Policy Research Working Paper 10645 The Trade-Growth Nexus Evidence of Causality from Innovative Instruments for Trade Ibrahim Nana Sephooko Ignatius Motelle Susan K. Starnes International Finance Corporation December 2023 Policy Research Working Paper 10645 Abstract During the past decades, extensive literature has empha- over 1970–2020. The findings suggest that international sized the role of both international trade and openness in trade has a positive and significant effect on gross domestic fostering economic growth. Endogeneity bias is a nagging product per capita, which tends to be higher for emerging challenge for any empirical attempt to study the causal rela- markets and development economies. Thus, the study pro- tionship between trade and economic growth. This study vides an enhanced empirical foundation for the expectation contributes to the existing stock of knowledge and helps that investments made to support trade are also good for to address these challenges by introducing new instrumen- economic growth, especially in emerging markets. tal variables for trade. The study samples 197 countries This paper is a product of the International Finance Corporation. It is part of a larger effort by the World Bank Group to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at at inana@ifc.org, smotelle@ifc.org, and sstarnes@ifc.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 The Trade-Growth Nexus: Evidence of Causality from Innovative Instruments for Trade Ibrahim Nana1; Sephooko Ignatius Motelle2; Susan K. Starnes3 JEL Classification Codes : F15, F43, O47 Keywords: International trade, GDP, growth, instrumental variable Acknowledgments: We would like to thank Daniel Lederman, Deputy World Bank Chief Economist for the Middle East and North Africa Region, Michael Jansson Professor of Econometrics at University of California, Berkeley and Filipe Lage de Sousa Program Coordinator and Senior Lecturer - M.S. in Applied Economics Program at The Johns Hopkins University for their comments and guidance. The paper has also benefitted from thoughtful guidance and encouragement from Pablo Fajnzylber, Director of the Development Impact Measurement department (IFC/CDI), Issa Faye Senior Economic Adviser at the CGIEU, Dan Goldblum Manager of the CDIFI Unit and Arun Prakash Strategy Officer at the CDIFI Unit. 1 ET Consultant, Sector Economics and Development Impact Department: inana@ifc.org 2 Senior Economist, Sector Economics and Development Impact Department: smotelle@ifc.org 3 Lead Global Trade and Commodity Finance Strategist, Sector Economics and Development: sstarnes@ifc.org 1. INTRODUCTION Global trade has experienced significant expansion buoyed by technological progress and trade-supporting policy shifts. IMF data highlight that, by the year 2000, trade value had increased 60 times since 1960, reaching US$13 trillion. During this period, exports increased 61 times while imports rose 58 times. 4 Since then, global trade has seen curve-shifting expansion. There is an apparent correlation between the accelerated growth in trade and economic growth. According to the World Bank World Development Indicator data (WDI), world exports of goods and services increased from US$8 trillion in 2002 to over US$31 trillion in 2022 and as shown in Figure 1 below, the growth in exports was accompanied by global gross domestic product (GDP) growth. The patterns of trends in global trade tend to be similar for both advanced economies (AEs) and Emerging Markets and Developing Economies (EMDEs), albeit higher for the former. International trade has enabled EMDEs to integrate with the global economy, which has mirrored their growth. Figure 1 shows that EMDEs have increased their participation in global trade significantly since the 1960s as well as their share of global trade (from 19 percent in 1990 to 41 percent in 2021). Between the late 1990s and 2022, manufacturing exports from EMDEs increased eight-fold — from approximately US$ 629 billion in 1995 to US$ 5.5 trillion in 2022. Countries that trade more tend to achieve higher income, as illustrated in Figures A2, A3, and A4 in the Appendix. Figure 1. The relationship between trade and growth and the share of EMDEs trade a) Trade vs GDP (1970-2021) b) Share of EMDEs trade (1970-2021) 120 100% Trillions 100 80% 80 60% 60 40% 40 20 20% 0 0% 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 2018 2021 1970 1973 1976 1979 1982 1985 1988 1991 1994 1997 2000 2003 2006 2009 2012 2015 2018 2021 Exports GS (current US$) Imports GS (current US$) Emerging and Developing Economies Trade GS (current US$) GDP (current US$) Advanced Economies Source: Authors’ calculation based on data from the World Bank Group and the IMF. Note: GS in Panel (a) refers to “Good and Services”. There are specific examples of developing countries that have benefited from trade-supported industrialization. For example, the East Asian Tigers realized significant export-led growth during the period 1960s-1990s. China has been leading EMDE growth in the post-1990s and India is increasingly playing an important role recently. Several studies discover a pattern of faster growth for countries that have embraced the trade liberalization reform (Falvey, Foster, and Greenaway 2012; Frankel and Romer 1999; Salinas and Aksoy 2006; Thirlwall 2000; Wacziarg and Welch 2008). EMDEs that are actively 4 According to the World Bank World Development Indicators, trade (% GDP) has doubled between 1970 and 2000, reaching 57% in 2021. 2 engaged in international trade have also made the most progress in alleviating poverty and raising living standards (WTO 2017). The Global Financial Crisis of 2007-2008 stifled global value chains and led to a reduction of world trade by 10 percent. More recently, the COVID-19 pandemic disrupted value chains and port and transit capacity due to lockdowns. Consequently, Global Trade dropped by 6 percent year on year in the first quarter of 2020 and fell by 21 percent during the second quarter. The recovery of trade from the Global Financial Crisis by 30 percent in 2010 and from the COVID-19 pandemic by 11 percent in 2021 appeared to accompany economic recovery, although the recovery speeds varied across countries. This may be pure correlation instead of a causal relationship between trade and growth. Early empirical evidence on the positive relationship between trade and growth can be traced back to the literature in the late 1960s (Emery 1967; Maisel 1968). Building on this foundation, additional studies in the 1970s showed that, in the long run, outward-oriented development strategies were more conducive to higher growth than import substitution (Bhagwati 1978; Krueger 1978; Michaely 1977). Empirical studies in the 1990s and early millennium buttressed the finding that trade is good for growth based on a boom of innovation in econometric approaches (Dollar and Kraay 2004; Feyrer 2009; Frankel and Romer 1999; Greenaway, Morgan, and Wright 1998; Grossman and Helpmann 2015; Hausmann, Pritchett, and Rodrik 2005; Sachs and Warner 1995). However, some dissenting voices leveled fundamental criticisms against the claim that trade is a causal determinant of economic growth (Rodriguez and Rodrik 2000). 5 The problem of endogeneity undermines empirical findings based on the standard ordinary least squares (OLS) approach. While the OLS approach reliably captures the correlation between trade and growth, it is unable to provide useful insight on the question of causality due to problem of endogeneity (which can be caused by omitted variable bias, measurement error, selection bias and/or simultaneity bias), rendering any statistical inference on the causal relationship between trade and growth invalid (Frankel and Romer 1999; Rodriguez and Rodrik 2000; Winters 2004). For example, Rodriguez and Rodrik (2000) argue that OLS erroneously attributes to trade most of the explanatory power that is otherwise associated with other determinants of growth such as institutions and geography. 6 Frankel and Romer (1999) attempted to address this challenge by turning to an instrumental variable (IV) approach, estimating a gravity equation with only geography-related explanatory variables to predict exogenous trade used as an instrument. They utilized the trade intensity measure (the ratio of total trade to GDP) and found that greater trade intensity is responsible for greater growth. 7 However, their chosen instrument violated the exclusion condition because it did not necessarily affect GDP exclusively through the channel of trade. Brueckner and Lederman (2015) addressed the endogeneity challenge in the context of Sub-Saharan Africa (SSA) using residuals from the growth impact on trade as an instrument to solve reverse causality. More recently, Feyrer (2018) built on the work of Frankel and Romer (1999) by developing an instrument based on time varying geography leveraging developments in transportation technology. Feyrer (2018) used the gravity regressions to obtain time series predictions for bilateral trade based on exogenous geography to generate a panel of predictions for overall trade for each country in the sample over time. 5 There is also some literature that utilizes complex general equilibrium (GE) models (Aguiar et al. 2019). The WTO Global Trade Model, for example, is a recursive dynamic extension of the static GTAP model (Corong et al. 2017). It implements a parsimonious approach to incorporate monopolistic competition in the standard GTAP model following Bekkers and Francois (2018). 6 Geography is not necessarily fully exogenous to income because a country’s resource endowments or quality of institutions can be affected by its location (Brock and Durlauf 2001; Winters 2004). 7 For further exposition on the theoretical foundations of the gravity equation see Anderson (1979) and Evenett and Keller (2002), as well as Baldwin and Taglioni (2006) on matters of estimation of the gravity equation. 3 Our study contributes to the stock of knowledge in three ways. First, it contributes to past efforts to address endogeneity caused by reverse causality and omitted variable bias. Feyrer (2018) noted that advances in air transport technology did not exclusively benefit trade, but also non-trade economic activities such as migration through passenger air-travel. Thus, his “flight distance” instrument does not fully meet the exclusion restriction. Our study refines this instrument further by focusing purely on air-freight capacity which measures air transport freight capacity dedicated only to international cargo shipments. Second, as recommended by Tang (2011), our study differs from Feyrer (2018) by measuring trade using the trade-to- GDP ratio adjusted for each country’s share in global trade to remove the bias caused by the size of the domestic economy. Third, in addition to “air transport freight capacity,” this study introduces a new pair of instrumental variables — “the mean GDP of top five trade partners”, and “the mean distance to the top five largest traders in the world.” These help to further address potential endogeneity by both eliminating domestic effects on GDP growth and seeking to minimize potential effects of cross border investment (versus cross border trade). Our findings confirm that trade intensity (as a percentage of GDP) has a statistically significant positive causal impact on GDP per capita. Furthermore, the positive impact of trade tends to be higher for emerging markets and developing economies (EMDEs). The rest of the paper is organized as follows: Section 2 presents a summary of relevant literature on the trade-growth nexus. Section 3 presents the methodology followed by the study, highlighting how it builds on past success and takes advantage of advances in econometric modeling. Section 4 discusses the results of the study and Section 5 concludes. 2. SUMMARY OF RELEVANT LITERATURE 2.1. The basic theoretical growth model The modeling process followed by Mankiw, Romer, and Weil (1992) has been a highly influential contribution to the empirical literature that investigates the relationship between trade and economic growth. It is based on the growth theory inspired by the Solow (1956) growth model. It relies on a standard neoclassical production function with decreasing returns on "capital" to study the determinants of economic growth. Several studies based their empirical models on a standard Solow model, an augmented Solow model, or a linearized neoclassical production function and adding additional control variables (Keho 2017; Mankiw, Romer, and Weil 1992; Nsiah and Fayissa 2018). For example, Mankiw, Romer, and Weil (1992) examine whether the Solow growth model is consistent with the international variation in the standard of living. They found that augmenting the Solow model by including the accumulation of human and physical capital provides an excellent theoretical basis for explaining economic growth. 2.2. Trade and other fundamental growth determinants 2.2.1 Independent variable — Trade Trade is the independent variable of interest for our study, and it can be conceptually understood as the degree to which an economy maintains its outward orientation by exchanging goods and services with the rest of the world (Fujii 2019). In the literature, trade is often measured using the trade-to-GDP ratio defined as the ratio of exports plus imports to GDP (Tang 2011) — sometimes referred to as the trade intensity ratio 4 (Leamer 1988) or the trade share (Frankel and Romer 1999). 