceptually and we can examine if it has statistical validity.Weak exogeneity does not mean that economic agents do not take into account expected future growth in their decision to develop the financial system; it just means that future (unanticipated) shocks to growth do not influence current financial development.It is the innovation in growth that must not affect financial development.
In this specification, the individual-effects can be freely correlated with the regressors.Their estimation is, in fact, the own estimation of the model of multiple regressions with binary variables for each one of the n units of the analysis, in such a way that the introduction of them will cause the intercept of the regression to be different for each one of these variables and pick up the heterogeneity among them.The ordinary least square (OLS) estimator will be a consistent and efficient estimator and it is know as LSDV.
The random-effect model specification treats the individual-specific effects as random variables.This model assumes no correlation between the individual effects and the other random variables, where the estimation was pursued using the Generalized Least Square (GLS).
One crucial question is to know which is the most appropriate model?According to Frees (2003) it depends on the available information and the estimation objectives.If, for example, the main concern of the analysis will be to test the effect of the variables where the individuals are classified in groups, then the random effect specification is more appropriate.In Hsiao (1999: 42): "The fixed-effects model is viewed as one in which investigators make inferences conditional on the effects that are in the sample.The random-effects model is considered as the one in which one can make unconditional or marginal inferences with respect to the population of all effect." A static panel-data model can be written as: (2) where: t λ and i η are time and individual specific effects respectively, x it is a vector of explanatory variables, N is the number of cross-section observations and NT is the total number of observations.
The main goal is to obtain a consistent estimator of β with the desired efficiency proprieties.The choice of the estimation technique to be used depends on the hypothesis assumed for the relationship between the error-term ( it ε ) and the regressors (x it ) in terms of random error and the fixed effect i η .In the more restrictive case, one can assume that E( i η ,x it ) = 0 (the orthogonality between the fixed-effect and the regressors) and E( it ε ,x it-s ) = 0 for any lag s.
One can use OLS (Ordinary Least Square) or LSDV (Least Square Dummy Variable) since both provide consistent estimators, but the second is the more efficient.If we do not consider the hypothesis of orthogonality between the fixed effect and the regressors, that is, if we assume E( i α ,x it ) ≠ 0, it is not possible to assume consistence for the OLS estimation, and LSDV should be the estimation choice since it is the only one that is consistent.Another consistent estimator is OLS using the first difference (FD-OLS) 12 , but some caution is necessary since it presents efficiency problems.
One can also assume that E( i α ,x it ) = 0 and E( i α ,x it ) ≠0. Comparing the estimated slopes for the fixed effect and the random effect models, one can say that: 1) assuming that the formulation of the fixed effects is right, so β EF is consistent and asymptotically efficient, and β RE is inconsistent.2) assuming that the formulation in terms of random effect is right, so β RE is consistent and asymptotically efficient, and β EF is consistent, as well.
According to Hsaio (1999:36), the GLS estimator (Generalized Least Square) is the weighted average between-groups and within-groups.The GLS estimator can converge to OLS or to LSDV.In the LSDV procedure (fixed effect model) the source of variation is not taken into account and OLS and LSDV can be considered as an example of all or nothing in terms of variation between groups.The procedure that considers i α as random allows for an intermediate solution and does not have to treat everyone as different or similar, according to GLS estimators.
In models ( 1) and ( 2), there are no lagged variables, nether regressors or explanatory variables.Incorporating such elements, we propose the following model: , for i = j and t = s, and , for all the other cases.
