WPS5426 Policy Research Working Paper 5426 Trade and Economic Growth Evidence on the Role of Complementarities for CAFTA-DR Countries César Calderón Virginia Poggio The World Bank Latin America and the Caribbean Region Office of the Chief Economist September 2010 Policy Research Working Paper 5426 Abstract This paper examines the effects of trade on growth there is a robust causal link from trade to growth, but also among Central America-Dominican Republic Free Trade that the growth benefits from trade are larger in countries Agreement countries. To accomplish this task, the authors with higher levels of education and innovation, deeper collected a panel data set of 136 countries over 1960­ financial markets, a stronger institutional framework, 2010, and estimated cross-country growth regressions more developed infrastructure networks, a high level using an econometric methodology that accounts of integration with world capital markets, and less for unobserved effects and the likely endogeneity of stringent economic regulations. On average, rising trade the growth determinants. Following recent empirical has benefited growth in Central America-Dominican efforts, they tested whether the impact of trade openness Republic Free Trade Agreement countries. However, the on growth may be more effective after surpassing a lack of progress in structural reforms has not allowed "minimum threshold" in specific areas closely related to these countries to maximize the potential benefits from economic development. The analysis finds not only that trade. This paper--a product of the Office of the Chief Economist, Latin America and the Caribbean Region--is part of a larger effort in the department to understand the drivers of growth. This paper was prepared for the "DR-CAFTA and the Complementary Agenda" project conducted by the Central America CMU (LCC2C). Policy Research Working Papers are also posted on the Web at http://econ.worldbank.org. The author may be contacted at ccalderon@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Trade and Economic Growth: Evidence on the role of complementarities for CAFTADR countries* César Calderóna, Virginia Poggioa a The World Bank, 1818 H Street NW, Washington DC 20433, USA JEL Classification: F36, F41, F43 Key Words: Trade openness, complementarities, growth * This paper was prepared for the "DRCAFTA and the Complementary Agenda" project. We would like to thank Pravin Krishna and Rashmi Shankar for valuable comments. The views expressed here are those of the authors, and they should not be attributed to the World Bank, its Executive Directors, or the countries they represent. The usual disclaimer applies. 1. Introduction One of the salient features of the world economy has been the important surge in trade and financial globalization in the last two decades. Multiple free trade agreements and regional integration agreements are being celebrated --with more than 400 regional trade agreements in force by December 2008 according to WTO/GATT. In addition, world trade grew at least twice as fast as world output over the last two decades, thus deepening economic integration. In August 2004, the U.S. signed the CAFTADR free trade agreement with five Central American countries (Costa Rica, El Salvador, Guatemala, Honduras, and Nicaragua) and the Dominican Republic. According to the Office of the U.S. Trade Representative, U.S. exports to CAFTADR countries totaled US$ 26.3 billion in 2008, and the total trade between U.S. and CAFTADR in 2008 was US$ 45.6 billion. The Office also points out that CAFTA DR countries represent the third largest U.S. export market in the LAC region, after Mexico and Brazil.1 Theoretically, it has long been argued in the literature that trade stimulates longterm growth and that it can do so through multiple channels. International trade would allow countries to specialize in areas where they possess comparative advantage, expand potential markets and allow firms to exploit economies of scale, enable the diffusion of technological innovation and frontier managerial practices, and reduce incentives for firms to conduct rentseeking activities through higher market competition. Empirically, earlier works find evidence in support of the growthenhancing effects of trade. However, Rodriguez and Rodrik (2000) suggested that most of the evidence was not robust due to issues related to the measurement of trade openness and trade policy as well as econometric problems (i.e. endogeneity of trade measures and colinearity of trade and institutions). Also, Rodrik (2005) argued that policies towards trade openness may not render the same results for all countries since there is no unique mapping from economic principles to economic packages. Most of these criticisms have been tackled in recent empirical efforts by developing new identification strategies (Frankel and Romer, 1999), new trade indicators (Wacziarg, 2001), examining the tradegrowth correlation around episodes of policy changes (Wacziarg and Welch, 2008), and addressing the issue of mapping from principles to policies by assessing the role of complementarities between trade and other structural reforms in stimulating growth (Calderon, Loayza, and SchmidtHebbel, 2006; Chang, Kaltani and Loayza, 2009; Calderon and Fuentes, 2009). The goal of this paper is to assess the growth effects of trade among CAFTADR countries and, more specifically, to evaluate the structural areas that might become a constraint to reaping the growth benefits from growth. In this context, the paper argues that policy complementarities are a cornerstone to start up growth. Progrowth policies should mutually reinforce --e.g. trade openness will have positive and substantial effects on growth in countries with higher levels of human capital. At the same time, policy complementarities may also impose severe restrictions in the design of optimal growth strategy --especially, among countries with less favorable initial conditions. 1 See webpage: http://www.ustr.gov/tradeagreements/freetradeagreements/caftadrdominicanrepublic centralamericafta 2 To accomplish this task we gather annual information for a sample of 136 countries over the period 19602009 and construct a panel data base of 5year nonoverlapping observations. We run our crosscountry regressions using econometric techniques suitable for dynamic panel data models that account not only for the presence of unobserved components but also for the likely endogeneity or reverse causality of the growth determinants. Our results find that trade has indeed promoted growth, and our result is robust to the specification and technique used. However, the growth benefits of rising trade openness are conditional on the level of progress in structural areas such as education, innovation, infrastructure, institutions, the regulatory framework, financial development and international financial integration. Indeed, we find that the lack of progress in these areas can restrict the potential benefits of trade. We discuss the implications of our regression analysis for CAFTADR nexus, putting emphasis on the impact of trade openness on growth per capita and identifying the structural areas that may represent a constraint to growth. To do so we calculate the impact of trade on growth among CAFTADR countries over the last 15 years and the potential growth gains of raising trade openness to the levels of a benchmark country/region (in our case, the East Asian Tigers, EAP7). In both cases, we find that there is room for trade to stimulate growth but special attention should be placed on reforms in structural areas that are complementary to the trade reform policies implemented by CAFTADR countries, mainly in the areas of education, institutional quality, and infrastructure. This paper is divided in 5 sections. Section 2 presents a brief review of the literature on trade and growth with some emphasis on the channels of transmission, the problems in the empirical literature and the complementarities between trade and other structural factors in driving growth. Section 3 describes the data used in the paper and outlines the econometric methodology to estimate our crosscountry growth regressions. Section 4 presents the empirical evidence on trade and growth and tests whether the impact of the former on the latter is enhanced by advances in structural areas such as education, domestic financial market development, institutional quality, infrastructure, financial integration, innovation and the regulatory framework. We also discuss the economic implications of our statistical analysis on CAFTADR countries. Finally, Section 5 concludes. 2. Literature Review The classical paradigm of international trade argues that trade will promote growth by increasing the relative price of the good that is intensive in the relatively abundant factor (see, e.g. Deardorff, 1973, 1974). It has been found that the standard theory predicts an effect of trade openness on the longrun level rather than on the longrun growth of GDP (Lucas, 1988; Young, 1991). The new trade literature, on the other hand, argues that longterm growth gains from trade can be channeled through more intense research and development activity (see Romer, 1990; Grossman and Helpman, 1991; RiveraBatiz and Romer, 1991). In this context, trade promotes longterm growth by raising the availability of resources for R&D and, thus, increasing the availability of specialized inputs and the size of the market, among other things. 3 More broadly speaking, the theoretical literature is ambiguous about the impact of trade on longrun growth. A strand of the literature suggests that the growth effects are positive when trade specializes in increasingreturnstoscale activities (Young, 1991; Grossman and Helpman, 1991; Eicher, 1999). Others suggest that the effect is either negligible or negative whenever there are market or institutional imperfections (Rodrik and Rodriguez, 2001), underutilization of human or capital resources, focus on extractive activities (Sachs and Warner, 1995, 1999) or there is specialization away from technologicallyintensive, increasing returns to scale sectors (Matsuyama, 1992). It has been argued in the literature that trade may affect economic growth through different channels. First, trade openness may increase a country's market size and, thus, may provide innovators with new business opportunities and allow domestic firms to take advantage of scale economies. Alesina, Spolaore and Wacziarg (2005) find evidence supporting this hypothesis --especially, for smaller countries. Second, trade can enhance technological diffusion and transmit knowhow and managerial practices thanks to stronger interactions with foreign firms and markets (Keller, 2004; Sachs and Warner, 1995). In a seminal paper, Coe and Helpman (1995) find that foreign R&D has a beneficial effect on domestic productivity and that these growth benefits are particularly stronger in countries that are more integrated to international goods' markets. Analogous studies support the hypothesis of productivity gains due to tradefacilitated technology spillovers among developed countries (Xu and Wang, 1999; Keller, 2000; Funk, 2001) as well as among developing countries (Coe et al. 1997).2 Consistent with this evidence, Lewer and van den Berg (2003) find that the strength of trade as an engine of growth depends on the composition of trade. More specifically, they find that countries that import mostly capital goods and export consumer goods tend to grow faster than those that export capital goods. Third, trade may enhance product market competition, thus reducing anticompetitive practices of domestic firms and leading to higher specialization due to exploitation of comparative advantages of domestic firms. Trefler (2004) finds evidence supporting this hypothesis for Canada. In addition, Aghion, Fedderke, Howitt, Kularatne and Viegi (2008) find that trade liberalization stimulated productivity growth in South Africa through product market competition and pricing power of domestic producers. The literature on the consequences of trade liberalization strategies can also be classified in two strands. The first one focuses on the longrun productivity benefits of free trade policies (e.g. Tybout et al., 1991; Levinsohn, 1993; Harrison, 1994; Tybout and Westbrook, 1995; Krishna and Mitra, 1998; Head and Ries, 1999a,b; Pavcnik, 2002) while the second strand examines the impact of freer trade on short run worker displacement and earnings (e.g. Gaston and Trefler, 1994, 1995; Revenga, 1997; Levinsohn, 1999; Beaulieu, 2000; Krishna et al. 2001). The empirical literature on trade and growth typically argued that growth was positively correlated with higher trade volumes, even after accounting for a variety of growth determinants. Edwards (1992), Dollar (1992), BenDavid (1993), Sachs and Warner (1995), Ades and Glaeser (1999), and Alesina, Spolaore and Wacziarg (2000) are examples of this sort. However, Rodriguez and Rodrik 2 Relatedly, several empirical papers suggest that the investment rate is an important channel that links trade and growth (Levine and Renelt, 1992; Baldwin and Seghezza, 1996; Wacziarg, 2001). 4 (2000) argued that most of these findings were less robust than claimed due to: (a) difficulties in measuring openness and especially trade policy, (b) the statistical sensitivity of the specifications and other econometric difficulties --among them, collinearity of protectionist policies with other bad policies, and likely endogeneity of trade openness. These authors argued that the literature focused on the growth effects of trade volumes rather than trade policy, and that the former is plagued by severe endogeneity problems (i.e. a booming country may trade more in international markets). In addition, they suggested that the indicators of trade openness typically used in the empirical literature were as controversial as those proxies for trade barriers. Finally, they stated that the empirical methodologies used to examine the linkages between trade policy and growth were not robust to accounting for endogeneity and controlling for other structural factors --more, specifically, institutions. To address the issue of endogeneity, Frankel and Romer (1999) use a gravity model to instrument for trade openness. According to this model, trade flows between countries would depend on the geographical and cultural characteristics of trading partners --say, distance, remoteness, common border, landlocked and/or island countries, common language, among others-- as well as their size (population and surface area). Using gravitational variables, they attempted to establish a causal link between trade and growth and found that the impact of the former on the latter was positive and statistically significant. The need for a paper that studies contingent relationships between trade policy and growth was addressed by Wacziarg and Welch (2008). More specifically, these authors examined the evolution of growth, investment and openness around episodes of trade liberalization. They found that growth rates in countries that liberalized their trade regimes were 1.5 percentage points higher than before liberalization, and that investment rates rose 1.52.0 percentage points after liberalization. Finally, the trade to GDP ratio rose by 5 percentage points due to the liberalization. In sum, their results suggest that trade and growth have a positive comovement and one of the channels of transmission is likely to be investment. A strand of the empirical literature has suggested that trade openness appears to be beneficial to economic growth on average. However, its effect may vary considerably across countries and may depend upon a variety of conditions associated to structural policies and institutions. Edwards (1993) surveys the conditions needed for successful trade reforms and finds that growth benefits from trade may depend upon a "minimum critical threshold" associated to the level of development (Helleiner, 1986) or the structure of trade (Kohli and Singh, 1989). A recent paper by Chang, Kaltani and Loayza (2009) finds that although trade stimulates growth, this effect can be enhanced by complementary reforms undertaken in the economy. The authors specifically find that interactions among trade and structural factors such as human capital, financial depth, infrastructure and economic regulations are statistically and economically significant, and robust to changes in specification, econometric method, and openness measure.3 Finally, Bolaky and Freund (2004) use crosscountry regressions to find that 3 The empirical growth literature offers some related examples of nonlinear specifications considering interaction effects. Borensztein et al. (1996) and Alfaro et al. (2006) find that growth benefits from FDI are attained when the host country has sufficiently high levels of human capital and financial development, respectively. 