8 The ratio is an elegant measure of trade for three reasons. According to Tang (2011) normalizing trade with GDP controls for economic size and focuses the analysis on trade instead of the size of the economy. In addition, data on this metric is readily available (Fujii 2019). Although the normalization of trade with GDP is advantageous, it introduces another challenge — the dominance of variability in GDP. According to Fujii (2019), the variability of trade openness can be decomposed into: the variability of trade, the variability of GDP, and the co-variability between trade and GDP. His variance decomposition reveals that heterogeneity in country size or size of GDP can be too influential in explaining observed variation in trade intensity, i.e., variations in the trade ratio from country to country can be attributed more to variations in GDP than trade. Trade intensity also creates another challenge, the domestic economy size bias. According to Squalli and Wilson (2011), this bias emanates from the ratio’s singular focus on the relative position of a country’s trade performance compared with its domestic economy, imposing a ‘relative size penalty’ on larger economies and ranking them as relatively closed (see Figure A1 in the Appendix). Similarly, Tang (2011) observed that due to the size bias, the trade ratio tends to understate the degree of openness of large economies relative to small ones. For example, this measure ranks the U.S. way below Eswatini and Tajikistan, while China is similarly ranked behind Cambodia and the Lao People’s Democratic Republic — suggesting that the US and China are relatively closed economies. Tang (2011) associated the country size bias with the Balassa-Samuelson effect. He argued that the Balassa-Samuelson effect exaggerates the size of large economies through the prices of non-tradable goods and services which tend to be higher in high income countries than in low-income countries. The Balassa-Samuelson effect is consequential for the measurement of trade using the trade intensity ratio because the share of the non-tradeable sector in GDP generally tends to be larger in large economies than small economies. Some studies, for example, Alcalá and Ciccone (2004) and Dollar and Kraay (2003) use the Purchasing Power Parity (PPP)-based GDP in the denominator of the trade ratio to mitigate this. In doing so, they scrape off the part of the size bias associated with non-tradable price differentials. To remedy the underlying size bias more fully, Tang (2011) recommended adjusting the trade intensity ratio for each country by its share in global trade instead. This better manages the effect of changes in a country’s trade that are related to outside supply and demand versus their exclusive inclination to trade. 2.2.2 Control variables in the Trade-Growth Model This study controls for important growth determinants, while also deploying country and time fixed effects. First, countries that invest in capital formation tend to grow faster. The more capital a country accumulates and deploys, the faster its economy grows (Hsieh and Lai 1994; Phetsavong and Ichihashi 2012; Ramirez and Nazmi 2003). Capital investment improves productivity by increasing infrastructure and equipment, as well as upgrading existing, and depreciating equipment. 9 The experience of the East Asian Tigers 10 demonstrated that increased domestic savings (in a stable, market-oriented environment) supports the accumulation of capital, given the relationship between savings and investment. In turn, this facilitates technology transfer and boosts productivity (Stiglitz and Yusuf 2001). 8 This measure of trade openness features frequently in a plethora of cross-country studies that investigate a variety of questions concerning trade openness (Cermeño, Grier, and Grier 2010; Fatás and Mihov 2001; Goldfajn and Valdéz 1999; Levine and Renelt 1992; Rodrik 1998; Yeyati and Panizza 2011). 9 Capital formation represents increases in capital goods, namely equipment, tools, transportation assets, and infrastructure. 10 For example, Hong Kong SAR, China; Singapore; the Republic of Korea; and Taiwan, China. 5 Human capital is also in important determinant of economic growth according to several studies in the 1990s (Eaton and Kortum 1996; Edwards 1992; Frankel and Romer 1999; Harrison 1996; Knight, Loayza, and Villanueva 1993; Lee 1993; Levine and Renelt 1992; Sachs and Warner 1995). It captures the accumulation of ability and skill of a labor force acquired either through formal education or on-the-job training. 11 Human capital plays a crucial role in enabling innovation and R&D activities in developed countries. In less developed countries, it allows for the assimilation of new technologies developed by advanced countries and transferred through trade (Abramovitz 1986; Benhabib and Spiegel 2003). As a bedrock of the new endogenous growth theories, human capital not only serves as an engine of growth, but also as a productive input along with labor and physical capital. Several theoretical studies acknowledged the role of human capital in economic growth (Karam and Zaki 2015; Lucas 1988; Mankiw, Romer, and Weil 1992) and technology spillovers (Abramovitz 1986; Benhabib and Spiegel 2003). The impact of inflation, often measured by the consumer price index or the GDP deflator, on economic growth has been the subject of intense interest and debate in the literature. Some researchers find evidence that inflation is a robust growth determinant (Gillman and Kejak 2005; Mirestean and Tsangarides 2009). However, other studies have provided evidence of the negative impact of inflation on medium and long- run growth (Barro 1991; Barro 2001; Chari, Jones, and Manuelli,1996; Gylfason and Herbertsson 2001; Kormendi and Meguire 1985). Still others show that the relationship between inflation and economic growth is non-linear. For example, Fischer (1993) has shown that the relationship can be positive for low levels of inflation, but negative or non-significant for high levels of inflation. These results were confirmed by other authors such as Burdekin et al. (2004), Bruno and Easterly (1998), Doyle and Christoffersen (1998), Ghosh and Phillips (1998), Gillman and Kejak 2005, Judson and Orphanides (1999), Khan and Ssnhadji 2001, and Sarel (1996). López-Villavicencio and Mignon (2011), with a sample of both industrialized and emerging economies, also established that inflation is growth-enhancing for advanced countries below a certain threshold, while it exerts a negative effect on growth beyond it. From a theoretical approach, the effects of inflation on economic growth are also mixed, depending on the model. Nevertheless, these theoretical approaches provide some insight regarding the transmission channels. For example, inflation has a positive effect on long-run growth when caused by higher monetary growth that enhances capital accumulation. This channel works when money is regarded as a substitute for capital (Tobin 1965). However, in a scenario where money is required for purchasing capital goods (Stockman 1985), higher anticipated inflation decreases steady-state real balances and capital stock, lowering growth. In endogenous growth models, the relationship is channeled via the marginal product of capital (physical and/or human capital). Inflation acts as a tax on physical capital that decreases the rate of return to capital and tends to lower growth. Thus, inflation can proxy macroeconomic stability and is a crucial variable to consider in growth models. There are other growth determinants included in many studies on the trade-growth nexus, each of which were effective to various degrees depending on several factors. For example, several studies have included population growth in their empirical investigation (Busse and Königer 2012; Frankel and Romer 1999; Greenaway, Morgan, and Wright 1998; Heitger 1987; Knight, Loayza, and Villanueva 1993; Lee, Ricci, 11 There is a wide range of measures of human capital including primary, secondary, and tertiary school enrollment ratio, average years of schooling, life expectancy, and composite indexes. For example, following Daude and Fernández-Arias (2010), coupled with the standard approach of Hall and Jones (1999), Pinat and Didier (2013) constructed the human capital index as a function of the average years of schooling. The most popular used is the gross enrollment ratio in secondary school. The gross enrollment ratio in secondary school measures the flow of human capital. 6 and Rigobon 2004 ; Levine and Renelt 1992; Rodrik, Subramanian, and Trebbi 2004). 12 Population growth brings both benefits that can lead to growth (Baker, Delong, and Krugman 2005; Savaş 2008), 13 and challenges that can hinder growth. 14 Financial development has also been identified as a relevant growth determinant in the 2000s (Aghion, Howitt, and Mayer-Foulkes 2005; Beck 2003; Chang, Kaltani, and Loayza 2009; Dollar and Kraay 2004; Easterly and Levine 2001). Few studies, which have tended to limit their analysis to commodity export countries, have found that natural resource endowment can be an important growth determinant (Calderón, Cantú, and Zeufack 2020; Fosu 2011; Redmond and Nasir 2020). Another growing strand of the literature, especially in the 1990s and 2000s, has established a positive effect of institutional quality on growth 15 (Alcalá and Ciccone 2004; Cavallo and Frankel 2008; Chang, Kaltani, and Loayza 2009; Dollar and Kraay 2003; Dollar and Kraay 2004; Edwards 1992; Rodrik, Subramanian, and Trebbi 2004; Sachs and Warner 1995). 2.3. Addressing the endogeneity problem Studies that investigate the trade-growth nexus are confronted with the problem of endogeneity as mentioned at the outset. The trade-growth literature raises an accounting problem associated with the fact that net exports are part of GDP through the accounting identity. 16 Furthermore, trade is not the exclusive driver of GDP growth. Therefore, empirical findings from early OLS-based trade-growth model utilizing the trade intensity (trade as a percentage of GDP) were undermined by the problem of endogeneity (Dowrick 1992; Feder 1983; Greenaway and Sapsford 1994). The Instrumental Variable (IV) method appears to have become the signature technique in trade-growth literature to address the problem of endogeneity (Esfahani 1991; Frankel and Romer 1999; Harrison 1994).17 The IV method is a general approach to estimating causal relationships using observational data. It can be used when standard regression estimates of the relationships of interest are subject to biases due to reverse causality, selection bias, measurement error, or omitted variable bias. The IV method addresses these problems by finding a suitable ‘instrumental’ variable that is correlated with the independent variable but is neither related to control variables or the dependent variable. In this way, its estimated causal influence, or ‘unique variation’, can be free from endogeneity bias, and attributed to the independent variable (Angrist and Krueger 2001; Pokropek 2016). Frankel and Romer (1999), Brueckner and Lederman (2015), and Feyrer (2018) have deployed the IV method to investigate the trade-growth nexus. Among past contributions on the choice of IVs in the trade- 12 While some authors theoretically and empirically emphasized the positive impact of population growth on economic growth (Aghion and Howitt 1992; Baker, Delong, and Krugman 2005; Klasen and Lawson 2007; Romer 1990; Tumwebaze and Ijjo 2015), there are others who have come to the opposite conclusion (Banerjee 2012; Linden 2017; Malthus 1872; Yao, Kinugasa, and Hamori 2013). 13 The age structure matters because an aging population tends to lower labor-force participation and savings rates and may slow down the rate of economic growth, while a young population tends to achieve the opposite (Bloom, Canning, and Fink 2010; Savaş 2008). 14 Ladd (1992) argues that a higher population density might increase crime, leading to rising public safety costs. 15 Institutions consider several aspects, including contracts and contract enforcement, protection of property rights, the rule of law, government bureaucracies, and financial markets. They also include habits and beliefs, norms, social cleavages, and traditions in education (so-called informal institutions). According to North (1990), “institutions are the rules of the game in a society, […] the humanly devised constraints that shape human interactions. […] They structure incentives in human exchange, whether political, social, or economic”. 16 The national income or product identity describes the way in which the gross domestic product (GDP) is measured, as the sum of expenditures in various broad spending categories. The identity, shown below, says that GDP is the sum of personal consumption expenditures (C), private investment expenditures (I), government consumption expenditures (G), and expenditures on exports (Ex) minus expenditures on imports (Im): = + + + ( − ). 17 Some studies use a suitably designed randomized controlled experiment to estimate causal effects. However, even though randomized controlled experiments can circumvent endogeneity problem, they are often not feasible, and subject to other challenges. For example, a randomized controlled experiment could be prohibitively expensive, unethical, and/or have questionable external validity. Even when randomized controlled experiments are available, such as clinical trials of medical procedures, it is of interest to validate the experimental predictions using information on outcomes in the field. 7 growth space, Frankel and Romer’s (1999) publication emerges as an early influential study. Frankel and Romer (1999) addressed the endogeneity problem by deploying a gravity model using the (static) distance between countries (among other factors) to predict trade between bilateral country pairs. Then they used their bilateral trade predictions to construct an exogenous instrument for total trade in each country. They found that trade has a significant impact on growth, with the elasticity of “trade openness” to GDP/capita ranging from 1.5 percent to 2 percent.18 While their exogenous geography-based instrument addresses the problem of reverse causality, their approach was criticized on the grounds that it does not meet the exclusion restriction (Rodriguez and Rodrik 2000; Rodrik, Subramanian, and Trebbi 2004). Rodrik, Subramanian, and Trebbi (2004) indicated that Frankel and Romer’s (1999) results suffer from omitted variable bias due to excluded determinants of growth specifically institutional quality, and distance from the equator which affect GDP through non-trade channels.19 Brueckner and Lederman (2015) addressed the challenge of endogeneity by using two different IV strategies, focusing on Sub-Saharan Africa (SSA). For their first instrument, they estimated the effect of GDP growth on trade, i.e., the response of “trade openness" to variations in GDP per capita, using rainfall as an instrument for GDP per capita. 