If we assume E( i α ,x it-s ) = E(ε it ,x it-s ) =0, to s = 0,1, then the parameters can be estimated in a consistent way using any methods suggested so far.However, it is not possible to estimate a consistent parameter ρ, and the idea is to use instrumental variables to get consistency.One possibility is to use the variables ∆y β t-j and y it-j 13 , where the following property will be fulfilled: If E(x it-s ε i ,) ≠ 0 and E( i α ,x it ) ≠ 0 to s = 0,1, OLS and LSDV do not provide consistent estimations of .We have to use the regressors in first difference and instruments to ∆x β t e ∆x t-1 , where a good example will be x it-2 or ∆x t-2 , following the suggestion from Hsaio and Anderson.Arellano & Bond (1990) suggests an alternative approach using GMM (Generalized Moments Method) based on equation (3): There are two basic differences between (3) and (4): a) the fixed effect, α I , presented in (3) was eliminated in (4) by differentiation and; b) first order autocorrelation was introduced in (4).Even though the estimator HD (Anderson & Hsaio proposition) allows one to obtain consistent estimators, it does not have the desired efficiency property.Efficiency is present due to automatic autocorrelation in the disturbance terms and the eventual presence of heteroskedasticity also would result in efficiency problems. 14

IV -Empirical Findings
At a first glance, it is important to highlight similar features observed among Latin American economies.Fluctuation in economic growth over time since 1970's has been associated to an increase in the degree of trade opening and in foreign liquidity, regardless of the indicator examined.But we need to know whether experiences of high economic growth are followed by high foreign liquidity and a higher degree of trade openness.Although we are studying a large number of emerging countries, it should be mentioned that there are many differences among them over the period considered.We can see that Chile has grown at increasing rates, has high trade opening and faced increasing international liquidity.On the other hand, we have economies like Brazil where for each decade the growth rate has been lower than the secular one, the degree of trade openness is low, but faces an increasing international liquidity throughout the past decades.Mexico can be considered as an intermediate case showing a sustainable long run economic growth, a strong process of trade opening and high foreign liquidity.Regarding the other economies, we can say that there are unclear signs in terms of trade opening or even when we try to take into account the role of liquidity to economic growth.As a general rule, we can say that each Latin American country has experienced an increase in international liquidity during the last decade when comparing to historical levels, even though economic growth rates have not followed the same path for most of them. 15  A second feature to be highlighted is that the degree of trade opening averages around 26%, with a high dispersion within the region, with a coefficient of variation near 80%.On the other hand, the international liquidity indicators have increased throughout the last decades but with a high disparity among countries.Based on this, one issue that comes to 14  Arellano and Bond (1990) suggest Hausman and Sargan tests to analyze whether or not equation (3) specification is the right choice.Sargan (1958,1988) proposed a test of overestimation where the idea is to verify if the instruments used are orthogonal to estimated residuals.The Hausman test on coefficients of lagged variables can be implemented in a sequential way.In this case, first lag is not a valid instrument since it will generate correlation between the variable and the residual, such as the estimation using GMM where only in this condition the estimation will be inconsistent.When the null hypothesis is rejected, it is an evidence of first order autocorrelation.Then, for the statistic of Hausman test, our null hypothesis is that the fixed effect model is the right one and the alternative hypothesis is that the random effect model is the right.The statistic β EF -β RE tends to zero under the null and to some different value from zero under the alternative.More specifically, under Ho, the Hausman statistic is: with X 2 distribution with K degrees of freedom.2A.mind and should be pointed out is the existence of a structural heterogeneity in many dimensions of the analysis, as we can see in table 3A, with differences in terms of economy size, real GDP variation coefficient (143%), international liquidity (Bolivia, Ecuador and Paraguay for low levels), trade opening (Chile, Bolivia, Ecuador, Venezuela and Mexico with higher than average index).It is difficult to say that economies with lower GDP levels have lower degree of trade opening, special when considering Brazil, which is the largest economy of the region, but it is fair to say that economies with higher GDP levels faces a higher degree of foreign liquidity, with some variation over time.
As we know, estimation using panel data (pooled cross-section and time-series data) allows us to exploit the time-series nature of the relationship between liquidity and trade opening with respect to growth.It is important to mention that in a pure cross-country instrumental variable (IV) regression, as in most initial empirical studies, any unobserved country-specific effect becomes part of the error term, bringing up a problem of bias in the coefficient estimates.On the other hand, the dynamic panel approach offers some advantages when compared to OLS estimation, where the empirical results has shown some improvement on previous efforts to examine the link between international financial integration and growth.
We estimate the model (1) using different methods of analysis, including the fixed effect and random effects models, both for a static and a dynamic approaches.In a panel data setting we have time-series observations for multiple economies, and the time-series observations cover the same period, what is called a balanced panel.
Our static panel data estimators were obtained by OLS in levels, by GLS (OLS residuals) and by Maximum Likelihood (ML), where the dynamic panel data estimators were obtained using ML one step GMM (Generalized Methods of Moments) estimation.The fixed effect model was estimated by LSDV (Least Square Dummy Variable).The first Wald test for the significance on all variables except the dummy (which is the constant term), is the equivalent to the overall F-test.The next Wald test reports the significance of the constant term, and is just the square of the t-value.The AR(1) test is for first order serial correlation, witch is significant when one variable is considered.And, more generally speaking, the Sargan test deals with the over identifying restrictions.The figure 1A shows actual and fitted values, cross-plot between both of them and residuals for only one estimation that we consider reasonable, that is using ML one-step estimation.