5 trade openness is effective in promoting an expansion of income in countries that are not excessively regulated. They argued that resource allocation towards the most productive sectors and companies is more difficult in highlyregulated countries. 3. Data and Methodology 3.1 The Data We have initially collected a panel dataset of 136 countries organized in 5year nonoverlapping observations over the period 19702010, with each country having at most 8 observations. The list of countries in our sample is presented in Table A.1.4 Given that the availability of data is different across variables, we have an effective sample of 99 countries with at least 4 consecutive observations for all variables involved in our analysis. This subsection describes the construction and sources of the data used in our empirical analysis. The focus of this paper is to examine the growth effects of trade openness and the role of complementarities between the latter and other structural factors in promoting growth. Our dependent variable is the average annual growth rate in real GDP per capita within the 5 year period, which is computed as the simple average of log differences in real GDP per capita over the 5year period. Real GDP per capita is expressed in 2005 international dollars (adjusted by PPP) from Heston, Summers and Aten (2009). Our set of control variables includes the (log) level of real GDP per capita at the beginning of the 5year period to test for the existence of transitional convergence. A negative coefficient estimate for this variable would imply that poorer countries may grow faster than richer countries --i.e. consistent with the neoclassical model. The rest of variables that conform our set of longrun growth determinants follows Loayza, Fajnzylber and Calderon (2005): human capital, financial depth, institutional quality, lack of price stability, infrastructure, financial openness and our variable of interest, trade openness. 4 Note that we have collected information for some variables since 1960. However, the information on holdings of foreign assets and liabilities restrict our effective regression sample to start since 1970. 6 Human capital is approximated by the initial gross rate of secondary schooling (in logs) and the data is obtained from Barro and Lee (2001).5 Financial development is measured by the ratio of domestic credit to the private sector to GDP and the data is collected from Beck, DemirgüçKunt and Levine (2000), Beck and DemirgüçKunt (2009), and updated using data from the IMF's International Financial Statistics and the World Bank's WDI. For the sake of robustness, we use other proxies of financial development: domestic credit provided by domestic money banks, and liquid liabilities of the financial sector. Both variables are expressed as a percentage of GDP and in logs. Institutional quality, on the other hand, comprises different dimensions such as absence of corruption, rule of law, enforcement of contracts, quality of the bureaucracy, democratic accountability, among others. We use the ICRG index of political risk as our indicators of institutional quality. The data is published in the International Risk Country Guide (ICRG) by the Political Risk Services (PRS) Group. The lack of price stability is approximated by the average CPI inflation rate. This variable typically reflects the quality of monetary and fiscal policies and is directly related to other indicators of poor macroeconomic management. The data on the inflation rate is gathered from the IMF's International Financial Statistics. Infrastructure is a multidimensional concept; however, most empirical studies have focused on a singlesector approach partly due to: (i) the difficulty of capturing the multiple dimensions of infrastructure in a simple way, and (ii) the high correlation often found among indicators of different types of infrastructure assets (Calderón and Servén, 2004).6 To overcome this problem, while accounting for the multidimensionality of infrastructure, we use principal component analysis to build synthetic indices summarizing information on the quantity of different types of infrastructure assets as well as the quality of services in different infrastructure sectors.7 These synthetic indices combine information on three core infrastructure sectors telecommunications, power, and roads and help address the problem of high colinearity among their individual indicators.8 We denote IK the synthetic quantity 5 This "flow" measure captures more closely current policies on schooling and human capital investment than "stock" measures related with educational attainment of the adult population or life expectancy (Loayza et al. 2005). 6 Calderón and Servén (2004) find that the sample correlation between standard measures of telephone density and power generation capacity (measured respectively by a country's total number of telephone lines, and its total power generation capacity, in both cases relative to the number of workers) exceeds 0.90, which makes it hard to disentangle in a regression framework the separate roles of the two types of assets. 7 Alesina and Perotti (1996) used principal component analysis to create a measure of political instability, while SánchezRobles (1998) employed it to build an aggregate index of infrastructure stocks. 8 We should point out that the sectorspecific indicators of infrastructure quantity and quality employed below, while standard in the literature, are subject to caveats regarding their homogeneity and international comparability. For example, the quality and condition of a `paved road' can vary substantially across countries ­ even within the same country. More homogeneous measures of infrastructure performance would be clearly 7 indices that result from this procedure. The indices can be expressed as linear combinations of the underlying sectorspecific indicators, and hence their use in a regression context is equivalent to imposing linear restrictions on the coefficients of the individual infrastructure indicators. We define the synthetic infrastructure quantity index IK1 as the first principal component of three variables: total telephone lines (fixed and mobile) per 1000 people (Z1/L), electric power installed capacity expressed in MW per 1000 people (Z2/L), and the length of the road network in km. per 1000 people (Z3/L). Each of these variables is expressed in logs and standardized by subtracting its mean and dividing it by its standard deviation. All three infrastructure stocks enter the first principal component with roughly similar weights: Z Z Z IK1 0.603 * ln 1 0.613 * ln 2 0.510 * ln 3 L L A The index accounts for almost 80 percent of the overall variance of the three underlying indicators. As a robustness check, we compute an alternative index of infrastructure quantity, IK2, which uses main telephone lines instead of the combined main lines and mobile phones employed in the first index. 9 Financial openness is approximated by the data on holdings of foreign assets and liabilities from Lane and MilesiFerretti (2001, 2007). Specifically, we use summary measures of financial openness: FAit FLit FLit FOit and FO( L) it GDPit GDPit where FA and FL refer to the stocks of foreign assets and liabilities --expressed as a ratio to GDP. Note that FA and FL include stocks of assets and liabilities in foreign direct investment, portfolio equity, financial derivatives and debt (portfolio debt, bank and traderelated lending).10 On the other hand, given that international trade in debt instruments may be driven by special factors, we also consider the preferable, but unfortunately they do not exist, at least with any significant coverage across countries and time periods. 9 The correlation between the two synthetic quantity indices is over 0.996. This is unsurprising given the similarly high correlation between the two indicators of telephone density underlying the respective synthetic indicators. 10 In this paper we also evaluate the role of the structure of external capital in driving the longterm growth performance of countries. Hence, we will break down our outcome measure of financial openness into equity and loanrelated foreign liabilities. While the former includes the foreign liability position in foreign direct investment and portfolio equity, the latter includes only the debt liability position. The same calculation is performed for the ratio of foreign assets and liabilities to GDP. 8 decomposition of financial openness into equityrelated and debtrelated financial measures (Lane and MilesiFerretti, 2003): FDIAit FDILit PEQAit PEQLit Eq FOit GDPit PDBAit PDBLit OIAit OILit RAit Db FOit GDPit where FDIA and FDIL are stocks of foreign direct investment assets and liabilities, PEQA and PEQL are the stocks of portfolio equity assets and liabilities, PDBA and PDBL are holdings of portfolio debt assets and liabilities, OIA and OIL are stocks of other investment assets and liabilities, and RA represents reserve assets. In short, EqFO and DbFO are indicators of the level of equity and debtrelated cross holdings. Analogously to the definition of overall financial openness, we also define these ratios for only liability holdings, EqFO(L) and DbFO(L), and asset holdings, EqFO(A) and DbFO(A). Our variable of interest, trade openness, affects growth through various channels. It allows production specialization through the exploitation of comparative advantages, enabling technological diffusion and expanding potential markets for the country's goods, among other things. Trade openness is measured as the ratio of real exports and imports to real GDP (all these magnitudes are expressed in local currency at constant prices) and the data is collected from the World Bank's World Development Indicators. We also use an alternative measure of openness that adjusts the volume of trade over GDP for the size (area and population) of the country and for whether the country is landlocked or an oil exporter.11 Loayza, Fajnzylber and Calderon (2005) argue that this structureadjusted volume of trade maybe preferable than the unadjusted ratio given that the econometric analysis is based on cross country comparisons. Unadjusted measures of trade volume may unfairly attribute to trade policy what is merely the result of structural country characteristics --e.g. small countries are more dependent on foreign trade than larger countries, oil exporters may have large trade volumes and also impose high import tariffs, and landlocked countries tend to trade less than other countries due to higher transport and trading costs. Finally, we will describe two sources of data for which we lack extensive time series but we have a good crosscountry coverage: research and development, and economic regulations. We argue that 11 A similar adjustment is presented in Pritchett (1996). 9 positive complementarities between trade and innovation can be exploiting in triggering higher and sustained growth. Our proxies for innovation are R&D spending as percentage of GDP, R&D scientists (per one million people), and R&D technicians (per one million people). We summarize all these three measures in an aggregate R&D index. In addition, we used the share of hightech exports to manufacturing exports as a proxy for innovation. 3.2 Econometric Methodology We have an effective pooled data set of crosscountry and timeseries observations for 99 countries over the period 19702010, and we use an estimation method that is appropriate for dynamic panel data models. The methodology used not only controls for unobserved time and countryspecific effects but also accounts for likely endogeneity or reverse causality among the explanatory variables. In short, we use the generalized method of moments (GMM) for dynamic panel data models developed by Arellano and Bond (1995), Arellano and Bover (1995) and Blundell and Bond (1998). For more details on the econometric methodology, see Appendix I. We regress the growth in real output per capita on a standard set of growth determinants that includes our variable of interest, trade openness. Our basic set of control variables comprises information on the level of human capital, domestic financial depth, institutional quality, lack of price stability, financial openness and infrastructure stocks. In addition to our baseline regression, we explore the role of complementarities between trade and structural factors in driving growth. In short, our dynamic regression equation can be specified as follows: yit yit 1 yit 1 ' K it ' Z it t i it (1) yit 1 ' X it t i it where y denotes the real GDP per worker (in logs), K is a set of standard growth or inequality determinants, and Z is our variable of interest: trade openness. The terms t and i respectively denote an unobserved common factor affecting all countries, and a country effect capturing unobserved country characteristics. The second equality follows from defining Xit = (K'it, Z'it)' and ( ' , ' )' . Our assessment of the effects of trade openness on economic growth in our panel data set poses some econometric challenges: (i) the presence of unobserved effects, and (ii) the potential endogeneity of explanatory variables. We control for unobserved time effects by including period 10 specific dummies in our regressions while unobserved country effects are accounted for by differencing and instrumentation. The problem of joint endogeneity is addressed again by instrumentation in this methodology. More specifically, this econometric technique relaxes the assumption of strong exogeneity of the explanatory variables by allowing them to be correlated with current and previous realizations of the error term, . Since there are no obviously exogenous instruments available, the methodology primarily relies on internal instruments --that is, suitable lags of the explanatory variables (Arellano and Bond, 1991). Additionally, we will use some external instruments to control for the likely endogeneity of our variable of interest, trade openness. We are concerned that moment conditions may not hold with the use of internal instruments and that our results may be driven by invalid instruments. It has been argued that future shocks to growth may promote the expansion of international trade. In this context, we may be required to find instruments that can be considered exogenous and yet be correlated with trade openness. We follow Loayza, Fajnzylber and Calderon (2005) and Chang, Kaltani and Loayza (2009) and consider measures of size and geography as instruments of trade openness --i.e. (actual and lagged values of) population, surface area of the country, and dummies for oil exporting countries and landlocked countries. The consistency of the GMMIV estimator relies on the validity of the moment conditions specified in the Appendix I --equations (I.2) through (I.6). Their validity can be examined through two specification tests (Arellano and Bond, 1991; Arellano and Bover, 1995): First, the Sargan test of over identifying restrictions examines the overall validity of the instruments by evaluating the sample analog of the moment conditions used in the estimation process. If we fail to reject the null hypothesis, we can argue that the moment conditions hold --thus providing statistical support to the model. Second, we conduct higherorder serial correlation tests of the error term it. The system GMMIV estimator, GMM IV(S), tests whether the differenced error term (i.e. the residual of the regression in first differences) shows secondorder serial correlation. We expect that the differenced error term shows firstorder serial correlation even if the error term of the regression in levels is uncorrelated --unless the latter follows a random walk. In this case, the presence of secondorder serial correlation indicates that the original error term is serially correlated and follows a moving average process of at least order one. This 11 would render invalid the proposed instruments.12 Failure to reject the null would tend to support the model. 