20 They reasoned that by construction, the model’s residuals obtained from their regression are exogenous to within-country variations (i.e., non-trade-related) in GDP per capita. As a second step, they used their resulting residual as an instrument to estimate the effect of trade openness on GDP growth. Brueckner and Lederman’s second IV strategy uses GDP growth rates of approximately 30 OECD countries as an instrument for trade openness of selected economies in SSA. This assumes that these countries are the mainstay of trading partners for the SSA countries included in their sample. They argue that this instrument avoids reverse causality because African countries are less likely to influence the domestic economies of OECD nations. Their results suggest that the elasticity of GDP per capita to trade openness is approximately 0.5 – 0.6 percent, depending on the IV strategy. Building on Frankel and Romer (1999), Feyrer (2018) attempted to address the recalcitrant trade-growth omitted variable bias with an instrument that sought to satisfy the exclusion condition. He noted the rise in air transport over time as documented by Hummels (2007). Feyrer also noted that evolving technology affects the ease, speed, and cost of transporting goods between trade partners providing a practical way of conceptualizing distance in the trade-growth model in a dynamic, rather than static, manner. He generated a time-varying geographic instrument using the distance between countries (sea distance and air distance) to predict trade between bilateral pairs. 21,22 Feyrer generated time series predictions for bilateral trade based on exogenous “effective distance,” i.e., distance scaled by the propensity of each trade partner to trade (average aggregate trade), the population at the beginning of the sample period, and the log of country land area. Next, he summed the bilateral predictions for trade to generate a panel of predictions for overall trade for each country in the sample over time. Then, he used these trade predictions as an instrument in panel regressions of trade on income per capita, including country-specific effects to control for time-invariant factors that are correlated with GDP. His findings suggest that trade has a significant effect on income with an elasticity of roughly one half. 18 Trade as a percentage of GDP. 19 Remoteness also correlates with being nearer to the equator, which is associated with worse health conditions and institutions. 20 This IV strategy builds on prior literature that has established a robust effect of rainfall on SSA countries' GDP per capita (Barrios, Bertinelli, and Strobl 2010; Brueckner and Ciccone 2011; Miguel, Satyanath, and Sergenti 2004). 21 Feyrer relies on bilateral great circle distances (the measure of air distance) from the CEPII. 22 Feyrer (2009) uses a similar methodology to estimate the impact of trade on GDP by using the temporary closure of the Suez to construct an instrument that is entirely about trade by sea. 8 These three authors, among many others, have chipped away at the primary problems associated with trade- growth modeling, especially the problem of endogeneity. However, the issue has not been completely laid to rest. For example, even Feyrer’s instrument does not fully satisfy the exclusion restriction as he accepted that his measure of air transport includes the movement of people, which can affect GDP via non-trade channels (e.g., tourism, migration, etc.). This raises an attribution problem because it becomes difficult to separate the ‘pure’ effect of trade from that of non-trade related travel. In line with Feyrer’s concerns, Campante and Yanagizawa-Drott (2018) found that business links and capital flows have increased as air travel became cheaper. Furthermore, biases such as Tang’s country-size bias, remain. Therefore, further work is needed to refine such instruments further. 3. METHODOLOGY 3.1. Model specification This study leverages the Mankiw, Romer, and Weil (1992) model which proposes three fundamental determinants of economic growth: human and physical capital accumulation and labor. The main empirical model for our study is specified in equation (1) which augments the Mankiw, Romer, and Weil (1992) model by including inflation: ln(, ) = + + + , + � ℎ ℎ , + , (1) ℎ=1 Where ln , represents the logarithm of GDP per capita (in US dollars), , is adjusted trade share (adjusted ratio of the sum of imports and exports to GDP; see Table A4 for more details) and is the target parameter which measures the relationship between economic growth and trade. The use of , allows our study to depart from earlier literature which has commonly relied on the trade intensity as a proxy for trade (Irwin and Terviö 2002; Dollar and Kraay 2004; Cavallo and Frankel 2008; Brueckner and Lederman 2015; Didier and Pinat 2017). ℎ , is a vector of control variables (h=3) including investment, human capital and inflation, and , represents an error term. Each represents an unobserved individual country-specific effect and each is a time-specific effect; = 1, … , denotes the country ( = 197) and = 1, … , denotes time in years ( = 51). is a constant which estimates the average economic growth related to the base country and base year (1970) when the rest of the regressors are not statistically different from zero. Table 1 provides a list of variables used in the estimation of equation (1) including the description of how each is measured as well as the sources of data. Table 1. Variables, measures, and sources Variables Measures Sources Dependent variable GDP per capita Gross domestic product divided by midyear population. WDI Independent variables Adjusted Trade Share: measured as total trade (i.e., export plus Adjusted Trade imports) as a percentage of GDP, the ratio is then adjusted for WDI share (% GDP) each country’s share of global trade (see Table A4) Inv (% GDP) Gross fixed capital formation (% GDP) WDI Human cap (%) Secondary school enrollment ratio (% total enrollment) WDI Inflation Consumer price index (base year =2010) WDI Note: WDI refers to the World Bank World Development Indicators databases. 9 The choice of control variables in our model is informed by past research, e.g., Balassa (1978), Calderón, Cantú, and Zeufack (2020), Cavallo and Frankel 2008, Chang, Kaltani, and Loayza (2009), Dollar (1992), Dollar and Kraay (2003), Easterly and Levine (2001), Eaton and Kortum (1996), Fosu (2011), Frankel and Romer (1999), Greenaway, Morgan, and Wright (1998), Levine and Renelt (1992), Rodrik, Subramanian, and Trebbi (2004), and Sachs and Warner (1995). The literature has mentioned additional determinants of growth; multiple control variables were considered and tested. For instance, simultaneous inclusion of both labor quality and availability proved unnecessary econometrically. 3.2. Model estimation techniques As mentioned earlier, the estimation of equation (1) using OLS presents some empirical challenges, mainly the problem of endogeneity. This study addresses this problem by using an IV approach with innovative instruments to test the hypothesis that trade intensity has a significant and positive causal effect on growth. The estimation follows a two-step process to assess causal relationship between trade and growth. First, we examine the correlation between trade intensity and growth using a bivariate model and then we add control variables. We also deploy a Generalized Method of Moments (GMM) approach as a robustness check specifically to test the sensitivity of the results to a different estimation method. 3.2.1. Instrumental Variable Approach To address endogeneity, our study estimates equation (1) using an IV approach. The strength of an IV approach is heavily reliant on the identification of suitable instruments for endogenous regressors (Acemoglu, Johnson, and Robinson, 2001; Frankel and Romer 1999). Our study relies on two separate IV strategies, obtained through the combination of several different trade instruments: “lagged air transport freight capacity,” “the mean GDP of top five trade partners,” and “the mean distance to the world top five largest traders.” The use of these instruments is justified on both an intuitive and technical grounds below. The IV1 strategy – Integrating Air transport freight capacity The economic intuition for the study’s choice of instruments is predicated on refinements of the Frankel and Romer (1999) instrument by Feyrer (2018). Trade by air represents a material component of trade. While over 80% of the volume of international trade in goods is carried by sea,23 air cargo transports over US $6 trillion worth of goods, including those where transport speed is critical, accounting for a material 35% of world trade by value. 24 Feyrer (2018) incorporated both sea and air distance to create an instrument. However, he relied on flight distance as a proxy for air-channeled trade. This instrument includes passenger flight and introduces the contribution of non-trade growth factors (e.g., tourism, migration) into the analysis, and distorts the coefficient estimating the effect of trade on GDP growth. Feyrer (2018) considers this inflated coefficient as a ‘globalization effect’ which comprises pure trade effects, and other effects associated with changes in factors such as technology transfer or foreign direct investment (FDI). Our study further refines the “predicted exogenous trade instrument” by focusing purely on air-freight capacity, which measures only the transportation of freight/cargo and excludes passenger air travel. Instrumenting trade with air freight captures the change in trade logistics, spurred, in part, by a ten-fold reduction in the cost of air freight, as aeronautical technology has improved over time (Hummels 2007). 23 UNCTAD Review of Maritime Transport 2021. 24 IATA Value of Air Cargo. 10 By using air transport freight capacity instead of physical flight distance to generate the instrument, this study separates the effect attributable exclusively to trade not other growth predictors. 25 The estimation process for the IV1 strategy follows the Feyrer (2018) approach, and it is executed in two steps. The first step involves the use of a gravity model to predict trade based on geography, incorporating geographic distance and replacing air distance with air transport freight capacity (we included one period lag of air transport freight capacity to avoid reverse causation). The suitable equation for predicting trade for each pair over time becomes: ln� � = + + + + , ln�dist � + , ln(airfreight −1 ) (2) + , ln�airfreight −1 � + X + Where is bilateral trade between country i and country j at time t, dist is the bilateral distance between i and j, airfreight −1 is air transport freight capacity in period t-1, X is a set of controls for pair countries’ time invariant characteristics such as colonial relationships and shared borders, , , and represent country and time fixed effects. Building on Feyrer’s (2018) approach, we controlled for fixed effects. 26 The second step aggregates predicted trade obtained from equation (2), as defined in equation (3) below. The aggregated predicted trade is used as an exogenous instrument to assess the impact of trade on growth. � = ∑≠ �, ln(airfreight−1 )+ �, ln�dist �+ �, ln�airfreight−1 � � + � + � + (3) The IV2 Strategy – Top 5 Trade Partners, Mean Distance to World’s largest traders. The IV2 strategy uses three instruments: the mean GDP of top five trade partners (in exports and imports, separately), the mean distance to the world largest traders, and the lag of air transport freight capacity. Historically, studies have explored different definitions of a country’s GDP as the dependent variable e.g., country GDP (Lee 1993; Rodrik, Subramanian, and Trebbi 2004), growth rates (Kohli and Singh 1989; Levine and Renelt 1992), or various measures of GDP-inducing productivity gains (Bodman 1996; Kunst and Marin 1989; Marin 1992; Nishimizu and Robinson 1984). Trade economics dating back to Ricardo (1817) imply that, in the case of increasing supply and/or demand, given certain circumstances, a country would export and/or import more with its trading partners. This study’s IV Strategy intuits that the increasing GDP of a country’s trading partner can influence that country’s GDP through trade. Our study tests this hypothesis by moving GDP to the right side of the equation, that is, a country’s trading partners’ GDP becomes an exogenous instrument to our “dependent variable country’s” GDP. The main assumption is that the growth of trade partners is external to domestic growth. However, as an instrument “trading partner GDP” may not fully satisfy the exclusion restriction given the potential for investment capital flows 25 Air freight is the volume of freight, express, and diplomatic bags carried on each flight stage (operation of an aircraft from takeoff to its next landing), measured in metric tons times kilometers traveled. Data are obtained from the World Bank World Development Indicators. 26 Fixed effects are essential to isolate growth determinants related to relatively constant country characteristics, such as climate or institutional quality. Such effects may affect growth but are unresponsive to changes in trade. They minimize omitted variable bias by eliminating the relevance of the non-trade channels whose effects on GDP are successfully restricted to time varying bilateral relationships. 11 between healthy trade partners. To address this, we rely only on top five trade partners so that the trade dimension between our bilateral pair is more heavily weighted than the investment dimension.27 To further complement our Top five trade partners instrument, we use the mean distance to the world’s largest traders as an instrument for trade. Bilateral distance is an important determinant of international trade in gravity models. Countries tend to exchange more with their neighbors, especially when these neighbors are among the world’s largest traders. The closer a country is to a global production hub or market, the more it trades. On the contrary, countries that are separated from each other with natural obstacles, such as open landmasses or oceans, will tend to trade less. Even as Global Value Chains (GVCs) grow, distance remains a robust determinant of trade due to associated trade costs. Fernandes, Kee and Winkler (2022) used distance to top trading partners as a determinant of a country’s participation in GVCs. Therefore, this IV strategy adopts this concept, but it uses the mean distance to the world’s largest traders (top five traders) as an instrument for trade rather than an independent variable and this simplifies the gravity model to: � , = . 5. + . 5. + ℎ,−1 + . 5 , (4) , , 5. 5. Where . , and . , are respectively the mean GDP of top five export and import partners , ℎ,−1 is lagged air transport freight capacity and . 