χ
The econometric results from our panel estimation are summarized on table 5A.There are three important lessons to be highlighted.First, observing only the parameter estimation for trade opening it is difficult to conclude that a high degree of trade opening can explain high economic growth for most specifications, except for the OLS (pooled regression) where we found a negative and significant coefficient.All remaining specifications have shown different coefficients (negative and positive) but all of them are not statistically significant.One should remind that the OLS estimation (pooled regression) ignores the panel aspect of the data, in other words, the country-specific effect is not captured by the model.In the specification OLS-Diff, we take first differences removing countryspecific effects and the intercept.The least-squares dummy variable (LSDV) estimation reports similar results, except that the coefficients on the country dummies are reported.By using this specification it is difficult to accept the idea that a higher degree of trade opening explains a low real GDP growth rate.
Second, it is fair to conclude that foreign liquidity measured either by the ratio of foreign reserves to imports (R / M) or by external debt to exports (D / X), can explain improvements in real growth rates in Latin American economies, either using the ML onestep estimation or the LSDV estimators for the first proxy of international liquidity.It is important to emphasize that the coefficients have the expected sign (positive for β 1 and negative for β 2 and β 3 ).
Finally, we have an important conclusion regarding country size and the effect on long-run real economic growth.As we have already said, although we are analyzing emerging economies within the same region, there are many structural differences among them, especially in terms of country size.On one side we have small economies with GDP of approximately US$7,5 billions and US$9,5 billions (Bolivia and Paraguay, respectively), while on the other we have Brazil with a GDP of US$700 billions (Brazil), not to mention economies with intermediate size like Argentina (GDP around US$290 billions and Venezuela with US$80 billions).Because of these disparities we introduced the real GDP in the equation for economic growth rate to capture possible country size effects on long-run growth.All coefficients estimated with different econometric techniques have expected signs (negative) but are not statistically significant except for the LSDV cases.This is an indication country size matters and that large economies tend to face lower economic growth rates over the long run.
Once we have analyzed our empirical results for Latin American economies since early 1970s and after comparing them with the empirical evidence reported by the literature and summarized in the first section of the paper, we can say that it is difficult to find a stable association capital account liberalization and growth and for trade opening and growth.Our empirical results follow the same trend from the literature, but at the same time we could find some evidence linking international liquidity and growth, but not for trade opening and growth.

V -Final Considerations
One of the conclusions we can draw from the empirical findings is the difficulty to find a stable relationship associating international liquidity and growth, which is conditioned on the proxy used for foreign liquidity and to the estimation method used for panel data analysis.On the other hand, there is no empirical evidence for a link between trade openness and growth in Latin American countries since 1972.Stronger evidence of a significant association between foreign liquidity and growth was found when we use the concepts of external debt relative to exports, followed by the case when using foreign reserves relative to imports, but the result does not hold when we measure liquidity as the ratio of external debt relative to GDP.
Comparing these results to the other ones in the literature, we believe that it is difficult to reject the idea that increasing in foreign liquidity does not have a significant impact on long-run economic growth, although we can accept the idea that capital account liberalization in developing countries (Latin America included) plays a decisive role in real GDP growth.On the other hand, we have to consider the presence of heterogeneity across countries, expressed by different country size, degree of foreign liquidity, and degree of trade openness.As we surveyed in the second section of this paper, different studies have found a considerable link between international liquidity (financial opening) and growth as suggested by Edwards (2001) among others, or no link at all as pointed out by Rodriguez & Rodrik, (2001).
A final word based on the analysis of Latin America economic performance over the last decades should emphasize that economies with a higher degree of financial and trade opening are not necessarily the ones with higher growth rates, but the ones who face a higher international liquidity may be more suitable to sustain a higher economic growth rate over time.Share is proportion of years that IMF's AREAR shows open capital accounts (binary measure of restrictions on capital transactions) Quinn is Quinn's 0 -4 measure of capital account intensity ∆Quinn is change in value of Quinn 0 -4 Volume is measure of volume of capital flows Sachs-Warner openness dummy, defined as a binary variable equal to one if none of the five following criteria holds: the country had average tariff rates higher than 40 per cent, its nontariff barriers covered on average more than 40 per cent of imports, it had a socialist economic system, the state had a monopoly of major exports, and its black market premium exceeded 20 % for i = 1, ..., N and t = 1,..., T where 0 = it Eε Ecuador, Colombia and Paraguay are good examples.See table