4. Empirical Assessment 4.1 Basic Correlations and Baseline Regression Panel Correlations Table 1 presents the simple panel correlations between trade openness and growth for our sample of 136 countries with 5year nonoverlapping observations spanning the period 19602010. The panel correlation between these two variables is positive, significant and equal to 0.08. This correlation is significantly higher in countries with high levels of income per capita, human capital, infrastructure and financial openness. Figure 1 depicts the degree of association between trade openness and growth in the 2000s. We note that this correlation is higher than that of the full sample (0.21 vs. 0.08). Note that the second panel of the period identifies the CAFTADR countries in our scatter plot. Most of these countries are close to or below the medians of both trade openness and growth, and they have a flatter relationship than that of the rest of the sample. Figure 2, on the other hand, plots trade visŕvis growth and distinguishes the observations in higher percentiles of the distribution of the control factor (say, human capital, financial development, institutions, financial openness, infrastructure, and regulations) to that of lower percentiles (i.e. those below the 67th percentile of the sample distribution for the period 200110). Our crosssectional figures confirm the results for the pooled panel correlations: the tradegrowth nexus is stronger in countries with more educated people, stronger institutions, an improved infrastructure network and more flexible regulations. Note that these are unconditional correlations and the conditional ones will be conducted below in our regression analysis. 12 If so, we would have to use higherorder lags of the variables as instruments. 12 Baseline Regression Table 2 reports the coefficient estimates for our baseline regressions using different estimation techniques. We should note that the coefficient estimate of our variable of interest, trade openness, is positive and significant (at least at the 10 percent level) regardless of the technique used. In column [1] we run a pooled OLS regression while column [2] controls for time dummies and column [3] controls only for country dummies. We apply the ArellanoBover (1991) GMMIV difference estimator in column [4], thus controlling for unobserved components and endogeneity by differencing and instrumenting the differences of the explanatory variables using their lagged levels. However, the GMMIV difference estimator may face the problem of weak instruments if the explanatory levels are highly persistent. Hence, columns [5] and [6] estimate our baseline regression using the GMMIV system estimator (Arellano and Bover, 1995; Blundell and Bond, 1998). While the estimation in [5] uses internal instruments, the estimation in [6] uses external instruments to account for the likely endogeneity of trade openness. As we pointed out in Section 3, those external instruments are the (actual and lagged levels of) population (in logs), the surface area of the country (in logs), and dummies for landlocked and oil exporting countries. Given that it jointly addresses the issue of likely endogeneity and unobserved factors, our preferred estimation is the one reported in column [6]. We will discuss these results for our baseline regressions. We find a negative and significant coefficient for the initial (log level of) GDP per capita, thus providing evidence of conditional convergence. Growth is enhanced by a faster accumulation of human capital (as proxied by rising gross rates of secondary schooling), deeper domestic financial markets (as measured by higher ratios of domestic credit to the private sector to GDP), and better institutions (as approximated by higher levels of the ICRG political risk index). Lack of price stability, measured by higher rates of consumer price inflation, hinders growth. A faster accumulation of infrastructure stocks (as proxied by deeper telecommunication penetration, larger electricity installed capacity and a longer road network) promotes longterm growth. Financial openness, on the other hand, seems to have an adverse effect on growth rate. Our variable of interest, trade openness, has a positive and significant coefficient. This result implies that longrun growth is enhanced by a more outward orientation in goods' markets. Our coefficient estimates suggests that doubling trade openness would raise the growth rate by 43 basis points per year ­that is, more than 4 percentage points over a decade. Finally, we should also note that 13 the coefficient estimate of trade openness may vary according to the extent of the outward orientation of the country and/or over time. Trade and growth: does the extent of international trade integration matter? Table 3 investigates whether the effect of trade openness on economic growth will depend on the extent of international trade integration. Regression [1] includes the trade openness (TO) and a censored variable that takes the values of our trade openness indicator if it is higher than the 66th percentile of the sample distribution and 0 otherwise. We find that the TO coefficient is negative and statistically not different from zero whereas the TO coefficient for countries with high trade integration is positive and significant. Regressions [2] through [5] differentiate trade openness in our regression analysis for countries with low trade integration (up to the 33rd percentile of the sample distribution), medium trade integration (between 33rd and 67th percentiles) and high trade integration (higher than the 67th percentile). The difference between these regressions lies on the set of instruments used to account for endogeneity. Regression [2] uses the same set of instruments as our baseline regression. Regressions [3] and [4] use an adjusted measure of trade openness in our regression ­thus following Chang, Kaltani and Loayza (2009). Regression [3] uses an adjusted measure calculated as the residual from the regression of trade openness on size and geography measures such as (the log of) population, (log of) surface area, and dummies of landlocked and oil exporting countries. Regression [4] uses the adjusted measure that adds global effects to the previously described regression (i.e. time dummies). Finally, regression [5] uses only internal instruments in the spirit of Arellano and Bond (1991) and Arellano and Bover (1995). Our preferred estimated regression, [2], shows that the coefficient estimate of trade openness is positive regardless of the extent of outward orientation of the country; however, it is statistically significant only for countries with high trade integration. Note that when using only internal instruments (regression [4]), we find that all estimated coefficients are positive and significant, and that the impact of trade openness on growth is higher the deeper is the level of trade integration of the country. Does the impact of trade on growth change over time? Table 4 shows the coefficient of our variable of interest, trade openness, interacted with dummy decades for the 1980s, 1990s and 2000s. Regressions [1] through [3] include only trade openness (TO) and TO interacted with a dummy variable that takes the value of 1 for the period 20009 and 0 otherwise. These regressions differ from the set of instruments used to account for the endogeneity of 14 trade openness (i.e. external instruments as in our baseline, and adjusted measures of trade openness in the spirit of Chang et al, 2009). Regressions [4] through [6] include TO as well as its interaction dummies for the periods 19809, 19909, and 20009. In general, we find that the coefficient estimate for the 1980s is negative and significant in most cases, whereas the TO coefficient for the 2000s is positive and significant. 4.2 Trade and Growth: The Role of Complementarities The evidence presented in Tables 3 and 4 shows that the coefficient estimate of trade openness may not be either constant across countries or over time. In this respect we proceed to estimate the following regression equation yit yit 1 yit 1 ' K it it ' Z it t i it (2) Note that the parameter associated to trade openness (TO), it, is now allowed to vary across countries and time. In this paper, we will model this parameter as follows: it 0 1 Fit where the coefficient of trade openness depends directly on TO as well as its interaction with structural factors. In this paper we consider the complementarities between trade openness and the following factors: human capital, financial development, institutions, infrastructure, financial openness, research and development and certain aspects of the regulatory framework. Complementarities between Trade and Structural Factors Table 5 presents regression estimates that test for the significance of complementarities between trade openness and human capital (regression [2]), trade openness and financial development (regressions [3] to [5]), trade openness and institutional quality (regression [6]). We should note that the impact of trade openness on growth will now depend on the level of the specific structural factor in each country at a determined period of time. Regression [1] of Table 5 includes the interaction between trade openness and the level of income per capita in our baseline regression. While the TO coefficient (0) is negative and significant, its 15 interaction with income per capita is positive and significant (1). This finding suggests that a more outwardoriented strategy in goods' markets would have adverse effects on poorer countries and positive effects on countries with higher income levels. Economically speaking, our regressions suggest that a one standard deviation increase in trade openness --i.e. an increase in the ratio of approximately 75 percent-- would lead to a decline in the growth rate of 30 basis points per year for countries with lower levels of income per capita (approximately US$ 2500 at 2005 PPP prices -- Mongolia in 2005) while it would raise growth of output per capita in countries with higher levels of income per capita (US$ 22000 --Republic of Korea in 2005) by almost 1 percentage point (more precisely, 97 basis points). The first panel of Figure 3 reports the growth effects of rising trade openness for different levels of income per capita --that is, selected percentiles of the distribution (10th, 25th, 33rd, the median, 67th, 75th, and 90th percentiles), regions (CAFTA, LAC excluding CAFTA countries, OECD) and countries (CAFTA DR countries and the United States). Our evidence shows that countries with higher income per capita reap the largest growth benefits from rising trade openness. We also find that all CAFTADR countries (with the exception of Costa Rica) have income per capita levels below sample median for 2005 and, hence, they have a growth effect that is lower than the median response --that is, an increase that is smaller than 40 basis points per year in the growth rate. The growth increase in Costa Rica is approximately 55 basis points per annum (higher than the median) and the smallest response among CAFTADR countries is registered in Nicaragua. Here, the increase in trade openness leads to an annual decline in output per capita growth of 38 basis points. Regression [2] of Table 5, on the other hand, includes the interaction between trade openness and human capital (more specifically, the gross rate of secondary schooling). We find similar results to those found for the income per capita: the coefficient of TO is negative and significant while the interaction with secondary schooling is positive and statistically significant. Again, we could argue that the growth benefits from trade are larger in countries with higher human capital levels. Controlling for the gross rates of secondary enrollment across countries for 2005 in our sample, we find that rising trade openness in countries with low rates of enrollment in secondary schooling (43 percent --e.g. Bangladesh and Ghana located in the 25th percentile of the sample distribution for 2005) would have negligible effects on growth --almost 5 basis points per annum, and this small hike is not statistically significant. On the other hand, a one standard deviation increase in trade openness would raise the growth rate by almost 1.3 percentage points per annum in countries with higher rates of secondary schooling enrollment (96 percent --e.g. Slovak Republic and Slovenia in the 75th percentile). 16 Regressions [3] through [5] report the interaction between trade openness and measures of financial development, that is, domestic credit to the private sector, domestic credit provided by domestic money banks, and liquid liabilities, respectively. All these variables are expressed as a percentage of GDP and in logs, and they are interacted with trade openness. Regardless of the indicators of financial development used in our analysis, we find that the coefficient of TO is negative and not statistically significant in most cases but the interaction with financial development is robustly positive. Again, we find that countries with deeper domestic financial markets may reap the largest growth benefits from trade. Economically, we find that countries with low financial development (say, with domestic credit to the private sector of 20% of GDP --e.g. the average for the 20068 period in Paraguay and Botswana at the 25th percentile of the distribution for that period) would raise their growth per capita by 35 basis points if trade openness were to increase by a one standard deviation. The growth effect of an analogous increase in trade openness more than doubles (a 72 basis point hike in growth per capita) for countries with high financial development (e.g. Greece and Israel with an average domestic credit of 90% of GDP in 20068 --i.e. the 75th percentile of the distribution for that period). Finally, regression [6] interacts trade openness and the level of institutional quality. As we said before, our indicator of institutional quality is the average of the ICRG index of political risk (in logs). Analogous to the other regressions reported in Table 5, we find that the coefficient of TO is negative (and, in this case, statistically significant) whereas the interaction between TO and institutions is positive and significant. This result implies that the growth effects of rising trade openness would be larger in countries with stronger institutions --which is consistent with previous work that show evidence on the role of complementarities between trade and institutional quality; that is, trade reforms may lead to higher growth per capita in countries with stronger institutional quality (Calderon and Fuentes, 2006, 2009; Chang, Kaltani and Loayza, 2009). More specifically, our evidence shows that the rate of growth per capita would increase only by 30 basis points per annum in countries with weak institutions (say, Bolivia and Honduras at the 25th percentile of the sample distribution for the 20069 period), whereas the annual per capita growth benefit for a country with strong institutions (Poland and Slovak Republic at the 75th percentile) is approximately 72 basis points per annum. Complementarities between trade and infrastructure An adequate and efficient supply of infrastructure services has long been perceived as a key ingredient for development (The World Bank, 1994). A strand of the literature shows that infrastructure quantity and quality help promote longterm growth (SanchezRobles, 1998; Calderon and Serven, 2004, 17 2010). Access to infrastructure, on the other hand, plays a significant role in helping reduce income inequality (Estache, Foster and Wodon, 2002; Calderon and Chong, 2004; Calderon and Serven, 2004; Galiani et al. 2005). Recent work has also found that efficient provision of infrastructure is crucial for the success of trade liberalization strategies aimed at optimal resource allocation and growth of exports (Lederman, Maloney and Servén, 2005). Table 6 includes the interactions between trade openness (TO) and a battery of infrastructure indicators (either at the aggregate level or by sector). We have constructed two aggregate indices of infrastructure, IK1 and IK2, that summarize information on telecommunications, electricity and roads. The definition of these indices was provided in Section 3. Regressions [1] and [2] in Table 6 include the interaction between trade openness and the aggregate indices of infrastructure, IK1 and IK2, respectively. In both cases, we find that the coefficient of TO is negative and significant whereas that of the interaction between TO and infrastructure is positive and significant. Thus, our evidence suggests that a better infrastructure network would enhance the impact of trade on growth. Using the estimates of regression [1] we find that a one standard deviation increase in trade openness would lead to an increase in the growth rate per capita of 16 basis points in countries with a poor infrastructure network (i.e. India and Pakistan's average index of infrastructure is at the 25th percentile of the distribution for the 20068 period) while growth per capita would be higher by 1.4 percentage points for countries with better infrastructure networks (Taiwan, Singapore with levels of infrastructure provision in the 75th percentile of the distribution). Regressions [3] through [5] include the index of aggregate infrastructure IK1 and the interaction between trade openness and sectoral measures of infrastructure --say, number of main lines and mobile phones per 1000 people, electricity installed capacity (in MW per 1000 people), and the length of the road network (in km. per 1000 people). On the other hand, regressions [7] to [10] include both the sector indicators of infrastructure and its interaction between trade openness and each of the sectoral indicators. Regression [7] includes the measure of telephone penetration (main lines and mobiles per 1000 people) and its interaction with trade openness, whereas regressions [8] and [9] use electricity installed capacity and road length instead of total phones, respectively. These results imply that an adequate supply of telecommunications, electricity and an improved road network may help raise the growth benefits from trade. 