5 , represents the mean distance to the world’s largest traders. Testing the validity of chosen instruments In general, proposed instruments must meet certain requirements to be considered valid, namely: the relevance and exclusion conditions (Staiger and Stock 1997). To be relevant, the instrument must be strongly associated with the endogenous variable, that is, the strength of the correlation between the instrument (e.g., air freight transport) and trade. Weak association between trade and the instrument can expose the estimation to weak instrument bias. To test for the relevance condition, it is standard procedure to compute an F-test on the null hypothesis that the first stage coefficient of the instrument is equal to zero (Brueckner and Lederman 2015; Stock and Yogo 2005). The rule of thumb for strong instruments is that the F-statistic must be at least 10, even though higher is preferred. 28 The second consideration is the exclusion condition, which states that the instrument is uncorrelated with the error term. In other words, the instrument must affect GDP per capita only through trade intensity. The Hansen J-test is used to check whether the instrument is exogenous by testing the null hypothesis that the instruments are jointly valid. The results of these tests are reported and discussed in detail in the results section. 3.2.2. Generalized Method of Moments (“GMM”) In line with Brueckner and Lederman (2015), this study utilizes GMM. More specifically, this study uses GMM to check the robustness of the findings of the IV approach, that is, whether the results are sensitive to the choice of estimation method. The GMM estimator was originally developed by Holtz-Eakin, Newey, and Rosen (1988) and advanced by Arellano and Bond (1991). Unlike the IV approach, the GMM uses 27 The possibility that an increase in partners’ GDP can impact domestic country GDP by boosting investment inflow for the country is mitigated by focusing exclusively on top five trade partners. If the top five trade partners were identical to the top five sources of foreign investment, then an increase in partners’ GDP would affect not only trade between bilateral pairs, but also investment from one country to another, violating the exclusivity condition. 28 For statistical significance, the F-statistic must be higher than the Stock-Yogo weak identification test critical values. 12 internal instruments to address potential endogeneity challenges. The GMM estimation is executed in three steps: (1) specify the regression equation as a dynamic panel data model; (2) take first differences to remove unobserved time-invariant country-specific fixed effects; and (3) instrument the right-hand-side variables in the first-differenced equation using levels of the series, lagged two periods or more, while ensuring that the model is not overidentified (Blundell and Bond 1998). The model is specified as follows: ln[], = + + + ∅ ln(),−1 + � ℎ ℎ + , (6) =1 Where ln[], is the logarithm of GDP and ℎ represents growth determinants, including trade and control variables for countries over time , and ℎ is the number of growth determinants where ℎ = 1, 2, … . The GMM estimates Equation (6) and identifies inherent dynamic panel bias associated with the correlation between explanatory variables and error terms (Anderson and Hsiao 1982; Nickell 1981). 29 Following standard practice for GMM estimation, this study tests for panel stationarity or presence of unit roots using the Levin and Lin (1992) (LL-test) and the Maddala and Wu (1999) (MW-test) tests. The longer the time series dimension of the panel dataset, the greater the need to test for panel stationarity. In addition, the validity of the results of the GMM estimator can be affected by overidentification of instruments and second-order autocorrelation of the residuals. The system GMM estimator performs uses lagged values of endogenous variables as instruments, the number of instruments can easily explode as the time-series dimension of the dataset becomes longer weakening the test for over-identifying restrictions due to instrument proliferation or overidentification (Roodman 2009). According to Windmeijer (2005), GMM becomes more efficient when the lag length is controlled to use fewer instruments in the estimation. The Sargan or Hansen tests are diagnostic tests designed to establish whether additional instruments associated with the system GMM estimator are valid. 30 Practically, when using the system GMM estimator, the instrument count can be reduced creating one instrument per variable instead of different instruments for each variable and for each period (Daumal 2010). The result is a smaller set of instruments without compromising the optimal number of lags required (Roodman 2009). In addition, second-order autocorrelation of the residuals (serially correlated residuals) undermines the results of the GMM estimator by rendering instruments inconsistent. The Arellano–Bond test for first order (AR1) and second order (AR2) serial correlation are used to detect this problem. 3.2.3. Sample, data type and sources 29 The validity of the results of the GMM estimator can be affected by overidentification of instruments and second-order autocorrelation of the residuals. The system GMM estimator uses lagged values of endogenous variables as instruments. The number of instruments can easily explode as the time-series dimension of the dataset becomes longer (Roodman 2009). This weakens the test due to instrument proliferation or overidentification. According to Windmeijer (2005), GMM becomes more efficient when the lag length is controlled to use fewer instruments in the estimation. The Sargan or Hansen tests are designed to establish whether additional instruments associated with the system GMM estimator are valid. In addition to potential challenges with the number of instrument-based data points, there is the potential for second-order autocorrelation of (serially correlated) residuals. This can undermine the results of the GMM estimator by rendering instruments inconsistent. The Arellano–Bond tests for first order (AR1) and second order (AR2) serial correlation are used to detect this problem, since in the GMM approach, first order autocorrelation is allowed while second order correlation is not. 30 There is a tension between the two tests because although, on the one hand, the Hansen test is robust to heteroscedasticity, it is sensitive to instrument proliferation. On the other hand, the Sargan test is not sensitive to instrument proliferation, but it is not robust to heteroscedasticity. Therefore, as a rule of thumb, the Hansen test is used with an eye on instrument proliferation which is curtailed by ensuring that for every specification, the number of instruments is less than the corresponding number of countries. 13 This study utilizes a large N and large T panel data structure comprising 197 countries over 51 years over a period of 51 years spanning 1970-2020. 31 The full sample is composed of four subsamples: 28 low-income countries (LICs), 46 lower-middle-income countries (LMICs), 56 upper-middle-income countries (UMICs) and 67 high-income countries (HICs). 32 Table 1 provides information on the relevant data sources. 4. RESULTS AND FINDINGS 4.1 Descriptive statistical analysis Descriptive statistics reflect a high degree of variability and heterogeneity within the panel. This is in line with expectations given the variety of countries included covering different regions and income groups (see Table A3 in the Appendix). Pairwise correlations show evidence of a positive correlation between trade intensity and GDP per capita, even though the correlation appears low as reflected by a coefficient of 0.4 (see Table A1 in the Appendix). Similarly, Figure 2 shows that trade intensity is positively associated with GDP per capita. This indicates that, on average, countries that trade more tend to have higher level of GDP per capita.33 While the correlation analysis provides evidence of a positive relationship between trade intensity and GDP per capita, correlation does not imply causality. Figure 2. The relationship between trade intensity and growth A- Trade (% of GDP) - 1970-2020 B-Adjusted trade (% of GDP) - 1970-2020 12 12 Log of Av. GDP per capita (1970-2020) Log of Av. GDP per capita (1970-2020) CYM CYM 11 BMU CHE LUX SMR 11 BMU SMR LUXCHE FRO NOR FRO QAT VIR VIR NOR QAT DNK ARE USA ISL SWE ISR DNK GUM NLD ARE MAC SXM IRL SXM ISL MAC GUM ISRAUS FIN SWEIRL JPN AUT USA CANNLD DEU BEL 10 JPN AUS FRA GBR FIN CAN DEU ITA NZL NCL AUT GRL KWT BRN BEL SVNABW CUW HKGSGP 10 GRL ABWBRN NZL NCL MNP CUW SVN KWT ITA GBR FRA SGP HKG ESP BHS MNP CYP PYF BHS CYP BHR PRI ESP PYF GRC PRT KOR PRI CZE SVK BHR EST ASM EST MLTGRC CZE KOR SAU SAU HRV LVA LTU MLT ASM BRB LVA HRV PRT LTU HUN OMN SVK BRB OMN HUN ATG 9 LBYPOL SYC ATG NRU 9 NRU SYC URY ARG LBY POL RUS ARG URYRUS CHL MNE MNE GAB CHL ROU KAZ MEX MEXVEN ROU KAZ GAB PAN SRB PAN LBN MUS VEN TUR BRA TUR ZAF LBN SRB CRI XKX DMA BGR MUSMYS GNQBLR MDV DMA BLZ MDV XKX MKD GNQ BIH CRI BRA BGR BLR CUB ZAF MYS 8 COL IRN CUB ECU SYRDOM MKD BLZ BIH SUR AZE IRQ GEONAM JAMBWA TKM 8 MHL SUR FJI NAM BWA COL DOM AZE JAM ECU TKMPERSYR DZA IRN THA IRQ CHN PER DZAPRY ALB THA FJIMHL FSM ALB PSE ARM PRY GEO TUN CHN GTM SLV PSE TUN ARM WSM TON UKRFSM CPV VUT JOR MDA MNG SWZ GUY WSM TON VUT CPV MDAGTM JOR GUY SWZ MNG SLV MAR UKR AGOIDN IDN MAR AGO SSDCOG LKA EGYPHL LKA PHL SSD COG BTN BOLDJI HND 7 NGA COM EGY CMR BOL UZBCIV LAONIC BTN HND PNG MRT KIR VNM SLB TLS DJI 7 COM KIR SLB TLS LAO NIC MRT CMR SEN UZB PNG CIVNGA VNM SDN SENZWE YEM GHA SDNZWE YEM ZMB KGZ GHA IND KEN ZMB KGZ KHM HTI KEN KHM IND PAK HTI BGD TZABEN GIN TJK LSO BEN TZABGD GIN LSO TJK PAK GMB 6 GMB TCD 6 AFG UGABFA RWA ETH MDG NPL MMR NER TCD MLI GNB CAF COD MOZ ERI TGO GNB RWAERI CAFBFA NER SLE MLI UGA MDG NPLTGO AFGMOZ ETHMMR COD SLE BDI SOM BDI SOM 5 5 4 4 3 4 4 5 5 6 6 -8 -6 -4 -2 0 2 4 Log of Av. Trade (% GDP) (1970-2020) Log of Av. Trade (% GDP) (1970-2020) Source: Authors’ calculation using data from the World Bank World Development Indicators (WDI). Av. denotes the average. The red line represents fitted values. 4.2 The causal relationship between trade and economic growth The study follows a two-step estimation procedure, where the first step estimates a simple bivariate model of trade intensity and GDP per capita, and the second step adds control variables to the model. Table 2 31 Panel data estimates can help to provide a more credible inference of parameters by accounting for both inter-individual differences and intra- individual dynamics. They contain more degrees of freedom and more sample variability than pure cross-sectional data or pure time series data, and thereby improve the efficiency of the estimates (Hsiao 2007). The structure of the panel has important implications for the exogeneity assumption. For example, if the cross-sectional dimension is small, then the estimation of time-specific effects becomes harder. Similarly, a small makes the estimation of the individual country-specific effects harder. Thus, the benefits of asymptotics can be realized if both and are large. 32 The sample contains 56 low-income developing countries, 88 emerging market economies and 38 advanced economies (IMF classification). 33 The results are also similar when relying on other measures of international trade (trade value in US$). 14 reports the results of the bivariate model. Table 2 also presents the results of tests for the validity of our IV approaches. The first stage estimates of the IV approach shows that the instruments (exogenous trade intensity obtained from the gravity predictions, average GDP per capita of top five trade partners, lag of air freight capacity and average distance to the world top five traders) are significantly associated with trade intensity, indicating that the instruments are relevant (see Table A5 in the Appendix). Second, the Hansen J-test is applied to test the null hypothesis that the instruments in IV1 and IV2 are jointly valid. The p- values for the test are 0.554 for IV1 and 0.1351 for IV2, respectively, suggesting the null hypothesis cannot be rejected. Thus, the conclusion is that the overidentifying restrictions are not rejected (see Tables A5 and A7 in the Appendix). In addition, weakness identification tests were performed to test for validity of the models’ instruments. The Cragg-Donald Wald F statistic (F statistic = 47.78 for IV1, and 92 for IV2) is higher than Stock-Yogo critical values. Thus, the results are satisfactory, indicating that the instruments are not weak (see Table A7 in the Appendix). Estimating the basic bivariate model using OLS shows that a one percentage point increase in the adjusted trade intensity leads to about 0.25 percent increase in GDP per capita (Table 2— column 1). The OLS method tends to underestimate the impact of trade intensity on GDP per capita (columns 1) compared to the IV method (columns 2 and 3). Furthermore, the IV method can estimate the causal impact of trade intensity on GDP per capita. The estimated coefficient increases when IV strategies are used with IV1, and IV2 yielding higher coefficients of 0.58 and 0.70, respectively (Table 2—columns 2 and 3). These findings align with Brueckner and Lederman (2015) who mentioned the existence of a negative reverse causality bias that causes a downward bias in the OLS estimate. Table 2. The impact of trade intensity on GDP per capita (no controls) (1) (2) (3) OLS IV1 IV2 VARIABLES Log GDP/Cap Log GDP/Cap Log GDP/Cap Log Adj Trade 0.249*** 0.577*** 0.703*** (0.0313) (0.0417) (0.0480) First Stage F- test - 38.61*** 42.54*** Hansen J test p-value - 0.5545 0.1351 Observations 7,808 7,749 5,726 R-squared 0.842 0.753 0.726 Number of id 196 195 166 Country FE Yes Yes Yes Time FE Yes Yes Yes Note: Adj Trade is adjusted trade share, known as trade intensity; Robust standard errors in parentheses. *** < 0.01, ** < 0.05, * < 0.1 As control variables are included in the model, the results reinforce the positive relationship between trade intensity and GDP per capita. The results in Table 3 indicate a statistically significant causal relationship (first stage results are available in Table A6). Importantly, the expected sign of the trade intensity effect is consistent with earlier studies. Notwithstanding, the coefficients of 0.54 and 0.76 for IV1 and IV2, respectively, are relatively similar to those obtained from the estimation of a basic bivariate model. With the inclusion of control variables, the positive causal effect of trade intensity on GDP per capita is 15 statistically significant as 1 percent falls within the 0.45 - 0.96 percent confidence interval. 34 Although the IV2 estimate of the average trade intensity effect appears larger than the estimates based on IV1, the difference is not statistically significant. Based on 95% confidence intervals, IV2 places the trade intensity effect somewhere between 0.56 and 0.96 percent, while for IV1 the interval is 0.45 and 0.62 percent, confirming that both IV results are comparable. Thus, based on the confidence intervals of all the IV specifications, on average, the impact of trade intensity ranges between 0.45 and 0.96 percent. Table 3. Results — The impact of trade intensity on GDP per capita (1) (2) (3) OLS IV1 IV2 VARIABLES Log GDP/Cap Log GDP/Cap Log GDP/Cap Log Adj Trade 0.227*** 0.536*** 0.756*** (0.0395) (0.0428) (0.0997) Invest (% GDP) 0.00322 -0.00265 -0.00503* (0.00253) (0.00165) (0.00258) Human Cap (%) 0.00358*** 0.00168*** 0.000567 (0.00121) (0.000577) (0.000859) Inflation (2010 = 100) 0.00165*** 0.00239*** 0.00311*** (0.000507) (0.000480) (0.000744) First Stage F- test - 36.98*** 11.53*** Hansen J test p-value - 0.6251 0.1640 Observations 4,566 4,553 3,722 R-squared 0.879 0.824 0.779 Number of id 166 163 145 Country FE Yes Yes Yes Time FE Yes Yes Yes Note: Adj Trade is adjusted trade share, known as trade intensity; Invest (% GDP) is private investment measured by gross fixed capital formation in percentage of GDP; Human capital (%) represents secondary school enrollment ratio; and Inflation (2010 = 100) is Consumer Price Index. Robust standard errors in parentheses. Robust standard errors in parentheses. *** < 0.01, ** < 0.05, * < 0.1. The difference between the results of this study versus earlier studies somewhat depends on differences in the measurement of trade and growth as well as the methodology used, making comparisons difficult. For example, Frankel and Romer (1999) used the trade intensity measure, finding a positive impact of trade share on GDP per capita of approximately 2 percent. Feyrer (2018) relied on trade volume and found coefficients that range between 0.5 and 0.75 percent (impact of trade volume on GDP per capita). Brueckner and Lederman (2015) relied on the ratio of trade to GDP and found a short-term coefficient of approximately 0.5 percent and a cumulative long run effect of 2 percent. Because the current study relies on adjusted trade intensity, our findings are different from those of Frankel and Romer (1999), Feyrer (2018), and Brueckner and Lederman (2015) even though Feyrer’s and Brueckner and Lederman’s coefficients are similar. 34 Confidence intervals (95%) are calculated as follows: = ± 2 × . In this case, for IV1, 1 = 0.536 ± 2 × 0.0428 giving a , 1 = [0.45, 0.62], while for IV2 , 2 = 0.756 ± 2 × 0.0997 with 2 = [0.56, 0.96]. 16 With respect to EMDEs, the results confirm the positive and significant impact of trade intensity on GDP per capita. To assess the impact of trade intensity on GDP per capita for EMDEs, this study introduces an interactive variable between trade intensity and a dummy taking the value of one if the country is an EMDE and zero otherwise. The findings highlight that international trade is important the growth of GDP per capita in EMDEs. The average trade intensity effect for EMDEs is higher than the average global trade effect, underscoring the importance of trade for EMDEs’ growth (according to IV1). For EMDEs, a 1 percent increase in trade intensity, on average, leads to an increase in GDP per capita by 0.58 percent for IV1 and 0.71 percent for IV2. Based on 95% confidence intervals, on average, the impact of trade intensity in EMDEs ranges between 0.44 and 0.86 percent (Table A10 in the Appendix). 4.3 Testing the results for robustness The current study takes steps to assess the robustness of the findings by controlling for additional regional fixed effects (to consider unobserved regional specific effects during a given period) and applying a GMM approach to compare the long-term impact of trade intensity on GDP per capita with the average trade intensity impact obtained from the IV estimates. Controlling for regional fixed effects, the results confirm that our findings are robust to the inclusion of additional regional fixed effects (see Table A11 in the Appendix). Additional regional-year fixed effects support and stabilize the results around a fixed range. While IV1 presents a significant coefficient of 0.42 percent, IV2 show that a 1 percent increase in trade intensity increases GDP per capita by 0.52 percent. Thus, the causal impact of trade on economic growth, with the inclusion of additional regional fixed effects of trade intensity, ranges between 0.395 and 0.65 percent. Our finding is also robust to the addition of institutional quality as a control variable. Controlling for institutional quality has been a major criticism of Frankel and Romer (1999). Rodrik, Subramanian, and Trebbi (2004) indicated that Frankel and Romer’s (1999) results suffer from omitted variable bias due to excluded determinants of growth, specifically institutional quality. We therefore controlled for institutional quality, measured by the polity2 indicator per Aisen and Veiga (2013). 35 The results indicate that both IV1 and IV2 present significant and positive coefficients of 0.52 percent and 0.71 percent respectively (Table A14). These results are not significantly different from the baseline estimates, indicating the robustness of our finding to the addition of institutional quality as a control variable. It is important to note that, given the low variance of the polity2 indicator, country and time fixed effects likely already account for institutional quality. This study also uses a GMM estimator as a further robustness check. The credibility of the results of the GMM estimator depends on the outcomes of diagnostic tests for overidentification and serial correlation, as presented in Table 4. In each specification, the number of instruments is lower than the number of countries. In our case, the Hansen test, which diagnoses the problem of overidentification, shows that the additional instruments associated with the system GMM estimator are valid. The two tests for serial correlation show evidence of first order autocorrelation since the null hypothesis of no autocorrelation is rejected using the AR (1) test. However, the AR (2) test cannot reject the null hypothesis and provides evidence of no second order autocorrelation. Therefore, the results of the diagnostic tests indicate that the GMM model is valid. 35 The Polity2 variable measures the quality of institutions. It provides a score ranging from -10 (autarchy) to 10 (democracy). It covers all major, independent states in the global system since 1800 (currently 167 countries). 17 In this case, GMM results suggest that the study’s IV model is robust. Previous studies have deployed GMM estimators in a dynamic panel setting to analyze the trade-growth nexus (Calderón, Cantú, and Zeufack 2020; Chang, Kaltani, and Loayza 2009; Easterly and Levine 2001; Felbermayr 2005; Lee, Ricci, and Rigobon 2004; Ramzan et al. 2019). Thus, similar results between this study’s IV approach and GMM shows that the study’s finding are robust. First, the coefficients on the lagged dependent variable in all the models are found to have a value of less than one and to be statistically significant at the 1% level, providing strong evidence of conditional convergence (Table 4). In addition to internal instruments used in the system GMM, the study has included the list of instruments used in the IV1, and IV2 strategies. The results confirm the finding of a statistically significant average trade effect on GDP per capita. Therefore, the results from the IV strategies are robust to model specification and estimation technique. The system GMM results show that a 1 percent increase in trade intensity increases GDP per capita by 0.60 percent (Table 4—column 1). This coefficient remains fairly stable when including instruments from IV1 and IV2 to the GMM specification. The estimation yields coefficients of 0.58, and 0.45 percent for GMM-IV1, and GMM-IV2 respectively (Table 4—columns 2 and 4). Following Brueckner and Lederman (2015), we also estimated a dynamic model using the IV approach (see Table A15). Given that the model includes country-fixed effects, the lagged dependent variable is instrumented by the second lag of GDP per capita to minimize the bias in the estimated coefficient for lagged GDP per capita. The findings suggest that GDP per capita increase by 0.52 and 0.44 percent (IV1 and IV2 respectively) following a 1 percent increase in adjusted trade share. GMM results provide coefficients that are within the confidence intervals to the IV approach. Table 4. GMM - Impact of trade intensity on GDP per capita (1) (2) (3) GMM GMM-IV1 GMM -IV2 VARIABLES Log GDP/Cap Log GDP/Cap Log GDP/Cap Long Term log Adj Trade � �1 − � 0.6047*** 0.5814*** 0.4514*** (0.0162) (0.0147) (0.0131) log GDP/Cap−1 () 0.909*** 0.916*** 0.889*** (0.0234) (0.0228) (0.0216) Log Adj Trade () 0.0549*** 0.0486*** 0.0503*** (0.0162) (0.0147) (0.0131) Invest 0.0104*** 0.0109*** 0.0102*** (0.00248) (0.00260) (0.00257) Human capital 0.00113 0.00136 0.00260** (0.000987) (0.00101) (0.00101) Inflation -0.000152*** -0.000156*** -0.000155*** (4.22e-05) (4.26e-05) (3.58e-05) Observations 1,308 1,305 1,102 Number of id 165 165 150 Number of Inst 136 138 140 N-IV < N-Groups Yes Yes Yes Time FE Yes Yes Yes F stat 46475 57500 70270 F stat p-value 0.000 0.000 0.000 Hansen J stat 134.1 135.5 127.4 18 Hansen J p-value 0.163 0.173 0.376 AR1 p-value 0.00147 0.00145 0.00189 AR2 p-value 0.203 0.193 0.170 Note: This regression is performed using a system generalized method-of-moments estimators, developed by Arellano and Bovver (1995); and Blundell and Bond (1998). The period goes from 1970 to 2020, and we constructed 13 periods, each variable for a given period representing the average on four consecutive years. Variables - log GDP/Cap−1 is the lag of the logarithm of GDP; Adj Trade is adjusted trade share; Invest (% GDP) is private investment measured by gross fixed capital formation in percentage of GDP; Human capital (%) represents secondary school enrollment ratio; and Inflation (2010 = 100) is Consumer Price Index. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The second order lagged GDP per capita is used as an instrument for the lagged dependent variable. The lag of adjusted trade share is used as an instrument for adjusted trade share. 5. CONCLUSION This study was launched to assess the causal relationship between trade and economic growth, using IV strategies based on innovative instruments. It utilized unbalanced panel data from 197 countries for a period spanning 51 years (ending in 2020). The benefit of relying on an IV strategy is derived from its ability to address the recalcitrant problem of endogeneity in trade-growth modeling. The findings of this study confirm the positive and significant causal impacts of trade intensity on GDP per capita. Specifically, the impact of trade intensity ranges from 0.54 and 0.76 percent for IV1 and IV2 (considering 95% confidence intervals, this ranges between 0.45 and 0.96 percent). The positive impact varies across income groups and highlights a higher positive and statistically significant impact for EMDEs. For development finance institutions, the results indicate that trade support is essential to help countries grow, especially for EMDEs. Therefore, the results also motivate the role of development finance institutions and governments in supporting trade. There is great potential for further research in the trade-growth space. This study has assessed the causal relationship between trade and economic growth using total trade intensity (adjusted trade share) as a measure of trade. As a result, the investigation does not provide answers to other important questions regarding more specific components of trade (e.g., exports versus imports, direction of trade, composition of trade, etc.) Moreover, despite touching on country income groups, this study does not necessarily explore country or country group differences (e.g., leading sectors, products traded, etc.) that would affect individual results. In addition, research can be launched to further separate the effects of trade from other, unique, cross-border activities, such as non-trade-related capital or FDI flows. Finally, the econometric exploration of the role of the advent of internet communication, trade digitization and Artificial Intelligence appears to remain relatively untouched. 19 References Abramovitz, M. 1986. Catching Up, Forging Ahead, and Falling Behind. The Journal of Economic History, 46(2): 385–406. http://www.jstor.org/stable/2122171 Acemoglu, D., Johnson, S., and Robinson, J. A. 2001. The colonial origins of comparative development: An empirical investigation. American economic review, 91(5): 1369-1401. https://www.aeaweb.org/articles?id=10.1257/aer.91.5.1369 Aghion, P., Howitt, P., and Mayer-Foulkes, D. 2005. The effect of financial development on convergence: Theory and evidence. The quarterly journal of economics, 120(1): 173-222. https://academic.oup.com/qje/article/120/1/173/1931451 Aghion, P., and Howitt, P. 1992. A Model of Growth Through Creative Destruction. Econometrica, 60(2): 323–351. https://doi.org/10.2307/2951599 Aguiar, A., Corong, E.L., Van der Mensbrugghe, D., Bekkers, E., Koopman, R.B. and Teh, R. 2019. The WTO Global Trade Model: Technical documentation (No. ERSD-2019-10). WTO Staff Working Paper. https://www.wto-ilibrary.org/content/papers/25189808/242 Aisen, A., and Veiga, F. J. 2013. How does political instability affect economic growth?. European Journal of Political Economy, 29: 151-167. https://www-sciencedirect-com.libproxy- wb.imf.org/science/article/pii/S0176268012000675 Alcalá, F., and Ciccone, A. 2004. Trade and productivity. The Quarterly journal of economics, 119(2), 613-646. https://doi.org/10.1162/0033553041382139 Anderson, J. E. 1979. A theoretical foundation for the gravity equation. The American economic review, 69(1): 106-116. https://www.jstor.org/stable/1802501 Anderson, T. W., and Hsiao, C. 1982. Formulation and estimation of dynamic models using panel data. Journal of econometrics, 18(1): 47-82. https://www-sciencedirect-com.libproxy- wb.imf.org/science/article/pii/0304407682900951 Angrist, J. D., and Krueger, A. B. 2001. Instrumental variables and the search for identification: From supply and demand to natural experiments. Journal of Economic perspectives, 15(4): 69-85. https://www.aeaweb.org/articles?id=10.1257/jep.15.4.69 Arellano, M., and Bond, S. 1991. Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to Employment Equations. The Review of Economic Studies, 58(2): 277–297. https://doi.org/10.2307/2297968 Arellano, M., and Bover, O. 1995. Another look at the instrumental variable estimation of error- components models. Journal of econometrics, 68(1): 29-51. http://dx.doi.org/10.1016/0304- 4076(94)01642-D Baker, D., De Long, J. B., and Krugman, P. R. 2005. Asset returns and economic growth. Brookings Papers on Economic Activity, 2005(1): 289-330. https://www.brookings.edu/wp- content/uploads/2005/01/2005a_bpea_baker.pdf Balassa, B. 1978. ‘Exports and Economic Growth: Further Evidence’, Journal of Development Economics, 5(2): 181–89. https://www.sciencedirect.com/science/article/abs/pii/0304387878900068 Baldwin, R., and Taglioni, D. 2006. Gravity for dummies and dummies for gravity equations. NBER, Working Paper 12516. Issue Date September 2006. https://www.nber.org/papers/w12516 20 Banerjee, R. 2012. Population growth and endogenous technological change: Australian economic growth in the long run. Economic Record, 88(281): 214-228. Barrios, S., Bertinelli, L., and Strobl, E. 2010. Trends in rainfall and economic growth in Africa: A neglected cause of the African growth tragedy. The Review of Economics and Statistics, 92(2): 350-366. http://www.jstor.org/stable/27867541 Barro, R. J. 1991. Economic growth in a cross section of countries. The quarterly journal of economics, 106(2): 407-443. https://doi.org/10.2307/2937943 Barro, R. J. 2001. Human capital and growth. American economic review, 91(2): 12-17. https://www.aeaweb.org/articles?id=10.1257/aer.91.2.12 Beck, T. 2003. Financial dependence and international trade. Review of international Economics, 11(2): 296-316. https://onlinelibrary.wiley.com/doi/abs/10.1111/1467-9396.00384 Bekkers, E., and J. Francois. 2018. “A Parsimonious Approach to Incorporate Firm Heterogeneity in CGE-Models.” Journal of Global Economic Analysis, 3(2): 1–68. https://jgea.org/resources/jgea/ojs/index.php/jgea/article/view/69. Benhabib, J., & Spiegel, M. M. 2003. Human Capital and Technology Diffusion. FRB of San Francisco Working Paper No. 2003-02: https://ssrn.com/abstract=634128 Bhagwati, J. N. 1978. Anatomy of exchange control regimes. In Foreign Trade Regimes and Economic Development: Anatomy and Consequences of Exchange Control Regimes (pp. 7-52). NBER Books. https://www.nber.org/system/files/chapters/c1019/c1019.pdf Blundell, R., & Bond, S. 1998. Initial conditions and moment restrictions in dynamic panel data models. Journal of econometrics, 87(1): 115-143. https://www-sciencedirect-com.libproxy- wb.imf.org/science/article/pii/S0304407698000098 Bodman, P. M. 1996. On export‐led growth in Australia and Canada: cointegration, causality and structural stability. Australian Economic Papers, 35(67): 282-299. https://onlinelibrary.wiley.com/doi/abs/10.1111/j.1467-8454.1996.tb00051.x Brock, W. A., and Durlauf, S. N. 2001. What have we learned from a decade of empirical research on growth? Growth empirics and reality. the world bank economic review, 15(2): 229-272. Bruckner, M., and Ciccone, A. 2011. Rain and the democratic window of opportunity. Econometrica, 79(3): 923-947. http://www.jstor.org/stable/41237773. Brueckner, M. and Lederman, D. 2015. Trade openness and economic growth: Panel data evidence from Sub‐Saharan Africa. Economica, 82: 1302-1323. https://doi.org/10.1111/ecca.12160 Bruno, M., and Easterly, W. 1998. Inflation crises and long-run growth. Journal of Monetary economics, 41(1): 3-26. https://doi.org/10.1016/S0304-3932(97)00063-9 Burdekin, R. C., Denzau, A. T., Keil, M. W., Sitthiyot, T., and Willett, T. D. 2004. When does inflation hurt economic growth? Different nonlinearities for different economies. Journal of macroeconomics, 26(3): 519-532. https://doi.org/10.1016/j.jmacro.2003.03.005 Busse, M., and Königer, J. 2012. Trade and economic growth: A re-examination of the empirical evidence: https://www.econstor.eu/bitstream/10419/57921/1/715302949.pdf Calderón, C., Cantú, C., and Zeufack, A. G. 2020. Trade integration, export patterns, and growth in Sub-Saharan Africa. World Bank Policy Research Working Paper, (9132). https://elibrary.worldbank.org/doi/abs/10.1596/1813-9450-9132 21 Campante, F., and Yanagizawa-Drott, D. 2018. Long-range growth: economic development in the global network of air links. The Quarterly Journal of Economics, 133(3): 1395-1458. https://doi.org/10.1093/qje/qjx050 Cavallo, E. A., and Frankel, J. A. 2008. Does openness to trade make countries more vulnerable to sudden stops, or less? Using gravity to establish causality. Journal of International Money and Finance, 27(8): 1430-1452. https://www-sciencedirect-com.libproxy- wb.imf.org/science/article/pii/S0261560607001180 Cermeño, R., Grier, R., and Grier, K. 2010. Elections, exchange rates and reform in Latin America. Journal of Development Economics, 92(2): 166-174. https://doi.org/10.1016/j.jdeveco.2009.03.002 Chang, R., Kaltani, L., & Loayza, N. V. 2009. Openness can be good for growth: The role of policy complementarities. Journal of development economics, 90(1): 33-49. https://doi.org/10.1016/j.jdeveco.2008.06.011 Chari, V. V., Jones, L. E., and Manuelli, R. E. 1996. Inflation, growth, and financial intermediation. Review-Federal Reserve Bank of Saint Louis, 78: 41-58. Corong, E. L., Hertel, T. W., McDougall, R., Tsigas, M. E., and Van Der Mensbrugghe, D. 2017. The standard GTAP model, version 7. Journal of Global Economic Analysis, 2(1): 1-119. https://www.jgea.org/ojs/index.php/jgea/article/view/47 Daude, C. and Fernández-Arias, E., 2010. On the Role of Productivity and Factor Accumulation in Economic Development in Latin America and the Caribbean", OECD Development Centre Working Papers, No. 290, OECD Publishing, Paris. https://doi.org/10.1787/5kmh5gpqtdjc-en. Daumal, M. 2010. The impact of international trade flows on economic growth in Brazilian states. Review of Economics and Institutions, 2(1). https://rei.unipg.it/rei/article/view/27 Bloom, D. E., Canning, D., & Fink, G. 2010. Implications of population ageing for economic growth. Oxford review of economic policy, 26(4): 583-612. https://doi.org/10.1093/oxrep/grq038 Didier, T., & Pinat, M. 2017. The nature of trade and growth linkages. Policy Research working paper,no. WPS 8161 Washington, D.C. : World Bank Group. http://documents.worldbank.org/curated/en/900341502818068705/The-nature-of-trade-and-growth- linkages Dollar, D. 1992. Outward-oriented developing economies really do grow more rapidly: evidence from 95 LDCs, 1976-1985. Economic development and cultural change, 40(3): 523-544. http://www.jstor.org/stable/1154574 Dollar, D., and Kraay, A. 2003. Institutions, trade, and growth. Journal of monetary economics, 50(1): 133-162. https://doi.org/10.1016/S0304-3932(02)00206-4 Dollar, D., and Kraay, A. 2004. Trade, growth, and poverty. The Economic Journal, 114(493). https://academic.oup.com/ej/article/114/493/F22/5089312 Dowrick, S. 1992. Technological catch up and diverging incomes: patterns of economic growth 1960- 88. The Economic Journal, 102(412): 600-610. https://doi.org/10.2307/2234297 Doyle, M. P., and Christoffersen, M. P. F. 1998. From inflation to growth: eight years of transition. IMF Working Paper No. 1998/100, Washington DC: International Monetary Fund. 22 Easterly, W., and Levine, R. 2001. What have we learned from a decade of empirical research on growth? It's Not Factor Accumulation: Stylized Facts and Growth Models. The world bank economic review, 15(2): 177-219. https://academic.oup.com/wber/article/15/2/177/1667968 Eaton, J., and Kortum, S. 1996. Trade in ideas Patenting and productivity in the OECD. Journal of international Economics, 40(3-4): 251-278. https://doi.org/10.1016/0022-1996(95)01407-1 Edwards, S. 1992. Trade orientation, distortions and growth in developing countries. Journal of development economics, 39(1): 31-57. https://www.sciencedirect.com/science/article/pii/030438789290056F Emery, R.F. 1967. The relation of exports and economic growth. Kyklos, 20(4): 470-486. https://doi.org/10.1111/j.1467-6435.1967.tb00859.x Esfahani, H. S. 1991. Exports, imports, and economic growth in semi-industrialized countries. Journal of development economics, 35(1): 93-116. https://www-sciencedirect-com.libproxy- wb.imf.org/science/article/pii/0304387891900687 Evenett, S. J., and Keller, W. 2002. On theories explaining the success of the gravity equation. Journal of political economy, 110(2): 281-316. https://doi.org/10.1086/338746 Fatás, A., & Mihov, I. 2001. Government size and automatic stabilizers: international and intranational evidence. Journal of international economics, 55(1): 3-28. Falvey, R., Foster, N., and Greenaway, D. 2012. Trade liberalization, economic crises, and growth. World Development, 40(11): 2177-2193. https://doi.org/10.1016/S0022-1996(01)00093-9 Feder, G. 1983. On exports and economic growth. Journal of development economics, 12(1-2): 59-73. https://www.sciencedirect.com/science/article/pii/0304387883900317 Felbermayr, G. J. 2005. Dynamic Panel Data Evidence on the Trade–Income Relation. Review of World Economics, 141(4): 583–611. https://link.springer.com/article/10.1007%2Fs10290-005-0046-4 Fernandes, A. M., Kee, H. L., and Winkler, D. 2022. Determinants of global value chain participation: cross-country evidence. The World Bank Economic Review, 36(2): 329-360. https://doi-org.libproxy- wb.imf.org/10.1093/wber/lhab017 Feyrer, J. 2009. Trade and Income -- Exploiting Time Series in Geography (April 2009). NBER Working Paper No. w14910. https://www.nber.org/papers/w14910 Feyrer, J. 2018. Trade and income—exploiting time series in geography. American Economic Journal: Applied Economics, 11(4): 1-35. https://www.aeaweb.org/articles?id=10.1257/app.20170616 Fischer, S. 1993. The role of macroeconomic factors in growth. Journal of monetary economics, 32(3): 485-512. https://doi.org/10.1016/0304-3932(93)90027-D Fosu, A. K. 2011. Terms of trade and growth of resource economies: A tale of two countries. UNU- WIDER, Helsinki, Finland (2011) 22 pp. [WIDER Working Paper No. 2011/28] Frankel, J. A., and Romer, D. H. 1999. Does trade cause growth?. American economic review, 89 (3): 379-399. https://www.aeaweb.org/articles?id=10.1257/aer.89.3.379 Fujii, E. 2019. What does trade openness measure?. Oxford Bulletin of Economics and Statistics, 81(4): 868-888. https://doi.org/10.1111/obes.12275 Ghosh, A., and Phillips, S. 1998. Warning: Inflation may be harmful to your growth. IMF Econ Rev 45: 672–710. https://doi-org.libproxy-wb.imf.org/10.2307/3867589 23 Gillman, M.,and Kejak, M. 2005. Contrasting models of the effect of inflation on growth. Journal of Economic Surveys, 19(1): 113-136. https://doi.org/10.1111/j.0950-0804.2005.00241.x Goldfajn, I., and Valdes, R. O. 1999. The aftermath of appreciations. The Quarterly Journal of Economics, 114(1): 229-262. https://doi.org/10.1162/003355399555990 Greenaway, D., and Sapsford, D. 1994. Exports, growth, and liberalization: An evaluation. Journal of policy modeling, 16(2): 165-186. https://www.sciencedirect.com/science/article/abs/pii/0161893894900191 Greenaway, D., Morgan, W., and Wright, P. 1998. Trade reform, adjustment and growth: what does the evidence tell us?. The Economic Journal, 108(450). https://www.jstor.org/stable/2566198?seq=1#metadata_info_tab_contents Grossman, G. M., and Helpman, E. 2015. Globalization and growth. American Economic Review, 105(5): 100-104. https://www.aeaweb.org/articles?id=10.1257/aer.p20151068 Gylfason, T., and Herbertsson, T. T. 2001. Does inflation matter for growth?. Japan and the world economy, 13(4): 405-428. https://doi.org/10.1016/S0922-1425(01)00073-1 Hall, R.E. and Jones, C.I., 1999. Why do some countries produce so much more output per worker than others?. The quarterly journal of economics, 114(1): 83-116. https://doi.org/10.1162/003355399555954 Harrison, A. E. 1994. Productivity, imperfect competition and trade reform: Theory and evidence. Journal of international Economics, 36(1-2): 53-73. Harrison, A. E. 1996. ‘Openness and Growth: A Time-series, Cross-country Analysis for Developing Countries’, Journal of Development Economics, 48 (2): 419–47. https://ideas.repec.org/a/eee/deveco/v48y1996i2p419-447.html Hausmann, R., Pritchett, L., and Rodrik, D. 2005. Growth accelerations. Journal of economic growth, 10(4). https://link.springer.com/article/10.1007/s10887-005-4712-0 Heitger, B. 1987. Import protection and export performance — Their impact on economic growth. Weltwirtschaftliches Archiv 123: 249–261 (1987).https://doi.org/10.1007/BF02706662 Holtz-Eakin, D., Newey, W., and Rosen, H. S. 1988. Estimating Vector Autoregressions with Panel Data. Econometrica, 56(6): 1371–1395. https://doi.org/10.2307/1913103 Hsiao, C. 2007. Panel data analysis—advantages and challenges. Test, 16(1): 1-22. https://link-springer- com.libproxy-wb.imf.org/article/10.1007/s11749-007-0046-x Hsieh, E., and Lai, K. S. 1994. Government spending and economic growth: the G-7 experience. Applied Economics, 26(5): 535-542. https://doi.org/10.1080/00036849400000022 Hummels, D. 2007. Transportation costs and international trade in the second era of globalization. Journal of Economic perspectives, 21(3): 131-154. http://www.jstor.org/stable/30033738 Irwin, D. A., and Terviö, M. 2002. Does trade raise income?: Evidence from the twentieth century. Journal of international Economics, 58(1): 1-18. https://www-sciencedirect-com.libproxy- wb.imf.org/science/article/pii/S0022199601001647 Judson, R., and Orphanides, A. 1999. Inflation, volatility and growth. international Finance, 2(1): 117- 138. https://doi.org/10.1111/1468-2362.00021 Tang, K. K. 2011. Correcting the size bias in trade openness and globalization measures. Global Economy Journal, 11(3): 1850235. https://www.worldscientific.com/doi/abs/10.2202/1524-5861.1751 24 Karam, F., and Zaki, C. 2015. Trade volume and economic growth in the MENA region: Goods or services?. Economic Modelling, 45: 22-37. https://doi.org/10.1016/j.econmod.2014.10.038 Keho, Y. 2017. The impact of trade openness on economic growth: The case of Cote