18 Complementarities between trade and financial openness Table 7 explores the complementarities between trade and financial openness in driving growth. It has been argued that vertical FDI may allow MNEs to fragment production optimally and benefit from cost advantages related to targeting laborintensive production states in laborabundant countries (Hanson, Mataloni and Slaughter, 2001). This would lead to rising twoway trade (i.e. higher imports of inputs and subsequent export of upgraded products). Also, trade and financial openness are typically determined by the same set of forcing variables in a gravity model; more specifically, information flows and frictions (Portes and Rey, 2005). In this context, the access to financial flows would lead to more resources that could be devoted to further specialization in traded sectors (KalemliOzcan, Sorensen and Yosha, 2001). Before we assess the interactions between trade and financial openness, we will briefly discuss the results of our baseline regression and those of regression [1] where we decompose the extent of financial openness into equity and debtrelated openness. In the baseline regression we found that the coefficient of financial openness, as proxied by the ratio of foreign assets and liabilities to GDP, is negative and statistically significant. This suggests that financial openness may be harmful for growth and one of the leading explanations behind this result would be its deleterious effects on aggregate volatility (Kose, Prasad, and Terrones, 2003). However, when we decompose the holdings of foreign assets and liabilities of the country into equityrelated and debtrelated, further insights arise. We find that the coefficient of equityrelated financial openness (that is, FDI and portfolio equity assets and liabilities) is positive and significant whereas that of debtrelated financial openness (i.e. portfolio debt and other investment assets and liabilities as well as reserve assets) is negative and significant. This result suggests that growth benefits from financial openness would arise in countries with lower debt to equity ratios. Regression [2] introduces the ratio of foreign assets and liabilities to GDP (FO) and its interaction with trade openness (TO) in our regression equation. On the other hand, regressions [3] and [4] include foreign liabilities as a percentage of GDP (FL) and foreign assets (FA) instead of the sum of foreign assets and liabilities (FO), respectively. Regardless of the measure of financial openness, we find that the coefficient of TO is negative and significant (same as that of financial openness) whereas the interaction between TO and the indicators of financial openness is positive and significant. We will focus our discussion on the estimates in regression [2]. Again, we compare the growth effects of trade openness on growth for countries with low and high financial integration. Countries with low financial integration 19 are those in the 25th percentile of the sample distribution for the 20069 period (Peru and Costa Rica, with foreign assets and liabilities of 115 percent of GDP) while countries with high financial integration are located in the 75th percentile of that distribution (e.g. Taiwan, with foreign assets liabilities of 300 percent of GDP). Regressions [5] to [7] include measures of financial openness decomposed into equity and debtrelated only in its interaction with trade openness whereas regressions [8] to [10] include this decomposition autonomously and also interacted with TO. Focusing on the results presented in the last three columns of Table 7 we find that the coefficient of trade openness is negative and, interestingly, the interaction between TO and equityrelated financial openness is positive and significant whereas that of TO and debtrelated financial openness is negative and significant. This result holds when we use the definitions for financial assets and liabilities (FO) as well as financial liabilities (FL). These results suggest that the structure of external assets and liabilities may have a role in catalyzing the effect of trade on growth. In short, it implies that growth benefits from trade openness may be larger in countries that accumulate more equity rather than debt assets and liabilities. Complementarities between trade openness and research and development (R&D) Table 8 further investigates the complementarities between trade and human capital by examining the interaction between trade openness and innovation. As proxy of innovation we use measures such as R&D spending (as percentage of GDP), R&D scientists (per 1 million people), R&D technicians (per 1 million people), and high technology exports (as percentage of manufacturing exports). We also explore the interaction between trade and an index of R&D that summarizes information on R&D spending, R&D technicians and R&D scientists. Regression [1] of Table 8 includes the interaction between trade openness and our index of innovation. As we stated above, this index is the first principal component of spending and number of scientists and technicians in R&D. Higher values of this index indicates more resources devoted to R&D. In general we find that the coefficient of TO and its interaction are positive and significant, thus implying that trade openness enhances growth and that this effect is higher in countries with higher levels of innovation ­as proxied by more resources devoted to R&D. Regressions [2] through [4] replace the interaction between TO and our index of innovation with the interaction between TO and each of the categories that comprise our index: R&D spending (regression [2]), R&D scientists (regression [3]), and R&D technicians (regression [4]). In all these 20 regressions, the coefficient of TO is positive and significant; however, the interaction between trade openness and R&D is positive and significant only for R&D spending and the number of R&D scientists. What are the economic implications of our estimates? Using the estimates of regression [3] we can assess the growth benefits of higher trade in countries with low R&D spending (like Colombia and Thailand with 1.2% of GDP at the 25th percentile of the sample distribution in 20009), and countries with high R&D spending (such as Ireland and New Zealand with 3.2 percent of GDP at the 75th percentile). A one standard deviation increase in trade openness would lead to higher growth per capita by 90 basis points per annum in countries with low R&D spending whereas it leads to higher growth per capita by 101 basis points per annum in countries with high R&D spending. Finally, regression [6] of Table 8 includes an interaction between TO and the share of hightech exports in manufacturing exports. Although the TO coefficient is positive and significant, we find that the interaction is negative and not different from zero. Complementarities between trade openness and regulations Table 9 presents evidence on the complementarities between trade openness and economic regulations ­i.e. firm entry regulations and labor market regulations. Previous research shows that trade openness is unable to promote growth in heavilyregulated economies. Bolaky and Freund (2004) argue that excessive regulation may prevent the mobilization of resources towards the most productive sectors and to the most efficient firms within each sector. This implies that trade may likely occur in goods where there is no comparative advantage. As we stated in Section 3, we constructed an index of firm entry regulations by compiling information on the number of procedures required to start a business, the number of days that it takes to start that business and its cost. Our index of labor regulations, on the other hand, compiles information on indices of the difficulty of hiring, the difficulty of firing and the rigidity of hours. These indices were constructed either using simple averages or principal components. Finally, we used all six indicators to construct an index of economic regulations ­again, simple averages and principal components were used for aggregation. For the sake of brevity, we will discuss the results using simple averages ­regressions [1] through [3]. However, we should point out that the results are qualitatively similar regardless of the aggregation technique used. In regressions [1] through [3] we find that the coefficient of TO is positive and significantly different from zero whereas that of the interaction between TO and regulations is negative and 21 statistically significant. This confirms existing evidence that more stringent regulations in the economy may hinder economies to reap the growth benefits of rising trade openness. We use the estimates of regression [1] to assess the growth benefits of rising trade openness in countries with more flexible regulations (e.g. Colombia, located in the 25th percentile of the sample distribution) visŕvis countries that are heavily regulated (e.g. France, located in the 75th percentile of the sample distribution). We find that a one standard deviation increase in trade openness would lead to growth per capita that is higher by almost 50 basis points per annum in countries with low regulations, whereas growth per capita increases only by approximately 30 basis points per annum in countries with heavy regulations. 4.3 Economic Implications of Our Estimates: Discussion for CAFTADR This subsection discusses the economic implication of the regression analysis in the subsection 4.2 for the CAFTADR countries. We will conduct this analysis along three dimensions: (a) plot the growth response to a one standard deviation increase in trade openness conditional to the country's level of determined structural factors, (b) show the growth response to an increase in trade openness in 200610 visŕvis 199195, and (c) display the potential growth benefits of trade openness if CAFTADR countries reached the extent of trade openness in a benchmark region (EAP7). Growth implications of rising trade in CAFTADR Figure 3 depicts the growth response to a one standard deviation increase in trade openness (i.e. a sample increase in the trade ratio of approximately 75 percent during the period 20069) conditional on the level of income per capita (Figure 3.1), human capital (Figure 3.2), financial development (Figure 3.3), and institutions (Figure 3.4). We calculate the response for all CAFTADR countries (Costa Rica, Dominican Republic, Guatemala, El Salvador, Honduras, and Nicaragua), selected regions and/or countries (CAFTA, LAC excluding CAFTA, OECD, USA) and selected percentiles of the sample distribution in 20069. The bars represent the growth response (in percentage points) and the dotted lines are the 95% level confidence interval. Growth benefits from trade vary greatly across CAFTADR countries. For instance, the growth benefits of CAFTADR countries conditional on the level of secondary schooling are below the median of our sample distribution (i.e. below 1.1 percentage points per year), with Costa Rica close to the median while Honduras is below the 25th percentile of the distribution and the model predicting a contraction in growth per capita of 19 basis points (see Figure 3.2). On the other hand, we find that the growth benefits of rising trade conditional on the depth of 22 domestic financial markets among CAFTADR countries cannot surpass those of the 67th percentile of the sample distribution (66 basis points per year). Indeed, the growth effects of trade of Costa Rica, Honduras and El Salvador fluctuate between 53 and 57 basis points per year. On the other hand, the lowest benefits from trade are registered by Dominican Republic (42 basis points), which is closer to that of countries in the 33rd percentile of the sample distribution (see Figure 3.3). Finally, growth effects of trade openness conditional on the level of institutional quality are also below that of the 67th percentile of the sample distribution for CAFTADR countries. Figure 4 also assumes an analogous increase in trade openness and plots the growth response conditional to the level of infrastructure. We compute the response conditional on an aggregate index of infrastructure stock IK1 (Figure 4.1), and also conditional on the stocks of telecommunications (Figure 4.2), electricity (Figure 4.3), and roads (Figure 4.4). For the sake of brevity, we will focus our discussion on the growth effects conditional on the aggregate index of infrastructure. We examine the growth response to higher trade conditional on the IK1 levels and find that the growth effects of rising trade are below that of the country with the median level of infrastructure. Costa Rica and the Dominican Republic enjoy the largest benefits from trade (with increases in growth per capita of 95 and 84 basis points) thanks to their relatively better infrastructure network among CAFTADR countries. Nicaragua is the country with the lowest gains from growth (below 50 basis points) among CAFTADR countries. Figure 5 shows the response of growth to a one standard deviation increase in trade conditional on the level of financial openness. We will focus our discussion on Figure 5.1 that uses the holdings of foreign assets and liabilities as percentage of GDP as our proxy of financial openness. Note that our results are qualitatively similar, either using foreign liabilities (Figure 5.2) or equityrelated financial openness measures (see Figure 5.3). In Figure 5.1, Honduras displays the largest growth response (below the median and slightly higher than that of LAC excluding CAFTA). On the other hand, Guatemala is the worst performer (below the 10th percentile of the distribution of financial openness), thus reflecting its low integration to international capital markets. Finally, Figure 6 displays the growth response to trade openness conditional on spending in research and development (Figure 6.1) and regulatory indices for firm entry (Figure 6.2) and labor markets (Figure 6.3). Interestingly, we find that the growth response to rising trade is positive regardless of the values of the structural factors, although not significant in some cases --more specifically for labor market regulations. Given that the intensity of R&D spending is below the median of R&D expenditures in our world sample, the same goes for the growth response to trade openness. However, 23 we should point out that the differences between the growth responses of Costa Rica (close the median in R&D) and El Salvador (just below the 10th percentile in R&D spending) is not significant (91 visŕvis 88 basis points, respectively). Note that the same holds for economic regulations. We find that CAFTADR countries have regulations that are, on average, more restrictive than those of the representative country in our sample. The differences in the gains of growth from trade openness do not seem too large when comparing the countries with the most stringent regulations and that with the most flexible regulations among CAFTADR countries (see Figures 6.2 and 6.3). Growth effects to changes in trade openness over time If we assume that 0 is the coefficient estimate of trade openness TO and 1 is that of the interaction between TO and a determined structural factor SF, then the response of growth to a change in trade openness in period 1 with respect to period 0 is: dg 0 1 SF0 TO1 TO0 where the subscript 1 refers to the averages for these variables in the period 200610 and the subscript 0 refers to the period 199195. Hence we evaluate the growth effects of changes in trade openness over the last 15 years. Note that we evaluate the growth response for the following structural factors: human capital, financial development, institutional quality, the infrastructure stock, and the extent of financial openness. Table 10 reports these results for two scenarios: the first scenario assumes that SF0 is the level of the structural factor in the country or region during the 19915 period. The second scenario assumes that SF0 is the level in 200610 (i.e. period 1). This figure adds to the changes in growth per capita due to trade openness those that were experienced by the country in the structural factor. Note that the 2 bottom rows in Table 10 report the source of the regression coefficients used to perform these calculations. Table 10 measures the contribution of trade to growth over the last 15 years --as measured in basis points per annum. Panel I of Table 10 report those figures conditional on the initial values of the structural factor (say, 19915 in our analysis) and compares them to the model without interactions (i.e. baseline model). Note that if the growth benefits are higher than those reported in the baseline model, then the complementarities at work enhance rather than hinder the impact of trade openness on growth. 