d’Ivoire. Cogent Economics & Finance, 5(1): 1332820. https://www.tandfonline.com/doi/full/10.1080/23322039.2017.1332820 Khan, M. S., and Ssnhadji, A. S. 2001. Threshold effects in the relationship between inflation and growth. IMF Staff papers, 48(1): 1-21. https://doi-org.libproxy-wb.imf.org/10.2307/4621658 King, R. G., & Levine, R. 1993. Finance and growth: Schumpeter might be right. The quarterly journal of economics, 108(3): 717-737. https://doi.org/10.2307/2118406 Klasen, S., and Lawson, D. 2007. The impact of population growth on economic growth and poverty reduction in Uganda. Diskussions beiträge aus dem Volkswirtschaftlichen Seminar der Universität Göttingen, No. 133. http://www.econstor.eu/dspace/bitstream/10419/31966/1/534768717.pdf Knight, M., Loayza, N., and Villanueva, D. 1993. Testing the neoclassical theory of economic growth: A panel data approach. Staff papers, 40(3): 512-541. https://link-springer-com.libproxy- wb.imf.org/article/10.2307/3867446 Kohli, I., and Singh, N. 1989. Exports and growth: critical minimum effort and diminishing returns. Journal of Development Economics, 30(2): 391-400. https://www.sciencedirect.com/science/article/abs/pii/0304387889900114 Kormendi, R. C., and Meguire, P. G. 1985. Macroeconomic determinants of growth: cross-country evidence. Journal of Monetary economics, 16(2): 141-163.https://doi.org/10.1016/0304-3932(85)90027-3 Krueger, A. O. 1978. Front matter, Foreign Trade Regimes and Economic Development: Liberalization Attempts and Consequences (No. c3849). National Bureau of economic research. Kunst, R. M., and Marin, D. 1989. On Exports and Productivity: A Causal Analysis. The Review of Economics and Statistics, 71(4): 699–703. https://doi.org/10.2307/1928115 Ladd, H. F. 1992. Population growth, density and the costs of providing public services. Urban Studies, 29(2): 273-295. Leamer, E. E. 1988. Measures of openness. In Trade policy issues and empirical analysis (pp. 145-204). University of Chicago Press. Lee, H. Y., Ricci, L. A., and Rigobon, R. 2004. Once again, is openness good for growth?. Journal of development economics, 75(2): 451-472. https://doi.org/10.1016/j.jdeveco.2004.06.006 Lee, J. W. 1993. ‘International Trade, Distortions, and Long-run Economic Growth’, International Monetary Fund Staff Papers, 40 (2): 299–328. https://www.jstor.org/stable/3867316?origin=pubexport&seq=1#metadata_info_tab_contents Levin, A., and Lin, C.-F. 1992.Unit root test in panel data: asymptotic and finite sample properties (Discussion Paper No. 92-93). University of California at San Diego. Levine, R., and Renelt, D. 1992. A Sensitivity Analysis of Cross-Country Growth Regressions. The American Economic Review, 82(4): 942–963. http://www.jstor.org/stable/2117352 Linden, E. 2017. Remember the population bomb? It’s still ticking. New York Times: Sunday Review, 4. June 2017. 25 López-Villavicencio, A., & Mignon, V. 2011. On the impact of inflation on output growth: Does the level of inflation matter?. Journal of macroeconomics, 33(3): 455-464. https://doi.org/10.1016/j.jmacro.2011.02.003 Lucas Jr, R.E. 1988. On the mechanics of economic development. Journal of monetary economics, 22(1): 3-42. https://doi.org/10.1016/0304-3932(88)90168-7 Maddala, G. S., and Wu, S. 1999. A comparative study of unit root tests with panel data and a new simple test. Oxford Bulletin of Economics and statistics, 61(S1): 631-652. https://onlinelibrary.wiley.com/doi/abs/10.1111/1468-0084.0610s1631 Maisel, S. J. 1968. Some Relationships Between Assets and Liabilities of Thrift Institutions. The Journal of Finance, 23(2): 367–378. https://doi.org/10.2307/2325402 Malthus, T. R. 1872. An Essay on the Principle of Population. Paris : Éditions Gonthier, 1963, 236 pages. Collection : Bibliothèque Médiations. Mankiw, N.G., Romer, D. and Weil, D.N. 1992. A contribution to the empirics of economic growth. The quarterly journal of economics, 107(2): 407-437. https://doi.org/10.2307/2118477 Marin, D. 1992. Is the Export-Led Growth Hypothesis Valid for Industrialized Countries? The Review of Economics and Statistics, 74(4): 678–688. https://doi.org/10.2307/2109382 Michaely, M. 1977. Exports and Growth: An Empirical Investigation. Journal of Development Economics, 4 (1): 49–53. https://www.sciencedirect.com/science/article/abs/pii/0304387877900062 Miguel, E., Satyanath, S., and Sergenti, E. 2004. Economic shocks and civil conflict: An instrumental variables approach. Journal of political Economy, 112(4): 725-753. https://www.journals.uchicago.edu/doi/full/10.1086/421174 Mirestean, A., and Tsangarides, C. G. 2009. Growth determinants revisited. IMF Working Papers, 2009(268), A001. Retrieved Oct 30, 2023. Washington, DC: International Monetary Fund. Nickell, S. 1981. “Biases in Dynamic Models with Fixed Effects.” Econometrica 49 (6): 1417–1426. https://doi.org/10.2307/1911408 Nishimizu, M., and Robinson, S. 1984. Trade policies and productivity change in semi-industrialized countries. Journal of development economics, 16(1-2): 177-206. https://www-sciencedirect-com.libproxy- wb.imf.org/science/article/pii/0304387884901068 North, D. 1990. Institutions and their consequences for economic performance. The limits of rationality, 383-401. Nsiah, C., and Fayissa, B. 2018. Revisiting the short and long-run impacts of trade openness: evidence from the African growth experience. https://www.aeaweb.org/conference/2019/preliminary/paper/7anZQAbA Pagano, M. 1993. Financial markets and growth: An overview. European economic review, 37(2-3): 613- 622. https://ams-forschungsnetzwerk.at/downloadpub/1993_mp_eer-1993.pdf Phetsavong, K., and Ichihashi, M. 2012. The impact of public and private investment on economic growth: evidence from developing Asian countries. Hiroshima University. Pinat, M. and Didier, T., 2013. How Does Trade Cause Growth?. Conference papers 332421, Purdue University, Center for Global Trade Analysis, Global Trade Analysis Project. 26 Pokropek, A. 2016. Introduction to instrumental variables and their application to large-scale assessment data. Large-scale Assessments in Education, 4: 1-20. https://link-springer-com.libproxy- wb.imf.org/article/10.1186/s40536-016-0018-2 Ramirez, M. D., and Nazmi, N. 2003. Public investment and economic growth in Latin America: An empirical test. Review of Development Economics, 7(1): 115-126. https://doi.org/10.1111/1467- 9361.00179 Ramzan, M., Sheng, B., Shahbaz, M., Song, J., and Jiao, Z. 2019. Impact of trade openness on GDP growth: Does TFP matter?. The Journal of International Trade & Economic Development, 28(8): 960- 995. https://www.tandfonline.com/doi/full/10.1080/09638199.2019.1616805 Redmond, T., and Nasir, M. A. 2020. Role of natural resource abundance, international trade and financial development in the economic development of selected countries. Resources Policy, 66, 101591. https://www.sciencedirect.com/science/article/abs/pii/S030142071930950X Ricardo, D. 1817. On the Principles of Political Economy and Taxation. Available online at the Library of Economics and Liberty: https://www.econlib.org/library/Ricardo/ricP.html Rodriguez, F., and Rodrik, D. 2000. Trade policy and economic growth: a skeptic's guide to the cross national evidence. NBER macroeconomics annual, 15: 261-325. https://www.journals.uchicago.edu/doi/abs/10.1086/654419 Rodrik, D., Subramanian, A., and Trebbi, F. 2004. Institutions rule: the primacy of institutions over geography and integration in economic development. Journal of economic growth, 9: 131-165. https://www.jstor.org/stable/40212696?seq=1 Rodrik, D. 1998. Why do more open economies have bigger governments? Journal of Political Economy, 106 (5): 997-1032. Romer, P. M. 1990. Capital, Labor, and Productivity. Brookings Papers on Economic Activity. Microeconomics, 1990: 337–367. https://doi.org/10.2307/2534785 Roodman, D. 2009. A note on the theme of too many instruments. Oxford Bulletin of Economics and statistics, 71(1): 135-158. https://onlinelibrary.wiley.com/doi/full/10.1111/j.1468-0084.2008.00542.x Roodman, D. 2009. How to do xtabond2: An introduction to difference and system GMM in Stata. Stata Journal 9(1): 86-136. https://journals.sagepub.com/doi/abs/10.1177/1536867X0900900106 Sachs, J. D., and Warner, A. 1995. Economic Reform and the Process of Global Integration. Brookings Papers on Economic Activity, 1995(1): 1–118. https://doi.org/10.2307/2534573 Salinas, G., and Aksoy, A. 2006. Growth before and after trade liberalization. Policy, Research working paper; no. WPS 4062. Washington, D.C. : World Bank Group. http://hdl.handle.net/10986/8864 Sarel, M. 1996. Nonlinear effects of inflation on economic growth. Staff Papers, 43(1): 199-215. https://www.elibrary.imf.org/view/journals/024/1996/001/article-A007-en.xml Savaş, B. 2008. The relationship between population and economic growth: empirical evidence from the Central Asian Economies. Orta Asya ve Kafkasya Araştırmaları, (06): 135-153. Shahbaz, M. 2012. Does trade openness affect long run growth? Cointegration, causality and forecast error variance decomposition tests for Pakistan. Econ. Modell. 29 (6): 2325–2339. https://www-sciencedirect- com.libproxy-wb.imf.org/science/article/pii/S0264999312002362 27 Shahbaz, M., Khan, S., and Tahir, M. I. 2013. The dynamic links between energy consumption, economic growth, financial development and trade in China: fresh evidence from multivariate framework analysis. Energy economics, 40: 8-21. https://doi.org/10.1016/j.eneco.2013.06.006 Solow, R. M. 1956. A contribution to the theory of economic growth. The quarterly journal of economics, 70(1): 65-94. https://academic.oup.com/qje/article-abstract/70/1/65/1903777 Squalli, J. and K. Wilson. 2011. A New Measure of Trade Openness, The World Economy, 34 (10): 1745- 1770. https://doi.org/10.1111/j.1467-9701.2011.01404.x Staiger, D., and Stock, J. H. 1997. Instrumental Variables Regression with Weak Instruments. Econometrica, 65(3): 557–586. https://doi.org/10.2307/2171753 Stiglitz, J.E. and Yusuf, S. eds. 2001. Rethinking the East Asian Miracle. Washington, DC: World Bank and Oxford University Press. http://hdl.handle.net/10986/13969 Stock J, and Yogo M. 2005. Testing for Weak Instruments in Linear IV Regression. In: Andrews DWK Identification and Inference for Econometric Models. New York: Cambridge University Press ; 2005. 80- 108. Stockman, A. C. 1985. Effects of Inflation on the Pattern of International Trade. The Canadian Journal of Economics / Revue Canadienne d’Economique, 18(3): 587–601. https://doi.org/10.2307/135021 Thirlwall, A. P. 2000. Trade, trade liberalisation and economic growth: theory and evidence (Vol. 63). Côte d’Ivoire: African Development Bank. Tobin, J. 1965. Money and Economic Growth. Econometrica, 33(4): 671–684. https://doi.org/10.2307/1910352 Tumwebaze, H. K., and Ijjo, A. T. 2015. Regional economic integration and economic growth in the COMESA region, 1980–2010. African Development Review, 27(1): 67-77. https://doi.org/10.1111/1467- 8268.12123 Wacziarg, R., and Welch, K. H. 2008. Trade liberalization and growth: New evidence. The World Bank Economic Review, 22(2): 187-231. Windmeijer, F. 2005. A finite sample correction for the variance of linear efficient two-step GMM estimators. Journal of econometrics, 126(1): 25-51. https://www-sciencedirect-com.libproxy- wb.imf.org/science/article/pii/S0304407604000387 Winters, L. A. 2004. Trade liberalisation and economic performance: an overview. The economic journal, 114(493): F4-F21. WTO (World Trade Organization) (2017). Trade, Technology and Jobs. World Trade Report 2017, WTO. Geneva: World Trade Organization. https://doi.org/10.30875/50d27d62-en Yao, W., Kinugasa, T., and Hamori, S. 2013. An empirical analysis of the relationship between economic development and population growth in China. Applied Economics, 45(33): 4651-4661. Yeyati, E. L., and Panizza, U. 2011. The elusive costs of sovereign defaults. Journal of Development Economics, 94(1): 95-105. https://doi.org/10.1016/j.jdeveco.2009.12.005 28 Appendix Figure A1. The size bias of the trade-to-GDP ratio Ranking: Top 30 traders (% of GDP), 2018 Luxembourg 387% Hong Kong SAR, China 377% Singapore 327% San Marino 309% Djibouti 288% Malta 270% Ireland 212% Vietnam 208% Slovak Republic 191% Seychelles 182% Sint Maarten (Dutch part) 172% American Samoa 168% Belgium 166% Hungary 163% Slovenia 161% United Arab Emirates 160% Netherlands 159% Maldives 154% Bahrain 151% Cyprus 149% Lithuania 149% Czech Republic 148% Estonia 146% Lesotho 143% Antigua and Barbuda 141% Belarus 139% North Macedonia 133% Somalia 132% Malaysia 130% Bulgaria 129% 0% 50% 100% 150% 200% 250% 300% 350% 400% Source: Authors’ calculation based on the World Bank Word Development Indicators (WDI) data. 29 Figure A2. The relationship between exports and growth – Evidence for 2018 and 1970 Exports vs. GDP - (Av. 1970-2020) 31 29 27 Log of GDP (US$) 25 23 21 19 17 15 15 17 19 21 23 25 27 29 Log of Exports (US$) High income Upper middle income Lower middle income Low income Source: Authors’ calculation based on the World Bank World Development Indicators (WDI) data. Figure A3. The relationship between imports and growth – Evidence for 2018 and 1970 Imports vs. GDP - (Av. 1970-2020) 31 29 27 Log of GDP (US$) 25 23 21 19 17 15 18 19 20 21 22 23 24 25 26 27 28 Log of Imports (US$) High income Upper middle income Lower middle income Low income Source: Authors’ calculation based on the World Bank World Development Indicators (WDI) data. 30 Figure A4. A network analysis of bilateral trade by continent 2018 TUR QAT NLD JPN HUN DEU BRA AGO USA ROU NOR KAZ IDN DNK CAN ARE VNM RUS OMN KOR IND EGY CHE ARG ZAF SAU PAK LBY IRL ESP CHL AUS SGP PER MEX IRN FIN CHN AUT SVK PHL MMR IRQ FRA COD BEL SWE POL MYS ISR GBR COL BGD THA PRT NGA ITA HKG CZE BLR Africa Asia Europe North America Oceania South America Source: Authors’ calculation based on the UN COMTRADE database 2018. Note: This graph is a directed network. The size of each node represents the weighted degree of its corresponding economy (the total inward and outward trade from all other economies in the network). For visual reason the thickness of the links was not linked to trade values. This network shows that countries with larger node size represents those highly involved in international trade, and these countries are the one with higher income. 31 Table A1. Pairwise correlations – regressors Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (1) GDP per Capita (US$) 1.000 (2) Trade (% GDP) 0.307 1.000 (3) Exports (% GDP) 0.387 0.955 1.000 (4) Imports (% GDP) 0.194 0.950 0.814 1.000 (5) Adj Trade 0.410 0.368 0.414 0.285 1.000 (6) Adj Exports 0.417 0.381 0.434 0.288 0.992 1.000 (7) Adj Imports 0.398 0.347 0.384 0.273 0.992 0.968 1.000 (8) Invest (% GDP) 0.028 0.175 0.115 0.222 0.102 0.093 0.109 1.000 (9) Human capital 0.556 0.268 0.320 0.186 0.352 0.351 0.348 0.083 1.000 (10) Inflation (2010=100) 0.051 0.013 0.015 0.011 0.008 0.009 0.007 -0.009 0.320 1.000 Source: Authors’ calculation based on the World Bank World Development Indicators (WDI) data Table A2. Pairwise correlation– instruments and independent variables Variables (1) (2) (3) (4) (5) (6) (7) (8) (9) (10) (1) Adj Trade 1.000 (2) Adj Exports 0.992 1.000 (3) Adj Imports 0.992 0.968 1.000 (4) Exogenous Adj Trade (% GDP) 0.236 0.244 0.223 1.000 (5) Exogenous Adj Exports (% GDP) 0.241 0.248 0.227 0.999 1.000 (6) Exogenous Adj Imports (% GDP) 0.231 0.239 0.218 0.999 0.996 1.000 (7) Air freight 0.529 0.489 0.570 0.123 0.130 0.117 1.000 (8) Top 5 Export partners’ GDP 0.155 0.156 0.149 0.064 0.066 0.062 0.178 1.000 (9) Top 5 Import partners’ GDP 0.174 0.176 0.167 0.075 0.077 0.072 0.227 0.782 1.000 (10) Mean distance to top 5 traders -0.237 -0.230 -0.241 -0.075 -0.072 -0.078 -0.089 -0.062 -0.083 1.000 Source: Authors’ calculation based on the World Bank World Development Indicators (WDI) data. 32 Table A3. Descriptive Statistics Variable Obs Mean Std. Dev. Min Max GDP (US$ million) 8648 218.2 1081.1 0.01 23315.1 GDP per Capita (US$) 8648 9107 15520.4 20 133590.2 Trade (% GDP) 7826 81.2 54.5 0.02 863.2 Exports (% GDP) 7826 37.8 29.3 0.005 433.8 Imports (% GDP) 7827 43.4 27.97 0.016 429.4 Adj Trade 7808 0.43 1.1 0 11.7 Adj Exports 7808 0.23 0.57 0 5.8 Adj Imports 7809 0.21 0.51 0 5.9 Invest (% GDP) 7164 22.6 8.2 -2.4 93.6 Human capital 6311 66.3 34.4 0 166.1 Inflation (2010=100) 7376 83.8 465.5 0 22570.7 Source: Authors’ calculation based on the World Bank World Development Indicators (WDI) data Table A4. Measurement of trade variable Measures of trade Concept Computation Sources Trade Volume ( + )in % GDP, ( + ) ( + ) � �×� � adjusted by the World Trade Share 1 ∑ ( + ) =1 Adjusted Trade Squalli and Wilson Share (main Exports share of GDP � �×� � (2011); Tang 1 measure of trade) ∑ ( ) =1 (2011) Imports share of GDP � �×� � 1 ∑=1( ) Table A5. The impact of trade on GDP per capita (no controls) (1) (2) (3) OLS IV1 IV2 VARIABLES l\ Log GDP/Cap Log GDP/Cap Log GDP/Cap Log Adj Trade 0.249*** 0.577*** 0.703*** (0.0313) (0.0417) (0.0480) First Stage Exogenous Export (t-1) 0.3953*** (93.27442) Exogenous Import (t-1) -0.3794*** (95.22348) Log Exp Part GDP/cap 0.204*** (0.0274) Log Imp Part GDP/cap 0.190*** (0.0382) Log air Freight (t-1) 0.097*** (0.0112) Log Mean Dist top 5 -0.543*** (0.092) First Stage F- test - 38.61*** 42.54*** Hansen J test p-value - 0.5545 0.1351 33 Observations 7,808 7,749 5,726 R-squared 0.842 0.753 0.726 Number of id 196 195 166 Country FE Yes Yes Yes Time FE Yes Yes Yes Note: Adj Trade is adjusted trade share known as trade intensity; Robust standard errors in parentheses. *** < 0.01, ** < 0.05, * < 0.1;. Table A6. Full Results — The impact of trade on GDP per capita (1) (2) (3) OLS IV1 IV2 VARIABLES Log GDP/Cap Log GDP/Cap Log GDP/Cap Log Adj Trade 0.227*** 0.536*** 0.756*** (0.0395) (0.0428) (0.0997) Invest (% GDP) 0.00322 -0.00265 -0.00503* (0.00253) (0.00165) (0.00258) Human Cap (%) 0.00358*** 0.00168*** 0.000567 (0.00121) (0.000577) (0.000859) Inflation (2010 = 100) 0.00165*** 0.00239*** 0.00311*** (0.000507) (0.000480) (0.000744) First Stage: Impact of Adjusted Trade on Growth Exogenous Export (t-1) 0.452*** (0.0614) Exogenous Import (t-1) -0.446*** (0.0625) Log Exp Part GDP/cap 0.009 (0.0200) Log Imp Part GDP/cap 0.088*** (0.0260) Log air Freight (t-1) 0.052*** (0.0096) Log Mean Dist top 5 -0.208** (0.1068) First Stage F- test - 36.98*** 11.53*** Hansen J test p-value - 0.6251 0.1640 Observations 4,566 4,553 3,722 R-squared 0.879 0.824 0.779 Number of id 166 163 145 Country FE Yes Yes Yes Time FE Yes Yes Yes Note: Adj Trade is adjusted trade share, known as trade intensity; Robust standard errors in parentheses. *** < 0.01, ** < 0.05, * < 0.1; Invest (% GDP) is private investment measured by gross fixed capital formation in percentage of GDP; Human capital (%) represents secondary school enrollment ratio; and Inflation (2010 = 100) is Consumer Price Index. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 34 Table A7. Weak identification test and Hansen J statistic (overidentification test of all instruments) A- Without control variables Trade -IV1 Test 1: Weak identification test Cragg-Donald Wald F statistic 47.781 10% 15% 20% 25% Stock-Yogo critical values 19.93 11.59 8.75 7.25 Test 2: Hansen J statistic (overidentification test of all instruments) Hansen (H0: valid instruments) P-value = 0.5545 Trade -IV2 Test 1: Weak identification test Cragg-Donald Wald F statistic 92.003 10% 15% 20% 25% Stock-Yogo critical values 10.27 13.96 10.26 8.31 Test 2: Hansen J statistic (overidentification test of all instruments) Hansen (H0: valid instruments) P-value = 0.1351 B- With additional control variables Trade -IV1 Test 1: Weak identification test Cragg-Donald Wald F statistic 41.044 10% 15% 20% 25% Stock-Yogo critical values 19.93 11.59 8.75 7.25 Test 2: Hansen J statistic (overidentification test of all instruments) Hansen (H0: valid instruments) P-value = 0.6251 Trade -IV2 Test 1: Weak identification test Cragg-Donald Wald F statistic 92.003 10% 15% 20% 25% Stock-Yogo critical values 10.27 13.96 10.26 8.31 Test 2: Hansen J statistic (overidentification test of all instruments) Hansen (H0: valid instruments) P-value = 0.1640 Source: Authors’ calculation based on IV regressions 35 Table A8. Levin-Lin-Chu (1992) unit-root test H0: Panels contain unit roots Ha: Panels are stationary Variable Obs Period Unadjusted t Adjusted t* p-value Main variables Log GDP per Capita (US$) 10600 52 -5.536 -6.703 0.000 Trade (% GDP) 9800 52 -16.704 -11.022 0.000 Exports (% GDP) 9800 52 -14.538 -9.358 0.000 Imports (% GDP) 9800 52 -15.966 -11.663 0.000 Adj Trade (% GDP) 9800 52 -17.438 -11.725 0.000 Adj Exports (% GDP) 9800 52 -15.101 -9.101 0.000 Adj Imports (% GDP) 9800 52 -6.056 -3.529 0.000 Control variables Inv 9100 52 -14.514 -10.158 0.000 Human Capital 8950 52 1.816 2.198 0.986 CPI (2010=100) 8500 52 0.831 8.571 1 Source: Authors’ calculation; A balanced panel is required for the LL-test. To compute the LL-test we relied on extrapolation to fill missing data. Table A9. Maddala and Wu (1999) unit-root test H0: All panels contain unit roots Ha: At least one panel is stationary Variable Obs Period Chi2 Freedom p-value Main variables Log GDP per Capita (US$) 9243 43 640.084 422 0.000 Trade (% GDP) 7826 40 648.563 392 0.000 Exports (% GDP) 7826 40 549.996 392 0.000 Imports (% GDP) 7827 40 719.420 392 0.000 Adj Trade 7808 40 567.774 392 0.000 Adj Exports 7808 40 569.761 392 0.000 Adj Imports 7809 40 771.230 392 0.000 Control variables Invest (% GDP) 7164 39 776.953 362 0.000 Human capital 6671 32 317.740 396 0.998 Inflation (2010=100) 7722 40 277.398 378 1.000 Source: Authors’ calculation based on the World Bank World Development Indicators (WDI) data. 36 Table A10. Results of the IV regression by income group-the impact of trade on GDP per capita (1) (2) (3) (4) (5) (6) IV1 IV2 VARIABLES Log Log Log Log Log Log GDP/Cap GDP/Cap GDP/Cap GDP/Cap GDP/Cap GDP/Cap Log Adj Trade [AEs = 1] 0.540*** 0.616*** (0.0385) (0.177) Log Adj Trade [EMEs = 1] 0.627*** 0.788*** (0.0667) (0.0893) Log Adj Trade [EMDEs = 1] 0.576*** 0.713*** (0.0675) (0.0718) Invest (% GDP) -0.00336* -0.00364** -0.00336* -0.00414** -0.00470** -0.00414** (0.00194) (0.00175) (0.00194) (0.00211) (0.00203) (0.00211) Human Cap (%) 0.00147** 0.00164*** 0.00147** 0.000816 0.00113 0.000816 (0.000664) (0.000590) (0.000664) (0.000747) (0.000758) (0.000747) Inflation (2010 = 100) 0.00248*** 0.00222*** 0.00248*** 0.00295*** 0.00236*** 0.00295*** (0.000494) (0.000443) (0.000494) (0.000681) (0.000612) (0.000681) Observations 4,530 4,530 4,530 3,724 3,724 3,724 R-squared 0.959 0.958 0.959 0.957 0.959 0.957 Country FE Yes Yes Yes Yes Yes Yes Time FE Yes Yes Yes Yes Yes Yes Note: Adj Trade is adjusted trade share; Invest (% GDP) is private investment measured by gross fixed capital formation in percentage of GDP; Human capital (%) represents secondary school enrollment ratio; and Inflation (2010 = 100) is Consumer Price Index. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. AEs is a dummy equal to 1 if the country is an advanced economy. EMEs is a dummy that takes 1 if the country is an Emerging Market. EMDEs is a dummy that takes 1 if the country is an Emerging Market or Developing Economy. Table A11. Results — The impact of trade on GDP per capita – Income group effects (1) (2) (3) OLS IV1 IV2 VARIABLES Log GDP/Cap Log GDP/Cap Log GDP/Cap Log Adj Trade 0.212*** 0.481*** 0.524*** (0.0337) (0.0429) (0.0637) Invest (% GDP) 0.00586*** 0.000255 0.00118 (0.00212) (0.00160) (0.00191) Human Cap (%) 0.00403*** 0.00261*** 0.00322*** (0.00122) (0.000539) (0.000639) Inflation (2010 = 100) 0.00225*** 0.00285*** 0.00321*** (0.000609) (0.000488) (0.000572) Observations 4,566 4,553 3,722 R-squared 0.891 0.849 0.863 Number of id 166 163 145 Country FE Yes Yes Yes Time FE Yes Yes Yes WBHICs × Year Yes Yes Yes WBUMICs × Year Yes Yes Yes WBLMICs × Year Yes Yes Yes WBLICs × Year Yes Yes Yes 37 Note: Adj Trade is adjusted trade share; Invest (% GDP) is private investment measured by gross fixed capital formation in percentage of GDP; Human capital (%) represents secondary school enrollment ratio; and Inflation (2010 = 100) is Consumer Price Index. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Table A12. Impact of trade share on GDP per capita countries exporting and importing more. (1) (2) (3) OLS IV1 IV2 VARIABLES Log GDP/Cap Log GDP/Cap Log GDP/Cap Log Adj Trade [Imp > Exp = Yes] 0.227*** 0.519*** 0.609*** (0.0401) (0.0582) (0.124) Log Adj Trade [Exp > Imp = Yes] 0.237*** 0.465*** 0.659*** (0.0406) (0.0572) (0.137) Invest (% GDP) 0.00294 -0.000796 -0.00384 (0.00263) (0.00361) (0.00480) Human Cap (%) 0.00362*** 0.00161 0.00153 (0.00123) (0.00161) (0.00199) Inflation (2010 = 100) 0.00164*** 0.00239** 0.00266* (0.000516) (0.00106) (0.00137) Observations 4,566 4,556 3,724 R-squared 0.974 0.963 0.964 Country FE Yes Yes Yes Time FE Yes Yes Yes Note: Adj Trade is adjusted trade share; Invest (% GDP) is private investment measured by gross fixed capital formation in percentage of GDP; Human capital (%) represents secondary school enrollment ratio; and Inflation (2010 = 100) is Consumer Price Index. A dummy D is included and takes the value of 1 if the country exports more. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Table A13. Impact of trade share on GDP per capita -oil exporters vs. non-oil exporters (1) (2) (3) OLS IV1 IV2 VARIABLES Log GDP/Cap Log GDP/Cap Log GDP/Cap Log Adj Trade [Non-Oil Exporter = 1] 0.274*** 0.538*** 0.705*** (0.0530) (0.0752) (0.219) Log Adj Trade [Oil Exporter = 1] 0.202*** 0.513*** 0.636*** (0.0441) (0.106) (0.219) Invest (% GDP) 0.00288 -0.00250 -0.00316 (0.00255) (0.00374) (0.00520) Human Cap (%) 0.00349*** 0.00175 0.00100 (0.00122) (0.00165) (0.00228) Inflation (2010 = 100) 0.00159*** 0.00233** 0.00271* (0.000499) (0.00106) (0.00150) Observations 4,566 4,556 3,810 R-squared 0.974 0.964 0.960 Country FE Yes Yes Yes Time FE Yes Yes Yes Note: Adj Trade is adjusted trade share; Invest (% GDP) is private investment measured by gross fixed capital formation in percentage of GDP; Human capital (%) represents secondary school enrollment ratio; and Inflation (2010 = 100) is Consumer Price Index. A dummy D is included and takes the value of 1 if the country exports oil. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. 38 Table A14. Impact of trade share on GDP per capita - controlling for institutional quality (1) (2) (3) OLS IV1 IV2 VARIABLES Log GDP/Cap Log lGDP/Cap Log GDP/Cap Log Adj Trade 0.245*** 0.520*** 0.708*** (0.0435) (0.0397) (0.0983) Invest (% GDP) 0.00365 -0.00230 -0.00538* (0.00283) (0.00182) (0.00277) Human Cap (%) 0.00419*** 0.00250*** 0.00112 (0.00131) (0.000571) (0.000809) Inflation (2010 = 100) 0.00231*** 0.00211*** 0.00164*** (0.000632) (0.000285) (0.000327) Polity2 -0.0123** -0.0127*** -0.00938*** (0.00473) (0.00180) (0.00230) Observations 3,942 3,933 3,325 R-squared 0.884 0.844 0.827 Number of id 148 146 133 Country FE Yes Yes Yes Time FE Yes Yes Yes Note: Adj Trade is adjusted trade share, known as trade intensity; Invest (% GDP) is private investment measured by gross fixed capital formation in percentage of GDP; Human capital (%) represents secondary school enrollment ratio; Inflation (2010 = 100) is Consumer Price Index. Robust standard errors in parentheses and Polity2 measures the quality of institutions. It provides a score ranging from -10 (autarchy) to 10 (Democracy). Robust standard errors in parentheses. *** < 0.01, ** < 0.05, * < 0.1. Table A15. Impact of trade share on GDP per capita -a dynamic model and IV approach. No instrument for log GDP/ Second order lag as IV for Cap−1 log GDP/Cap−1 IV1 IV2 IV1 IV2 VARIABLES Log Log Log Log GDP/Cap GDP/Cap GDP/Cap GDP/Cap Long Term log Adj Trade � �1 − � 0.505*** 0.490*** 0.518*** 0.444*** (0.0928) (0.1637) (0.1081) (0.2066) log GDP/Cap−1 () 0.790*** 0.821*** 0.832*** 0.861*** (0.0310) (0.0308) (0.0262) (0.0283) Log Adj Trade ( ) 0.106*** 0.0876** 0.0868*** 0.0619 (0.0323) (0.0431) (0.0289) (0.0405) Invest (% GDP) 0.00126* 0.00157* 0.00131* 0.00170** (0.000713) (0.000819) (0.000678) (0.000813) Human Cap (%) -0.000150 0.000106 -0.000226 8.00e-05 (0.000232) (0.000233) (0.000224) (0.000231) Inflation (2010 = 100) 0.000352 0.000139 0.000247 1.11e-05 (0.000276) (0.000312) (0.000273) (0.000317) Observations 4,451 3,645 4,522 3,709 R-squared 0.971 0.976 0.974 0.978 Number of id 163 145 163 145 Hansen J stat 0.189 8.91e-06 0.197 1.44e-05 39 Country FE Yes Yes Yes Yes Time FE Yes Yes Yes Yes Note: Adj Trade is adjusted trade share; Invest (% GDP) is private investment measured by gross fixed capital formation in percentage of GDP; Human capital (%) represents secondary school enrollment ratio; and Inflation (2010 = 100) is Consumer Price Index. Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The second order lagged dependent variable is used as an instrument for lagged GDP per capita. IV1 and IV2 remains remains the instruments used for Adjusted trade share. 40