24 We find that on average, the baseline model predicts that growth per capita in CAFTA has increased on average by 19 points per annum. However, there is a large degree of variation across countries. On one hand, the model predicts growth benefits of expanding trade of 41 basis points for Nicaragua and a contraction in the growth rate of 3 basis points in Dominican Republic. Note that when trade openness is interacted with human capital, the benefits of trade growth are, on average, reduced in CAFTA due to the lower levels of human capital in this subregion (where our model predicts a decline in growth per capita of 18 basis points). Note that our model with TO interacted with human capital predicts a poorer growth performance in the event of rising trade openness. In general, the growth performance of CAFTA is not better than the one predicted by the baseline, except for when including the interaction between trade openness and financial openness (i.e. growth improves by 23 basis points per year and it is mostly explained by the considerable growth gains in Nicaragua). In most cases, note that the initial level of structural factors represent a severe hindrance for trade to stimulate growth in Honduras and the Dominican Republic, while structural factors facilitate the effects of growth in the rest of the CAFTADR countries. On the other hand, infrastructure seems to play less of a complementary role for all CAFTADR countries with the exception of Costa Rica. The panel II of Table 11, in contrast, calculates the growth effects of changes in trade openness conditional on the level of the structural factors in 200610. Interestingly, we find that the improvement in all structural factors (human capital, financial depth, institutions, infrastructure, and financial openness) lead to higher growth benefits than those in the baseline model. While the baseline regression predicts an annual average increase in growth per capita of 19 basis points for CAFTA, the model predicts increases of 30 and 34 basis points for CAFTA in the models that interact trade openness with human capital and infrastructure, respectively. This result implies that these two sectors may have represented bottlenecks to reap the growth benefits from trade. Potential growth effects to reaching a benchmark in trade openness Following the notation specified above, we calculate the potential growth benefits from raising trade openness if a specific country C were to reach the level of outward orientation of a benchmark region/country B --where these variables are evaluated with the averages of the period 200610: dg 0 1 SF1C TO1B TO1C 25 In our exercise, country C is represented by CAFTADR countries as well as the LAC region while the benchmark B is the average of the East Asian Tigers (EAP7).13 Hence, we calculate the potential growth benefits of CAFTADR countries in reaching the average levels of trade openness among EAP7 countries. Table 11 presents these calculations under two scenarios: the first one uses the value of the structural factor, SF, in country C whereas the second scenario measures the additional growth benefits that country C may have reaped if it had the level of the structural factor in the benchmark country B. Table 11 shows that if the extent of trade openness among CAFTADR countries is raised to that of EAP7 countries (i.e. an increase from an average of 97 to 147 percent of GDP in 20069), our baseline model predicts an average annual increase in the growth per capita of CAFTADR of 26 basis points. Note that when interacted with human capital, research and development, and infrastructure, the growth benefits from trade predicted in our model are even higher (say, 40, 64, and 46 basis points, respectively). In general, the distance of CAFTADR countries to the benchmark would directly determine the extent of the potential growth gains. Panel II of Table 11 present the growth benefits of closing the gap in trade openness of CAFTA DR countries with respect to EAP7 countries but conditional on CAFTADR countries having the level of structural factors of the average EAP7 country. As expected, the growth benefits are higher than those reported in Panel I --which reflects the fact that CAFTADR countries lagged relative to EAP7 countries in terms of human capital, financial depth, institutions, among other structural areas. We find that growth rates would be higher by almost 80 basis points per annum if trade openness were to rise in the model provided that CAFTADR possessed the human capital or the infrastructure network of EAP7 countries. 5. Concluding Remarks The goal of this study is to evaluate the growth effects of trade openness among CAFTADR countries and, more specifically, to examine whether these growth effects are stimulated or hindered by advances in structural policies and institutions. Following recent empirical literature, we evaluated the role of complementarities between trade openness and the following factors: human capital, financial 13 The East Asian tigers (EAP7) are Hong Kong, Indonesia, Republic of Korea, Malaysia, Singapore, Taiwan, and Thailand. 26 development, institutional quality, infrastructure, financial openness, innovation, and economic regulations. Using our effective regression sample of 99 countries with 5year nonoverlapping observations over the period 19602010, we find the following results: First, there is a robust causal link between trade and growth. Regardless of the set of instruments used in our regression analysis, we find that trade openness stimulates growth. In fact, our estimates are not only statistically but also economically significant: a one standard deviation increase in the ratio of trade to GDP (that is, an increase of roughly 75 percent in the ratio) would lead to an increase in the rate of growth per capita of 35 basis points per year (and an accumulated increase of 5.5 percentage points over 15 years). Second, we find strong evidence that the impact of trade openness on growth depends upon countryspecific conditions on structural areas such as education, financial development, institutional quality, infrastructure, financial openness, innovation and regulations. In general, we find that growth benefits from trade openness would be larger in countries that surpass a certain threshold in the structural areas mentioned above. Third, trade stimulates growth in countries with higher levels of human capital, deeper domestic financial markets, stronger institutions, more developed infrastructure networks, highlyintegrated to world financial markets, higher intensity in R&D investment, and less stringent regulations. Fourth, although our baseline model (without interactions) predicts growth benefits from trade for CAFTADR countries, we find that not accounting for complementarities between trade openness and structural factors may overstate these results. In fact, we find that human capital, infrastructure development and institutional quality may play an important role in enhancing the growth benefits from trade (see Table 10). Finally, there is ample room among CAFTADR for reaping the growth benefits from trade. However, a larger role should be played by further reforms in areas such as education, infrastructure, international financial integration, and the development of domestic financial markets (see Table 11). 27 References Ades, A.F., and E.L. Glaeser, 1999. "Evidence on Growth, Increasing Returns and the Extent of the Market." Quarterly Journal of Economics 114(3), 1025­45 Aghion, P., J. Fedderke, P. Howitt, C. Kularatne and N. Viegi, 2008. "Testing Creative Destruction in an Opening Economy: the Case of the South African Manufacturing Industries." Observatoire Francais des Conjonctures Economiques (OFCE), Documents de Travail No. 200823, July Alesina, A., Perotti, R., 1996. "Income Distribution, Political Instability, and Investment." European Economic Review 40, 12021229 Alesina, A., E. Spolaore, and R. Wacziarg, 2000. "Economic Integration and Political Disintegration." American Economic Review 90(5), 1276­96 Alesina, A., E. Spolaore, and R. Wacziarg, 2005. "Trade, Growth and the Size of Countries". In Aghion, P., and S.N. Durlauf, eds., Handbook of Economic Growth. Amsterdam: Elsevier NorthHolland. Alfaro, L., A. Chanda, and S. KalemliOzcan, 2006. "How does foreign direct investment promote economic growth? Exploring the Effects of Financial Markets on Linkages." NBER Working Paper 12522, Cambridge, MA AlonsoBorrego, C., and M. Arellano. 1999. "Symmetrically Normalized InstrumentalVariable Estimation Using Panel Data." Journal of Business and Economic Statistics 17(1), 36­49 Arellano, M., and S. Bond. 1991. "Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations." Review of Economic Studies 58(2), 277­97 Arellano, M., and O. Bover. 1995. "Another Look at the InstrumentalVariable Estimation of Error Components Models." Journal of Econometrics 68(1), 29­52 Baldwin, R.E., and E. Seghezza, 1996. "Testing for TradeInduced, InvestmentLed Growth." NBER Working Paper 5416 Barro, R.J and J.W. Lee, 2001. "International Data on Educational Attainment: Updates and Implications," Oxford Economic Papers, Oxford University Press 53(3): 54163 Beaulieu, E., 2000. "The CanadaU.S. Free Trade Agreement and Labour Market Adjustment in Canada." Canadian Journal of Economics 33(2), 54063 Beck, T., A. DemirgüçKunt, and R. Levine, 2000. "A New Database on Financial Development and Structure." The World Bank Economic Review 14, 597605 Beck, T., and A. DemirgüçKunt, 2009. "Financial Institutions and Markets Across Countries and over Time: Data and Analysis." The World Bank Policy Research Working Paper No. 4943, June BenDavid, D., 1993. "Equalizing Exchange: Trade Liberalization and Income Convergence." Quarterly Journal of Economics 108(3), 653679. Blundell, R., and S. R. Bond. 1998. "Initial Conditions and Moment Restrictions in Dynamic Panel Data Models." Journal of Econometrics 87(1), 115­43 Bolaky, B., and C. Freund, 2004. "Trade, regulations, and growth." The World Bank Policy Research Working Paper 3255, March Borensztein, E., J. De Gregorio, and J.W. Lee, 1995. "How does foreign direct investment affect economic growth?" Journal of International Economics 45 (1), 115­135 Calderón, C. and Chong, A., 2004. "Volume and Quality of Infrastructure and the Distribution of Income: An Empirical Investigation." Review of Income and Wealth 50, 87105 Calderón, C., and R. Fuentes, 2006. "Complementarities between Institutions and Openness in Economic Development: Evidence for a Panel of Countries." Cuadernos de Economía 43, 4980 Calderón, C., and R. Fuentes, 2009. "Removing the constraints from growth: Some guidelines." Pontificia Universidad Catolica de Chile, Instituto de Economia, Working Paper 366, December Calderón, C., N. Loayza, and K. SchmidtHebbel, 2006. "External conditions and growth performance." In: Caballero, R., C. Calderon, and L.F. Cespedes, Eds., External Vulnerability and Preventive 28 Policies. Santiago, Chile: Central Bank of Chile Series on Banking, Analysis and Economic Policies, pp. 4170 Calderón, C., Servén, L., 2004. The Effects of Infrastructure Development on Growth and Income Distribution. The World Bank Policy Research Working Paper 3400, September Chang, R., L. Kaltani, and N. Loayza, 2009. "Openness can be good for growth: The role of policy complementarities." Journal of International Economics 90(1), 3349 Coe, D.T., and E. Helpman, 1995. "International R&D Spillovers." European Economic Review 39(5), 859 887 Coe, D.T., E. Helpman, and A.W. Hoffmaister, 1997. "NorthSouth R&D Spillovers." The Economic Journal 107(440), 134149 Deardorff, A.V., 1973. "The Gains from Trade in and out of SteadyState." Oxford Economic Papers 25(2), 17391 Deardorff, A.V., 1974. "Factor Proportions and Comparative Advantage in the Long Run: Comment." Journal of Political Economy 82(4), 82933 Dollar, D., 1992. "OutwardOriented Developing Economies Really Do Grow More Rapidly: Evidence from 95 LDCs, 197685." Economic Development and Cultural Change, 52344. Dollar, D. and A. Kraay, 2003. "Institutions, Trade, and Growth." Journal of Monetary Economics 50, 133 162. Edwards, S., 1992. "Trade Orientation, Distortions and Growth in Developing Countries." Journal of Development Economics 39(1), 31­57. Edwards, S., 1993. "Openness, Trade Liberalization, and Growth in Developing Countries." Journal of Economic Literature 31, 135893. Eicher, T., 1999. "Trade Development and Converging Growth Rates: Dynamics Gains from Trade Reconsidered." Journal of International Economics 48(1), 179198 Estache, A., Foster, V. and Wodon, Q., 2002. "Accounting for Poverty in Infrastructure Reform: Learning from Latin America's Experience." WBI Development Studies, Washington, DC: The World Bank. Frankel, J.A., and D. Romer, 1999. "Does Trade Cause Growth?" American Economic Review 89(3), 379 399 Funk, M.F., 2001. "Trade and International R&D Spillovers among OECD Countries." Southern Economic Journal 67, 725736. Galiani, S., Gertler, P., Schargrodsky, E., 2005. "Water for Life: The Impact of the Privatization of Water Services on Child Mortality." Journal of Political Economy 113(1), 83120 Gaston, N., and D. Trefler, 1994. "Protection, Trade, and Wages: Evidence from U.S. Manufacturing." Industrial and Labor Relations Review 47(4), 57493. Gaston, N., and D. Trefler, 1995. "Union Wage Sensitivity to Trade and Protection: Theory and Evidence." Journal of International Economics 39(12), 125. Griliches, Z., and J. A. Hausman. 1986. "Errors in Variables in Panel Data." Journal of Econometrics 31(1), 93­118 Grossman, G.M., and E. Helpman, 1991. Innovation and growth in the global economy. Cambridge and London: The MIT Press Hanson, G.H., R.J. Mataloni Jr., and M.J. Slaughter, 2001. "Expansion Strategies of U.S. Multinational Firms." NBER Working Papers 8433 Harrison, A.E., 1994. "Productivity, Imperfect Competition and Trade Reform: Theory and Evidence." Journal of International Economics 36(12), 5373 Head, K., and J. Ries, 1999a. "Can SmallCountry Manufacturing survive Trade Liberalization? Evidence from the CanadaU.S. Free Trade Agreement." Perspectives on North American Free Research Publication No. 1, Industry Canada, April 29 Head, K., and J. Ries, 1999b. "Rationalization Effects of Tariff Reductions." Journal of International Economics 47(2), 295320 Helleiner, G., 1986. "Outward orientation, import instability and African economic growth: an empirical investigation." In: Lall, S. and F. Steward, eds., Theory and Reality in Development: Essays in Honor of Paul Streeten. London: Macmillan, pp. 139­153 Heston, A., R. Summers, and B. Aten, 2009. "Penn World Table Version 6.3." Center for International Comparisons of Production, Income and Prices at the University of Pennsylvania Irwin, D.A. and M. Terviö, 2002. "Does Trade raise Income? Evidence from the Twentieth Century." Journal of International Economics 58, 118. KalemliOzcan, S., B. Sorensen, and O. Yosha, 2001. "Economic Integration, Industrial Specialization, and the Asymmetry of Macroeconomic Fluctuations." Journal of International Economics 55(1), Keller, W., 2000. "Do Trade Patterns and Technology Flows Affect Productivity Growth?" The World Bank Economic Review 14, 1747. Keller, W. 2004. "International Technology Diffusion." Journal of Economic Literature 42(3), 752­82 Kohli, I., and N. Singh, 1989. "Exports and Growth: Critical Minimum Effort and Diminishing Returns." Journal of Development Economics 30(2), 391400 Kose, M.A., E.S. Prasad, and M.E. Terrones, 2003. "Financial Integration and Macroeconomic Volatility." IMF Staff Papers 50(Special Issue), 119142 Krishna, P., and D. Mitra, 1998. "Trade Liberalization, Market Discipline and Productivity Growth: New Evidence from India." Journal of Development Economics 56(2), 44762 Krishna, P., D. Mitra, and S. Chinoy, 2001. "Trade Liberalization and Labor Demand Elasticities: Evidence from Turkey." Journal of International Economics 55(2), 391409 Lane, P.R., and G.M. MilesiFerretti, 2001. "The external wealth of nations: measures of foreign assets and liabilities for industrial and developing countries." Journal of International Economics 55, 263294 Lane, P.R., and G.M. MilesiFerretti, 2003. "International Financial Integration." IMF Staff Papers Special Issue, 50, 82113. Lane, P.R., and G.M. MilesiFerretti, 2007. "The External Wealth of Nations Mark II: Revised and Extended Estimates of Foreign Assets and Liabilities, 19702004." Journal of International Economics 73, 223250 Lederman, D., W.F. Maloney, and L. Servén, 2005. Lessons from NAFTA for Latin America and the Caribbean. Latin American Development Forum Series. Stanford: Stanford University Press, Stanford Economics and Finance; Washington, D.C.: World Bank Levine, R., and D. Renelt, 1992. "A Sensitivity Analysis of CrossCountry Growth Regressions." American Economic Review 84(4), 942­63 Levinsohn, J., 1993. "Testing the ImportsasMarketDiscipline Hypothesis." Journal of International Economics 35(12), 122. Levinsohn, J., 1999. "Employment Responses to International Liberalization in Chile." Journal of International Economics 47(2), pp. 32144. Lewer, J. and H. Van den Berg, 2003. "Does trade composition influence economic growth? Time series evidence for 28 OECD and developing countries." Journal of International Trade & Economic Development 12, 39­96 Loayza, N., P. Fajnzylber, and C. Calderon, 2005. "Economic Growth in Latin America and the Caribbean: Stylized Facts, Explanations and Forecasts." World Bank Latin American and the Caribbean Studies, April 2005 Lucas, R. E., Jr. 1988. "On the Mechanics of Economic Development." Journal of Monetary Economics 22(1), 3­42 30 Matsuyama, K., 1992. "Agricultural Productivity, Comparative Advantage, and Economic Growth." Journal of Economic Theory 58(2), 31734 Pavcnik, N., 2002. "Trade Liberalization, Exit, and Productivity Improvement: Evidence from Chilean Plants." Review of Economic Studies 69(1), 24576 Pritchett, L., 1996. "Measuring Outward Orientation in LDCs: Can It Be Done?" Journal of Development Economics 49(2), 30735 Portes, R., and H. Rey, 2005. "The Determinants of CrossBorder Equity Flows." Journal of International Economics 65(2), 26996 Revenga, A., 1997. "Employment and Wage Effects of Trade Liberalization: The Case of Mexican Manufacturing." Journal of Labor Economics 15(3), Part 2, S20S43. Rodriguez, F., and D. Rodrik, 2000. "Trade Policy and Economic Growth: A Skeptic's Guide to the Cross National Evidence." In Bernanke, B., and K. Rogoff, eds., NBER Macroeconomics Annual 2000. Cambridge, Mass.: The MIT Press, pp. 261335 Rodrik, D., 2005. "Growth strategies." In: Aghion, P., and S. Durlauf, Eds., Handbook of Economic Growth Volume 1. Amsterdam: Elsevier, pp. 9671014 Romer, P.M., 1990. "Endogenous Technological Change." Journal of Political Economy 98(5), Part 2, S71 S102 RiveraBatiz, L.A., and P.M. Romer, 1991. "Economic Integration and Endogenous Growth." The Quarterly Journal of Economics 106(2), 53155 Sachs, J.D., and A.M. Warner. 1995. "Economic Reform and the Process of Global Integration." Brookings Papers on Economic Activity 1, 1­118 Sachs, J.D., and A.M. Warner, 1999. "The Big Push, Natural Resource Booms and Growth." Journal of Development Economics 59(1), 4376 SanchezRobles, B., 1998. "Infrastructure Investment and Growth: Some Empirical Evidence." Contemporary Economic Policy 16, 98108 Stiglitz, J.E., 1970. "Factor Price Equalization in a Dynamic Economy." Journal of Political Economy 78(3), 45688 Trefler, D., 2004. "The Long and Short of the CanadaU.S. Free Trade Agreement." American Economic Review 94(4), 870895 Tybout, J.R., J. de Melo, and V. Corbo, 1991. "The Effects of Trade Reforms on Scale and Technical Efficiency." Journal of International Economics 31(34), 23150 Tybout, J.R., and D.M. Westbrook, 1995. "Trade Liberalization and the Dimensions of Efficiency Change in Mexican Manufacturing Industries." Journal of International Economics 39(12), 5378 Wacziarg, R., 2001. "Measuring the Dynamic Gains from Trade." The World Bank Economic Review 15(3), 393­429 Wacziarg, R., and I. Welch, 2008. "Trade liberalization and growth: New evidence." The World Bank Economic Review 22(2), 187231 World Bank, 1994. World Development Report. Washington, DC: The World Bank Xu, B., and J. Wang, 1999. "Capital Goods Trade and R&D Spillovers in the OECD." Canadian Journal of Economics 32, 12581274. Young, A., 1991. "Learning by Doing and the Dynamic Effects of International Trade." The Quarterly Journal of Economics 106(2), 369405 31 Appendix I Econometric Methodology14 The estimation of equation (1) faces the potential problem of endogenous regressors. This affects, in principal, both the standard determinants of growth (variables in matrix X such as education, financial depth, inflation and so on) and the trade openness variable (as well as its interactions with other structural factors as captured) in (matrix) Z given that we can argue that these variables may be jointly determined. Indeed, the may be subject to reverse causality from growth per capita. 15 Furthermore, the lagged dependent variable yit is also endogenous due to the presence of the country specific effects. We need suitable instruments to deal with endogeneity issues. There are no obviously exogenous variables at hand to construct them and we may rely primarily on internal instruments within the framework described by Arellano and Bond (1991). These instruments are provided by suitable lags of the variables. However, note that the presence of unobserved country characteristics likely means that E[ X is i ] 0 , and hence lagged levels of the regressors are not valid instruments for (1). Therefore, we first eliminate the countryspecific effect by taking firstdifferences of equation (1): y it y it 1 (1 ) y it 1 y it 2 ' X it X it 1 it it 1 (I.1) Assuming that (i) the timevarying disturbance is not serially correlated, and (ii) the explanatory variables X are weakly exogenous (i.e. they are uncorrelated with future realizations of the timevarying error term), lagged values of the endogenous and exogenous variables provide valid instruments.16 In other words, we assume that: E y i ,t s i , t i , t 1 0 for s 2; t 3, ..., T (I.2) E X i ,t s i ,t i ,t 1 0 for s 2; t 3, ..., T (I.3) These conditions define the GMMdifference estimator. In spite of its simplicity, it has some potential shortcomings. When explanatory variables are persistent over time, their lagged levels are 14 The present Appendix draws heavily from Calderón and Servén (2004a). 15 For example, infrastructure accumulation could be driven by productivity growth. 16 Note that this still allows current and future values of the explanatory variables to be affected by the error term. 32 weak instruments for the regression equation in differences (AlonsoBorrego and Arellano, 1996; Blundell and Bond, 1998). This raises the asymptotic variance of the estimator and creates a small sample bias.17 To avoid these problems, below we use a system estimator that combines the regression in differences and in levels (Arellano and Bover 1995, Blundell and Bond 1998). The instruments for the regression in differences are the same as above. The instruments for the regression in levels are the lagged differences of the corresponding variables. These are appropriate instruments under the additional assumption of no correlation between the differences of these variables and the country specific effect. Formally, we assume E[ yi ,t p i ] E[ yi ,t q i ] and (I.4) E[ X i ,t p i ] E[ X i ,t q i ] for all p and q This leads to additional moment conditions for the regression in levels:18 E[y i ,t 1 yi ,t 2 i i ,t ] 0 (I.5) E[ X i ,t 1 X i ,t 2 i i ,t ] 0 (I.6) Using the moment conditions in equations (3), (4), (6), and (7), we employ a Generalized Method of Moments (GMM) procedure to generate consistent estimates of the parameters of interest and their asymptotic variancecovariance (Arellano and Bond, 1991; Arellano and Bover, 1995). These are given by the following formulas: ^ ( X 'W 1W ' X ) 1 X 'W 1W ' y ^ ^ (I.7) AVAR( ) ( X 'W 1W ' X ) 1 ^ ^ (I.8) where is the vector of parameters of interest (, ), y is the dependent variable stacked first in differences and then in levels, X is the explanatoryvariable matrix including the lagged dependent 17 An additional problem with the simple difference estimator relates to measurement error: differencing may exacerbate the bias due to errors in variables by decreasing the signaltonoise ratio (see Griliches and Hausman, 1986). 18 Given that lagged levels are used as instruments in the differences specification, only the most recent difference is used as instrument in the levels specification. Using other lagged differences would result in redundant moment conditions (see Arellano and Bover, 1995). 33 variable (yt1, X) stacked first in differences and then in levels, W is the matrix of instruments derived from the moment conditions, and is a consistent estimate of the variancecovariance matrix of the ^ moment conditions. 19 Consistency of the GMM estimators depends on the validity of the above moment conditions. This can be checked through two specification tests suggested by Arellano and Bond (1991) and Arellano and Bover (1995). The first is a Sargan test of overidentifying restrictions, which tests the overall validity of the instruments by analyzing the sample analog of the moment conditions used in the estimation process. Failure to reject the null hypothesis that the conditions hold gives support to the model. Furthermore, validity of the additional moment conditions required by the system estimator relative to the difference estimator can likewise be verified through difference Sargan tests. The second test examines the null hypothesis that the error term i,t is serially uncorrelated. As with the Sargan test, failure to reject the null lends support to the model. In the system specification we test whether the differenced error term (that is, the residual of the regression in differences) shows secondorder serial correlatation. Firstorder serial correlation of the differenced error term is expected even if the original error term (in levels) is uncorrelated, unless the latter follows a random walk. Secondorder serial correlation of the differenced residual indicates that the original error term is serially correlated and follows a moving average process at least of order one. This would render the proposed instruments invalid (and would call for higherorder lags to be used as instruments). So far we have limited our discussion to internal instruments. But as a double check that our results concerning trade openness are not driven by invalid instruments, we also experiment below with a set of external instruments provided by size and demographic features. This is motivated by the results of Frankel and Romer (1999), Loayza, Fajnzylber and Calderon (2005) and Chang, Kaltani and Loayza (2009), who show that much of the observed variation in infrastructure stocks is explained by population, surface area of the country, and dummies for oil exporting and landlocked countries. Thus, in our core regression analysis, we drop all lags of the trade openness indicator from the set of instruments and replace them with current and lagged values of these variables. 19 In practice, Arellano and Bond (1991) suggest the following twostep procedure to obtain consistent and efficient GMM estimates. First, assume that the residuals, i,t, are independent and homoskedastic both across countries and over time. This assumption corresponds to a specific weighting matrix that is used to produce first step coefficient estimates. Then, construct a consistent estimate of the variancecovariance matrix of the moment conditions with the residuals obtained in the first step, and use this matrix to reestimate the parameters of interest (i.e. secondstep estimates). Asymptotically, the secondstep estimates are superior to the firststep ones in so far as efficiency is concerned. 34 Table 1 Trade and Growth: Correlation Analysis Panel correlation for a sample of 135 countries Sample period: 19602010 (5year nonoverlapping observations) Overall Levels of structural factors: Structural factors Sample Low Middle High Income per capita 0.0792 0.0997 0.0645 0.1315 2.53 1.80 1.21 2.42 Human capital 0.0792 0.0656 0.0053 0.1266 2.53 1.20 0.10 2.38 Financial development 0.0792 0.1049 0.0089 0.0314 2.53 1.85 0.16 0.58 Institutional quality 0.0792 0.0109 0.1233 0.0822 2.53 0.18 2.20 1.51 Infrastructure 0.0792 0.0469 0.0127 0.1356 2.53 0.85 0.23 2.53 Financial openness 0.0792 0.0077 0.245 0.2699 2.53 0.13 4.03 4.21 R&D Spending 0.0792 0.1608 0.2251 0.1098 2.53 2.59 3.56 1.73 Regulation 0.0792 0.0503 0.1337 0.03 2.53 0.88 2.48 0.58 Table 1 reports the correlation between trade openness and growth for the full sample conditional on low, medium, and high levels of the mentioned structural factors. Note that the figures below the correlation coefficient represent tstatistics 35 Table 2 Trade and Growth: Baseline regression under different estimation techniques Dependent Variable: Growth in real GDP per capita (annual average, %) Pooled OLS Within GMMIV GMMIV GMMIV OLS Time dummies Group Difference System System Variables [1] [2] [3] [4] [5] [6] Variable of interest Trade openness 0.5756 * 0.4860 * 1.7448 ** 5.8530 ** 0.3614 ** 0.6245 ** (Exports and imports as % of GDP, log) (0.319) (0.322) (0.757) (1.053) (0.134) (0.143) Control variables Transitional convergence 1.7788 ** 1.9137 ** 6.4594 ** 7.4420 ** 2.1768 ** 2.1263 ** (Initial GDP per capita, log) (0.394) (0.437) (0.852) (0.760) (0.343) (0.218) Human capital 0.7783 ** 0.9918 ** 1.3719 ** 1.0505 * 1.8700 ** 1.5336 ** (Gross secondary enrollment rate, log) (0.350) (0.327) (0.575) (0.597) (0.285) (0.207) Financial depth 0.2492 0.1963 0.4415 0.6054 ** 0.2939 * 0.6229 ** (Credit to private sector, % GDP, log) (0.299) (0.305) (0.378) (0.281) (0.158) (0.148) Institutional quality 0.6914 0.9657 0.4272 0.6872 1.0118 ** 1.5695 ** (ICRG Political risk index, log) (0.725) (0.713) (0.854) (0.732) (0.345) (0.418) Lack of price stability 2.5916 ** 2.4343 ** 2.9280 ** 3.4301 ** 3.6547 ** 3.7073 ** (CPI inflation rate, log) (0.610) (0.642) (0.526) (0.823) (0.134) (0.184) Infrastructure stock 0.6284 ** 0.5882 ** 1.2285 ** 0.2651 0.4335 ** 0.2987 ** (Principal component) 1/ (0.169) (0.187) (0.244) (0.293) (0.139) (0.146) Financial openness 0.6767 ** 0.4241 0.2343 1.2307 ** 0.3706 ** 0.5876 ** (Foreign assets and liabilities, % GDP, log) (0.281) (0.303) (0.279) (0.380) (0.123) (0.129) Time dummies Dummy: 197680 period .. 0.0138 .. .. 0.1739 0.2339 Dummy: 198185 period .. 2.4998 ** .. 1.4681 ** 2.6141 ** 2.5612 ** Dummy: 198600 period .. 1.2370 ** .. 0.7218 ** 1.4532 ** 1.3186 ** Dummy: 199195 period .. 1.6349 ** .. 0.5876 * 1.8917 ** 1.6285 ** Dummy: 199600 period .. 1.7096 ** .. 0.5278 1.9168 ** 1.5804 ** Dummy: 200105 period .. 1.5078 ** .. 0.3271 1.9714 ** 1.5529 ** Dummy: 200609 period .. 0.6260 .. 1.0000 ** 1.0186 ** 0.5994 * Countries / Observations 99 / 646 99 / 646 99 / 547 99 / 547 99 / 646 99 / 646 Country Effects No No Diff Diff Diff Diff Time Effects No Yes No Yes Yes Yes Instruments 2/ No No No Internal Internal External Specification tests (pvalue ) Sargan test (Overidentifying restrictions ) .. .. .. (0.072) (0.310) (0.256) Secondorder serial correlation (0.082) (0.044) (0.273) (0.181) (0.182) (0.211) Numbers in parenthesis are robust standard errors. Regression includes constant. ** (*) indicates that the coefficient estimate is significant at the 5 (10) percent 1/ The aggregate stock of infrastructure is computed as the first principal component of: (a) main telephone lines and mobiles, (b) electric power installed capaci (in MW), and (c) length of the road network (in Km.). All these physical indicators of infrastructure are expressed in their corresponding units per 1000 people. 2/ The set of "internal instruments" correspond to lagged levels and differences of the corresponding explanatory variables in our regression analysis. In contras "external instruments" include variables that instrument for trade openness such as lagged population, surface area of the country, dummy for landlocked coun and oil exporting countries. 36 Table 3 Trade and Growth: Does the extent of trade openness matters? Dependent Variable: Growth in real GDP per capita (annual average, %) Estimation method: GMMIV System Estimator (Arellano and Bover, 1995; Blundell and Bond, 1998) 1/ Baseline Ancillary Regressions Variables Regression [1] [2] [3] [4] [5] Variable of interest Trade openness (TO ) 0.6245 ** 0.0177 .. .. .. .. (Exports and imports as % of GDP, log) (0.143) (0.283) Trade openness LOW .. .. 0.5234 1.0010 ** 1.0332 ** 1.3245 ** (TO lower than 33th %ile, 0 otherwise) (0.385) (0.407) (0.377) (0.647) Trade openness MEDIUM .. .. 0.2588 0.7802 ** 4.4229 ** 2.4355 * (TO if 33rd %ile < TO < 66th %ile) (0.327) (0.330) (1.515) (1.461) Trade openness HIGH .. 0.2115 ** 0.5356 * 0.8788 ** 5.7540 ** 4.1330 ** (TO greater than 66th %ile, 0 otherwise) (0.090) (0.278) (0.293) (0.429) (0.538) Countries / Observations 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 Instruments 2/ External External External External 2 External 3 Internal Specification tests (pvalue ) Sargan test ( Overidentifying restrictions ) (0.256) (0.280) (0.203) (0.165) (0.277) (0.269) Secondorder serial correlation (0.211) (0.251) (0.280) (0.295) (0.342) (0.195) Numbers in parenthesis correspond to robust standard errors. ** (*) indicates that the coefficient estimate is significant at the 5 (10) percent level. 1/ The full regression includes as control variables: the initial GDP per capita (log), gross secondary enrollment rate (log), domestic credit to the private sector as % of GDP (log), ICRG political risk index (log), CPI inflation rate, the aggregate index of infrastructure stock (in logs, see definition in footnote 1 of Table 1), foreign assets and liabilities as % of GDP (log). The regression also includes constant and time (5year period) dummies. 2/ The set of "internal instruments" correspond to lagged levels and differences of the corresponding explanatory variables in our regression analysis. In contrast, "external instruments" include variables that instrument for trade openness such as lagged population, surface area of the country, dummy for landlocked countries, and oil exporting countries. External 2 includes in the regression the measure of trade openness that adjusts for population, area, and geographical measures, and we use lagged levels and differences of this adjusted measure as instruments. Finally, External 3 includes to our adjusted measure the possibility of global shocks by accounting for time effects. 37 Table 4 Trade and Growth: Does the effect of trade changes over time? Dependent Variable: Growth in real GDP per capita (annual average, %) Estimation method: GMMIV System Estimator (Arellano and Bover, 1995; Blundell and Bond, 1998) 1/ Baseline Ancillary Regressions Variables Regression [1] [2] [3] [4] [5] [6] Variable of interest Trade openness (TO ) 0.6245 ** 0.3054 1.2033 ** 0.2938 1.3346 ** 1.4228 ** 0.0055 (Exports and imports as % of GDP, log) (0.143) (0.259) (0.249) (0.202) (0.398) (0.424) (0.369) TO * D(1980s) .. .. .. .. 2.3629 ** 3.5915 ** 1.0916 ** (0.509) (0.559) (0.347) TO * D(1990s) .. .. .. .. 2.3024 ** 1.9858 ** 0.0305 (0.316) (0.547) (0.326) TO * D(2000s) .. 4.4630 ** 6.2430 ** 2.4479 ** 2.3700 ** 1.3297 * 2.0329 ** (0.712) (0.652) (0.318) (0.728) (0.770) (0.501) Countries / Observations 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 Instruments 2/ External External External 2 External 3 External External 2 External 3 Specification tests (pvalue ) Sargan test (Overidentifying restrictions ) (0.256) (0.343) (0.161) (0.174) (0.367) (0.300) (0.248) Secondorder serial correlation (0.211) (0.214) (0.193) (0.151) (0.226) (0.186) (0.181) Numbers in parenthesis correspond to robust standard errors. ** (*) indicates that the coefficient estimate is significant at the 5 (10) percent level. 1/ The full regression includes as control variables: the initial GDP per capita (log), gross secondary enrollment rate (log), domestic credit to the private sector as % of GDP (log), ICRG political risk index (log), CPI inflation rate, the aggregate index of infrastructure stock (in logs, see definition in footnote 1 of Table 1), foreign assets and liabilities as % of GDP (log). The regression also includes constant and time (5year period) dummies. Finally, D(1980s) is a dummy variable that takes the value of 1 for the period 198089 and 0 otherwise, D(1990s) is a dummy variable that takes the value of 1 for the period 199099 and 0 otherwise, and D(2000s) takes the value of 1 for the 20009 period (and 0, otherwise). 2/ The set of "internal instruments" correspond to lagged levels and differences of the corresponding explanatory variables in our regression analysis. In contrast, "external instruments" include variables that instrument for trade openness such as lagged population, surface area of the country, dummy for landlocked countries,and oil exporting countries. We should note that "External 2" includes in the regression the measure of trade openness that adjusts for population, area, and geographical measures, and we use lagged levels and differences of this adjusted measure as instruments. Finally, "External 3" includes to our adjusted measure the possibility of global shocks by accounting for time effects. 38 Table 5 Trade and Growth: Interaction with Structural Factors and Policies Dependent Variable: Growth in real GDP per capita (annual average, %) Estimation method: GMMIV System Estimator (Arellano and Bover, 1995; Blundell and Bond, 1998) 1/ Baseline Ancillary Regressions Variables Regression [1] [2] [3] [4] [5] [6] Variable of interest Trade openness (TO ) 0.6245 ** 8.2487 ** 9.8907 ** 0.6676 1.2141 2.5225 ** 10.0006 ** (Exports and imports as % of GDP, log) (0.143) (1.627) (1.105) (0.474) (0.873) (1.251) (2.054) TO * ypc .. 0.9916 ** .. .. .. .. .. (0.183) TO * human .. .. 2.6520 ** .. .. .. .. (0.282) TO * findev1 .. .. .. 0.4230 ** .. .. .. (0.129) TO * findev2 .. .. .. .. 0.4382 ** .. .. (0.216) TO * findev3 .. .. .. .. .. 0.8046 ** .. (0.312) TO * instq .. .. .. .. .. .. 2.5798 ** (0.492) Control variables Transitional convergence (ypc ) 2.1263 ** 7.7486 ** 3.8154 2.7008 ** 2.2621 ** 1.8939 ** 3.3400 ** (Initial GDP per capita, log) (0.218) (0.864) (0.213) (0.239) (0.261) (0.301) (0.186) Human capital (human ) 1.5336 ** 1.5093 ** 9.0224 1.6259 ** 2.1980 ** 1.9878 ** 1.5617 ** (Gross secondary enrollment rate, log) (0.207) (0.205) (1.116) (0.212) (0.198) (0.232) (0.181) Financial depth (findev1) 0.6229 ** 0.7660 ** 0.7449 1.2707 ** .. .. 0.6589 ** (Domestic credit to private sector, % GDP, log) (0.148) (0.128) (0.120) (0.549) (0.127) Financial depth (findev2) .. .. .. .. 2.0794 ** .. .. (Banking Credit private sector, % GDP, log) (0.823) Financial depth (findev3) .. .. .. .. 3.0230 ** .. (Liquid liabilities M3 % GDP, log) (1.260) Institutional quality (instq ) 1.5695 ** 1.3471 ** 1.4749 0.2013 ** 0.1840 1.8149 8.7044 ** (ICRG Political risk index, log) (0.418) (0.303) (0.274) (0.334) (0.356) (0.424) (1.942) Countries / Observations 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 Specification tests (pvalue ) Sargan test (Overidentifying restrictions ) (0.256) (0.243) (0.196) (0.226) (0.280) (0.299) (0.190) Secondorder serial correlation (0.211) (0.213) (0.181) (0.201) (0.193) (0.261) (0.214) Numbers in parenthesis correspond to robust standard errors. ** (*) indicates that the coefficient estimate is significant at the 5 (10) percent level. 1/ The full regression includes as control variables: the initial GDP per capita (log), gross secondary enrollment rate (log), domestic credit to the private sector as % of GDP (log), ICRG political risk index (log), CPI inflation rate, the aggregate index of infrastructure stock (in logs, see definition in footnote 1 of Table 1), foreign assets and liabilities as % of GDP (log). The regression also includes constant and time (5year period) dummies. We control for endogeneity using lagged levels and differences for all the variables other than trade openness. The latter variable, in turn, is instrumented using lagged population, surface area of the country and dummies for landlocked and oil exporting countries. 39 Table 6 Trade and Growth: The role of physical infrastructure Dependent Variable: Growth in real GDP per capita (annual average, %) Estimation method: GMMIV System Estimator (Arellano and Bover, 1995; Blundell and Bond, 1998) 1/ Baseline Ancillary Regressions Variables Regression [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] Variable of interest Trade openness (TO ) 0.6245 ** 1.7379 ** 1.1129 ** 0.0076 1.2848 ** 0.5840 ** 1.1934 ** 1.6612 ** 1.9481 ** 2.3922 ** 2.4914 ** (Exports and imports as % of GDP, log) (0.143) (0.225) (0.186) (0.236) (0.140) (0.123) (0.267) (0.389) (0.216) (0.271) (0.454) TO * IK1 .. 0.7038 ** .. .. .. .. .. .. .. .. .. (0.066) TO * IK2 .. .. 0.4733 ** .. .. .. .. .. .. .. .. (0.049) TO * TC1 .. .. .. 0.1188 ** .. .. .. 0.5174 ** .. .. .. (0.048) (0.072) TO * EGC .. .. .. .. 0.4285 ** .. .. .. 1.2333 ** .. .. (0.038) (0.132) TO * RD .. .. .. .. .. 0.0246 ** .. .. .. 1.8601 ** .. (0.012) (0.149) TO * TC2 .. .. .. .. .. .. 0.3869 ** .. .. .. 0.7101 ** (0.048) (0.096) Control variables Index of aggregate infrastructure IK1 0.2987 ** 1.6361 ** .. 0.3848 ** 0.0341 1.2645 ** .. .. .. .. .. (First principal component: tc, egc, rd) (0.146) (0.228) (0.158) (0.138) (0.117) Index of aggregate infrastructure IK2 .. .. 0.2592 * .. .. .. 0.3200 * .. .. .. .. (First principal component: tc, egc, rd) (0.166) (0.167) Telecommunications 1 (TC1) .. .. .. .. .. .. .. 1.5394 ** .. .. .. (Main lines and mobiles per 1000 people, log) (0.285) Electric Power (EGC ) .. .. .. .. .. .. .. .. 4.5521 ** .. .. (Installed capacity, in MW per 1000 people, log) (0.508) Roads (RD ) .. .. .. .. .. .. .. .. .. 7.3920 ** .. (Length of total network, in Km per 1000 people, log) (0.608) Telecommunications 2 (TC2) .. .. .. .. .. .. .. .. .. .. 2.0130 ** (Main telephone line per 1000 people, log) (0.374) Countries / Observations 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 Specification tests (pvalue ) Sargan test (Overidentifying restrictions ) (0.256) (0.222) (0.377) (0.221) (0.252) (0.246) (0.186) (0.206) (0.283) (0.214) (0.192) Secondorder serial correlation (0.211) (0.177) (0.162) (0.173) (0.184) (0.180) (0.188) (0.195) (0.163) (0.172) (0.142) Numbers in parenthesis correspond to robust standard errors. ** (*) indicates that the coefficient estimate is significant at the 5 (10) percent level. 1/ The full regression includes as control variables: the initial GDP per capita (log), gross secondary enrollment rate (log), domestic credit to the private sector as % of GDP (log), ICRG political risk index (log), CPI inflation rate, the aggregate index of infrastructure stocks (in logs, see definition in footnote 1 of Table 1), foreign assets and liabilities as % of GDP (log). The regression also includes constant and time (5year period) dummies. We control for endogeneity using lagged levels and differences for all the variables other than trade openness. The latter variable, in turn, is instrumented using lagged population, surface area of the country and dummies for landlocked and oil exporting countries. 40 Table 7 Trade and Growth: Interactions between trade and financial openness Dependent Variable: Growth in real GDP per capita (annual average, %) Estimation method: GMMIV System Estimator (Arellano and Bover, 1995; Blundell and Bond, 1998) 1/ Baseline Ancillary Regressions Variables Regression [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] Variable of interest Trade openness (TO ) 0.6245 ** 0.9605 ** 2.2329 ** 1.2498 ** 0.8781 ** 0.3481 ** 0.0384 0.0490 1.5740 ** 4.3373 ** 2.4291 ** (Exports and imports as % of GDP, log) (0.143) (0.123) (0.374) (0.405) (0.376) (0.128) (0.158) (0.213) (0.388) (0.476) (0.248) TO * FO .. .. 0.6277 ** .. .. .. .. .. .. .. .. (0.086) TO * FL .. .. .. 0.4867 ** .. .. .. .. .. .. .. (0.098) TO * FA .. .. .. .. 0.3230 ** .. .. .. .. .. .. (0.108) TO * FOEq .. .. .. .. .. 0.0703 ** .. .. 0.6811 ** .. .. (0.014) (0.100) TO * FODb .. .. .. .. .. 0.1966 ** .. .. 0.6230 ** .. .. (0.078) (0.160) TO * FLEq .. .. .. .. .. .. 0.0988 ** .. .. 1.2039 ** .. (0.032) (0.112) TO * FLDb .. .. .. .. .. .. 0.2330 ** .. .. 1.7729 ** .. (0.056) (0.206) TO * FAEq .. .. .. .. .. .. .. 0.1857 ** .. .. 0.8552 ** (0.019) (0.055) TO * FADb .. .. .. .. .. .. .. 0.5031 ** .. .. 1.7377 ** (0.080) (0.110) Control variables: Financial Openness Foreign assets and liabilities 0.5876 ** .. 3.3657 ** .. .. 1.7900 ** .. .. .. .. .. (as % GDP, log) [FO] (0.129) (0.423) (0.423) Foreign liabilities .. .. .. 2.8541 ** .. .. 0.0012 .. .. .. .. (as % GDP, log) [FL] (0.458) (0.371) Foreign assets .. .. .. 1.4523 ** .. 1.3299 ** .. .. .. (as % GDP, log) [FA] (0.496) (0.404) Equityrelated foreign assets and .. 0.1521 ** .. .. .. .. .. .. 2.7371 ** .. .. liabilities (as % GDP, log) [FOEq] (0.051) (0.378) Debtrelated foreign assets and .. 1.3168 ** .. .. .. .. .. .. 1.9208 ** .. .. liabilities (as % GDP, log) [FODb] (0.130) (0.649) Equityrelated foreign liabilities .. .. .. .. .. .. .. .. .. 4.6919 ** .. (as % GDP, log) [FLEq] (0.453) Debtrelated foreign liabilities .. .. .. .. .. .. .. .. .. 6.5510 ** .. (as % GDP, log) [FLDb] (0.858) Equityrelated foreign assets .. .. .. .. .. .. .. .. .. 2.7084 ** (as % GDP, log) [FAEq] (0.217) Debtrelated foreign assets .. .. .. .. .. .. .. .. .. 6.3968 ** (as % GDP, log) [FADb] (0.446) Countries 99 / 646 97 / 621 99 / 646 99 / 646 99 / 646 97 / 621 97 / 621 97 / 621 97 / 621 97 / 621 97 / 621 Specification tests (pvalue ) Sargan test ( Overidentifying restrictions ) (0.256) (0.226) (0.276) (0.227) (0.178) (0.246) (0.239) (0.185) (0.326) (0.223) (0.268) Firstorder serial correlation (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Secondorder serial correlation (0.211) (0.202) (0.213) (0.277) (0.170) (0.191) (0.217) (0.260) (0.271) (0.265) (0.276) Numbers in parenthesis correspond to robust standard errors. ** (*) indicates that the coefficient estimate is significant at the 5 (10) percent level. 1/ The full regression includes as control variables: the initial GDP per capita (log), gross secondary enrollment rate (log), domestic credit to the private sector as % of GDP (log), ICRG political risk index (log), CPI inflation rate, the aggregate index of infrastructure stocks (in logs, see definition in footnote 1 of Table 1), foreign assets and liabilities as % of GDP (log). The regression also includes constant and time (5year period) dummies. We control for endogeneity using lagged levels and differences for all the variables other than trade openness. The latter variable, in turn, is instrumented using lagged population, surface area of the country and dummies for landlocked and oil exporting countries. 41 Table 8 Trade and Growth: Complementarities between trade openness and R&D Dependent Variable: Growth in real GDP per capita (annual average, %) Estimation method: GMMIV System Estimator (Arellano and Bover, 1995; Blundell and Bond, 1998) 1/ Baseline Ancillary Regressions Variables Regression [1] [2] [3] [4] [5] Variable of interest Trade openness (TO ) 0.6245 ** 2.4385 ** 1.5130 ** 3.5576 ** 1.5577 ** 0.7394 ** (Exports and imports as % of GDP, log) (0.143) (0.874) (0.216) (0.055) (0.748) (0.161) TO * R&D Index .. 0.0002 * .. .. .. .. (R&D Aggregate Index) 2/ (0.000) TO * R&D Spending .. .. 0.1989 ** .. .. .. (R&D Spending as % of GDP) (0.041) TO * R&D Scientists .. .. .. 0.0004 ** .. .. (Scientists in R&D per 1 million people) (0.000) TO * R&D Technicians .. .. .. .. 0.0001 .. (Technicians in R&D per 1 million people) (0.000) TO * HighTech Exports .. .. .. .. .. 0.0062 (Hightech exports, % manufacturing exports) 0.013 Countries 99 / 646 67 / 446 82 / 545 78 / 519 72 / 472 98 / 641 Specification tests (pvalue ) Sargan test (Overidentifying restrictions ) (0.256) (0.693) (0.311) (0.318) (0.638) (0.164) Secondorder serial correlation (0.211) (0.394) (0.219) (0.263) (0.641) (0.188) Numbers in parenthesis correspond to robust standard errors. ** (*) indicates that the coefficient estimate is significant at the 5 (10) percent level. 1/ The full regression includes as control variables: the initial GDP per capita (log), gross secondary enrollment rate (log), domestic credit to the private sector as % of GDP (log), ICRG political risk index (log), CPI inflation rate, the aggregate index of infrastructure stock (in logs, see definition in footnote 1 of Table 1), foreign assets and liabilities as % of GDP (log). The regression also includes constant and time (5year period) dummies. We control for endogeneity using lagged levels and differences for all the variables other than trade openness. The latter variable, in turn, is instrumented using lagged population, surface area of the country and dummies for landlocked and oil exporting countries. 2/ The aggregate index of R&D is calculated as the first principal component of the following variables: R&D spending as % of GDP, scientists in R&D per 1 million people, and technicians in R&D per 1 million people. 42 Table 9 Trade and Growth: The Role of Regulations Dependent Variable: Growth in real GDP per capita (annual average, %) Estimation method: GMMIV System Estimator (Arellano and Bover, 1995; Blundell and Bond, 1998) 1/ 2/ Ancillary Regressions Baseline Aggregation method: Simple Averages Aggregation method: Principal components Variables Regression [1] [2] [3] [4] [5] [6] Variable of interest Trade openness (TO ) 0.6245 ** 0.7914 ** 1.0219 ** 0.6772 ** 0.6950 ** 0.7087 ** 0.8693 ** (Exports and imports as % of GDP, log) (0.143) (0.144) (0.171) (0.192) (0.135) (0.184) (0.230) TO * Index of regulations .. 0.5878 ** 0.3190 ** (0.224) (0.055) TO * Index of Firm entry regulations 1.6636 ** 0.4504 ** (0.276) (0.178) TO * Index of labor regulations 0.6731 ** 0.3388 ** (0.135) (0.096) Countries 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 99 / 646 Specification tests (pvalue ) Sargan test ( Overidentifying restrictions ) (0.256) (0.250) (0.194) (0.321) (0.201) (0.282) (0.211) Firstorder serial correlation (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Secondorder serial correlation (0.211) (0.181) (0.158) (0.311) (0.194) (0.251) (0.192) Numbers in parenthesis correspond to robust standard errors. ** (*) indicates that the coefficient estimate is significant at the 5 (10) percent level. 1/ The full regression includes as control variables: the initial GDP per capita (log), gross secondary enrollment rate (log), domestic credit to the private sector as % of GDP (log), ICRG political risk index (log), CPI inflation rate, the aggregate index of infrastructure stock (in logs, see definition in footnote 1 of Table 1), foreign assets and liabilities as % of GDP (log). The regression also includes constant and time (5year period) dummies. We control for endogeneity using lagged levels and differences for all the variables other than trade openness. The latter variable, in turn, is instrumented using lagged population, surface area of the country and dummies for landlocked and oil exporting countries. 2/ Our indices of regulations comprise information on the following dimensions: (a) firm entry regulations: number of procedures to start a business, time to start (in days), and its cost (as % of income per capita), and (b) labor market regulations: difficulty of hiring, rigidity of hours and difficulty of firing. All these indices are constructed such that higher values indicate more obstacles to entry and industry and more rigidities in the labor market. Our index of regulations comprises information of all these 6 indicators and it is aggregated either using simple averages or the principal components analysis (i.e. we take the first principal components). Analogously, we compute the aggregate index of regulation for firm entry regulations and labor market regulations by either taking simple averages or the first principal component of the 3 indicators in each category. 43 Table 10 Growth effects due to changes in trade openness, 200610 vs. 199195 (In basis points per annum) Trade openness interacted with: Baseline Human Financial Institutional Infrastructure Financial Model Capital Development Quality Stock Openness I. Conditional on the structural factors of the country in 199195 Costa Rica 22 3 14 37 21 20 Dominican Republic 3 0 3 2 1 2 Guatemala 12 31 9 2 22 8 Honduras 4 4 5 1 6 6 Nicaragua 41 5 49 15 75 113 El Salvador 39 76 43 14 41 23 Latin America (LAC) 17 8 19 15 1 19 CAFTA 19 18 19 12 17 23 LAC (excl. CAFTA) 14 14 17 13 6 15 II. Conditional on the structural factors of the country in 200610 Costa Rica 22 59 32 36 57 26 Dominican Republic 3 7 3 4 7 3 Guatemala 12 11 15 13 17 7 Honduras 4 2 6 3 6 6 Nicaragua 41 82 56 51 46 55 El Salvador 39 74 58 53 72 52 Latin America (LAC) 17 39 21 20 38 22 CAFTA 19 30 27 24 34 24 LAC (excl. CAFTA) 14 36 17 16 34 18 Table Table 1 Table 4 Table 4 Table 4 Table 5 Table 6 Regression [6] [2] [3] [6] [1] [2] 44 Table 11 Potential growth effects of attaining the level of trade integration of the East Asian Tigers (In basis points per annum) Trade openness interacted with: Baseline Human Financial Institutional Infrastructure Financial Research & Economic Model Capital Development Quality Stock Openness Development Regulations I. Conditional on the structural factors of the country in 200610 Costa Rica 17 48 26 29 46 21 44 16 Dominican Republic 34 78 39 45 80 36 .. 35 Guatemala 48 42 60 52 66 26 117 47 Honduras 7 3 10 5 9 9 16 6 Nicaragua 32 64 44 40 36 43 79 32 El Salvador 17 33 25 23 31 23 42 17 Latin America (LAC) 45 105 58 54 104 60 113 43 CAFTA 26 40 36 32 46 33 64 25 LAC (excl. CAFTA) 55 143 68 65 136 74 138 52 II. Conditional on the structural factors of the benchmark in 200610 Costa Rica 17 53 34 30 53 39 48 19 Dominican Republic 34 104 67 58 104 76 94 37 Guatemala 48 146 94 81 146 106 131 52 Honduras 7 20 13 11 20 14 18 7 Nicaragua 32 98 63 55 98 71 88 35 El Salvador 17 53 34 29 53 38 47 19 Latin America (LAC) 45 137 88 76 137 100 123 49 CAFTA 26 79 51 44 79 58 71 28 LAC (excl. CAFTA) 55 166 107 93 166 121 150 59 Table Table 1 Table 4 Table 4 Table 4 Table 5 Table 6 Table 7 Table 8 Regression [6] [2] [3] [6] [1] [2] [2] [1] 45 Figure 1 Trade and Growth 1.1 Trade openness vs. Growth: Scatterplot Correlation between Growth and Trade Openness 15 10 GDP Growth 5 0 -5 3 4 5 6 Trade Openness 1.2 Trade openness vs. Growth: Where are the CAFTADR countries? Correlation between Growth and Trade Openness CAFTA-DR countries detail 15 10 GDP Growth 5 DOM CRI HND GTM SLV NIC 0 -5 3 4 5 6 Trade Openness 46 Figure 2 Trade openness vs. Growth: Do reformers exploit a higher correlation? 2.1 Correlation according to levels of human capital 2.2 Correlation according to levels of financial development Correlation between Growth and Trade Openness Correlation between Growth and Trade Openness Education categories detail Credit categories detail 15 15 10 10 GDP Growth GDP Growth 5 5 0 0 -5 -5 3 4 5 6 3 4 5 6 Trade Openness Trade Openness 2.3 Correlation according to levels of institutions 2.4 Correlation according to levels of financial openness Correlation between Growth and Trade Openness Correlation between Growth and Trade Openness Institutional categories detail Financial Openness categories detail 15 15 10 10 GDP Growth GDP Growth 5 5 0 0 -5 -5 3 4 5 6 3 4 5 6 Trade Openness Trade Openness 2.5 Correlation according to levels of infrastructure 2.6 Correlation according to levels of economic regulation Correlation between Growth and Trade Openness Correlation between Growth and Trade Openness Infrastructure Development categories detail Regulations categories detail 15 15 10 10 GDP Growth GDP Growth 5 5 0 0 -5 -5 3 4 5 6 3 4 5 6 Trade Openness Trade Openness 47 Figure 3 Growth response to a one standard deviation increase in trade openness 3.1 Conditional on the level of income per capita 3.2 Conditional on the level of secondary schooling 2.0 2.0 1.5 1.5 1.0 1.0 0.5 0.5 0.0 0.0 0.5 0.5 1.0 1.0 SLV CRI Median GTM CAFTA DOM OECD HND NIC USA 10th %ile 25th %ile 33th %ile LAC (ex. CAFTA) 67th %ile 75th %ile 90th %ile DOM SLV OECD HND CRI NIC USA LAC (ex. CAFTA) Median CAFTA GTM 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile Note: The computed responses were obtained using the estimated coefficients from column [1] of Table 4. Higher percentiles imply Note: The computed responses were obtained using the estimated coefficients from column [2] of Table 4. Higher percentiles imply higher levels of income per capita. higher (gross) enrollment rates for secondary schooling. 3.3 Conditional on the level of domestic financial development 3.4 Conditional on the level of institutional quality 1.2 1.2 1.0 1.0 0.8 0.8 0.6 0.6 0.4 0.4 0.2 0.2 0.0 0.0 0.2 0.2 SLV CRI Median GTM CAFTA SLV DOM OECD CRI NIC HND USA 10th %ile 25th %ile 33th %ile LAC (ex. CAFTA) 67th %ile 75th %ile 90th %ile Median GTM CAFTA DOM OECD NIC HND USA 10th %ile 25th %ile 33th %ile LAC (ex. CAFTA) 67th %ile 75th %ile 90th %ile Note: The computed responses were obtained using the estimated coefficients from column [3] of Table 4. Higher percentiles imply Note: The computed responses were obtained using the estimated coefficients from column [6] of Table 4. Higher percentiles imply higher ratios of domestic credit to the private sector relative to GDP. higher values of the ICRG index of political risk (as reported by the PRS Group). 48 Figure 4 Growth response to a one standard deviation increase in trade openness 4.1 Conditional on the level of aggregate infrastructure stock (IK1) 4.2 Conditional on the level of aggregate infrastructure stock: Telecommunications 2.5 1.4 2.0 1.2 1.5 1.0 1.0 0.8 0.5 0.6 0.0 0.4 0.5 0.2 1.0 0.0 SLV CRI Median GTM CAFTA DOM OECD HND NIC USA 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile LAC (ex. CAFTA) LAC (ex. SLV CRI Median DOM GTM CAFTA OECD HND NIC USA 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile CAFTA) Note: The computed responses were obtained using the estimated coefficients from column [2] of Table 5. The IK1 index is the first Note: The computed responses were obtained using the estimated coefficients from column [7] of Table 5. Our indicator of tele principal component of main lines and mobiles, electricity installed capacity (MW) and road length (Km). Higher percentiles imply communications is the number of main lines and mobile phones per 1 million people (in logs). Higher percentiles imply a higher higher values of the synthetic infrastructure index IK1 (i.e. more provision of infrastructure). penetration of main lines and mobile phones among the population of the country. 4.3 Conditional on the level of aggregate infrastructure stock: Electric Power 4.4 Conditional on the level of aggregate infrastructure stock: Roads 2.5 2.5 2.0 2.0 1.5 1.5 1.0 1.0 0.5 0.5 0.0 0.0 0.5 1.0 0.5 1.5 1.0 2.0 1.5 SLV CRI Median GTM CAFTA DOM OECD HND DOM NIC 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile USA LAC (ex. CAFTA) SLV OECD NIC HND CRI USA LAC (ex. CAFTA) Median GTM CAFTA 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile Note: The computed responses were obtained using the estimated coefficients from column [8] of Table 5. Our indicator of electric Note: The computed responses were obtained using the estimated coefficients from column [9] of Table 5. Our indicator of roads power is the electricity installed capacity (in MW) per 1 million people. Higher percentiles imply higher electricity installed capacity is the length of the total road network (in km.) per 1000 people. Higher percentiles imply a larger road network per person. per person. 49 Figure 5 Growth response to a one standard deviation increase in trade openness 5.1 Conditional on the level of financial openness Foreign assets and liabilities 5.2 Conditional on the level of financial openness Foreign liabilities 1.5 1.2 1.0 1.0 0.8 0.6 0.5 0.4 0.2 0.0 0.0 SLV CRI GTM CAFTA Median DOM OECD HND NIC USA 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile LAC (ex. CAFTA) DOM SLV OECD CRI USA HND NIC LAC (ex. CAFTA) Median GTM CAFTA 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile Note: The computed responses were obtained using the estimated coefficients from column [2] of Table 6. Financial openness is Note: The computed responses were obtained using the estimated coefficients from column [3] of Table 6. Financial openness is calculated by the ratio of foreign asset and liability holdings as % of GDP (in logs). Higher percentiles imply higher ratios of foreign calculated by the ratio of foreign asset and liability holdings as % of GDP (in logs). Higher percentiles imply higher ratios of foreign assets and liabilities to GDP (i.e. deeper international financial integration). liabilities to GDP. 5.3 Conditional on the level of equityrelated financial openness Foreign assets and liabilities 5.4 Conditional on the level of debtrelated financial openness Foreign assets and liabilities 1.5 1.4 1.2 1.0 1.0 0.8 0.6 0.4 0.5 0.2 0.0 0.2 0.0 0.4 SLV CRI GTM Median CAFTA DOM OECD HND NIC USA 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile LAC (ex. CAFTA) DOM SLV OECD CRI HND NIC LAC (ex. CAFTA) USA Median GTM CAFTA 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile Note: The computed responses were obtained using the estimated coefficients from column [8] of Table 6. Equityrelated financial Note: The computed responses were obtained using the estimated coefficients from column [8] of Table 6. Debtrelated financial openness is calculated by the holdings of FDI and Portfolio Equity assets and liabilities as % of GDP (in logs). Higher percentiles openness is calculated by the holdings of Portfolio Debt and Other Investment assets and liabilities as % of GDP (in logs). Higher indicate higher equityrelated financial openness ratios. percentiles indicate higher debtrelated financial openness ratios. 50 Figure 6 Growth response to a one standard deviation increase in trade openness 6.1 Conditional on the level of R&D Spending 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 SLV CRI GTM Median CAFTA OECD HND NIC USA 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile LAC (ex. CAFTA) Note: The computed responses were obtained using the estimated coefficients from column [2] of Table 7. R&D spending is the average ratio of R&D expenditure as % of GDP for the 20009 period. Higher percentiles imply higher R&D spending to GDP ratios. 6.2 Conditional on the level of firm entry regulations 0.7 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0.1 DOM OECD SLV HND USA NIC CRI LAC (ex. CAFTA) Median CAFTA GTM 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile Note: The computed responses were obtained using the estimated coefficients from column [2] of Table 8. Firm entry regulations are calculated as the simple average of the following measures: number of procedures, time and cost. Higher percentiles imply stricted regulations on firm entry (barriers to entry). 6.3 Conditional on the level of labor market regulations 0.6 0.5 0.4 0.3 0.2 0.1 0.0 0.1 DOM SLV OECD CRI HND USA NIC LAC (ex. CAFTA) Median GTM CAFTA 10th %ile 25th %ile 33th %ile 67th %ile 75th %ile 90th %ile Note: The computed responses were obtained using the estimated coefficients from column [3] of Table 8. Labor market regulations are calculated as the simple average of the following measures: difficulty of hiring, rigidity of hours and difficulty of firing. Higher percentiles indicate more strict regulations on labor markets. 51