Trade Openness and the Growth–Poverty Nexus: A Reappraisal with a New Openness Indicator

I. Introduction

The last few decades have witnessed a significant increase in the share of developing countries participating in global trade, coinciding with rapid economic growth and structural transformation, and concomitant widespread poverty reduction, most notably in the People’s Republic of China (PRC) and India. In this paper, I ask one main question: How does trade openness promote economic growth and impact poverty reduction?

This question is increasingly relevant today as developing countries face a less open and more conflicted international trade environment, highlighted by the trade war underway between the United States (US) and the PRC since 2018. Furthermore, large reductions in trade and output volumes occurred during the coronavirus disease pandemic (in part due to the set of policy responses adopted). These two factors have added momentum to the deglobalization trend. Since trade openness has long been praised as an engine of growth and employment generation, trade liberalization is currently at the forefront of development policy circles.

The standard trade theory postulates that opening up to trade reduces poverty in developing countries through both an indirect effect on economic growth (the “pull-up” effect) and by increasing the poverty impact of a given rate of growth through changes in the employment intensity of growth. The latter effect operates through changing factor proportions of production, which in a labor-abundant country means shifting resources toward labor-intensive production (the “factor proportion” effect).

Findings from the early comparative country studies of the developmental outcomes of trade-policy-regime shifts in developing countries from import substitution to export orientation are generally consistent with both postulates (Little, Scitovsky, and Scott 1970; Bhagwati 1978; Krueger 1978; Balassa 1982). A comparison of more recent evidence on the poverty reduction outcome of trade policy reform episodes in some developing countries is also consistent with these postulates.¹

The impacts of recent multicounty econometric studies on the growth–poverty nexus have provided evidence that only supports the pull-up effect (Roemer and Gugerty 1997; Dollar and Kraay 2002; Aisbett, Harrison, and Zwane 2008; Dollar, Kleineberg, and Kraay 2016).² The impact of trade openness on poverty, therefore, remains the subject of considerable controversy. For instance, Winters, McCulloch, and McKay (2004, p. 106) and Winters and Martuscelli (2014) conclude that “there can be no simple general conclusion about the relationship between trade liberalization and poverty.”

This paper is motivated by the concern that the failure of these previous studies to detect a systematic relationship between trade openness and poverty could be due to the limitations of the standard measure of trade openness: the trade-to-gross-domestic-product (GDP) ratio. This openness indicator, however, not only captures changes in trade policy but also other policy actions along with a variety of factors unrelated to trade openness such as economy size, population, technological change, evolving trade patterns, and income growth. I address this issue by first constructing a new index of trade openness that captures the convergence of the prices of tradable goods among countries, drawing on research by Jeffrey Williamson and other studies on the relative price movement of traded goods in the context of economic globalization. The key notion of this index, which I dub the price convergence index (PCI), is that, after allowing for transport costs, the degree of price convergence of tradable goods across countries over time is irrefutable evidence of greater global economic integration. I undertake empirical analysis using a new multieconomy panel dataset put together from various sources to examine the trade–growth–poverty nexus using the standard growth regression model. The analysis covers 123 economies from 1970 to 2017.

The results suggest that, when the new measure of trade openness is used, there is a statistically significant negative relationship between openness and poverty after controlling for economic growth and other relevant control variables. At the same time, the coefficient of the interaction term between growth and openness is negative and statistically significant. The poverty reduction impact of a given rate of economic growth is 0.3% larger in open economies. The findings are robust to alternative measures of poverty, the inclusion of a set of relevant explanatory variables, and estimation of the model for subsamples of economies and for different periods.

This paper is structured as follows. Section II sets out the conceptual framework of the poverty–growth nexus with emphasis on the role of trade openness. Section III explains the new measure of trade openness. In section IV, the model, estimation method, and data compilation are presented. Section V reports the results. Finally, section VI concludes.

II. Conceptual Framework

According to the standard trade theory, trade openness can reduce poverty directly through the factor proportion effect and indirectly through the growth effect.

The Stolper and Samuelson (1941) theorem postulates that liberalization in poor countries promotes labor-intensive production, resulting in an increase in the demand for unskilled labor. Since labor is the only resource owned by poor people, the creation of employment injects income to poor people, even if wages do not initially increase under “surplus labor” conditions at the early stages of economic growth (Lewis 1954). Put simply, employment creation is a surefire way to reduce poverty because labor is generally the only resource poor people possess. In addition, in the process of growth and structural transformation of the economy, wages begin to increase after the pool of surplus labor is fully absorbed in the modern sector. This process, facilitated by trade openness, further augments the poverty reduction effect of economic growth.

Trade can also affect poverty through the pull-up effect of growth. The basic notion is that openness contributes to growth and growth reduces poverty in a number of ways (Findlay 1984, Bhagwati 1988). Export earnings can relax balance-of-payment constraints. This allows the economy to access imported capital goods and machinery, and essential intermediate inputs, enabling an expansion of the manufacturing sector. Global market penetration enables domestic production to gain scale economies without being constrained by the size of the national economy. Moreover, openness increases the exposure of the domestic economy to international competition and diffuses international knowledge and foreign technology, resulting in higher productivity. Through these channels, growth is expected to trickle down to poor people for a given level of income distribution (Ahluwalia, Carter, and Chenery 1979; Deininger and Squire 1996).

The findings of a series of in-depth, comparative country studies are consistent with both postulates. A pioneering study of trade policy and industrialization in developing countries was conducted by Little, Scitovsky, and Scott (1970). The key message of this study, which sets the stage for the subsequent ideological shift from an import-substitution to an export-oriented industrialization strategy, was that redressing policy bias against exporting promotes greater efficiency in the use of resources and generates higher levels of employment, paving the way for later growth with an equitable distribution of income. Other subsequent comparative studies have arrived at similar inferences (Balassa 1982; Papageorgiou, Armeane, and Michaely 1990).

However, the results from empirical studies are rather mixed. Dollar and Kraay (2002) find that the share of income of poor people significantly increases with the rise of the average income of these countries, but the interaction of the openness measures with real GDP per capita is not statistically significant. Aisbett, Harrison, and Zwane (2008) find a positive but insignificant relationship between openness and poverty reduction using instrumental variable estimation. Despite continuing controversies over methodology and the measurement of openness, several studies reach a similar conclusion that openness does not affect the income of poor people beyond the effect on average per capita income growth (Roemer and Gugerty 1997, Dollar and Kraay 2002).

Instead of focusing on income and poverty at the aggregate (countrywide) level, several recent studies have used micro-level data to examine the openness–poverty nexus. For instance, Topalova (2010) investigates the relative effects of trade liberalization on poverty reduction in India at the district level using a difference-in-differences approach. The key result is that, in rural India, districts that were more exposed to trade liberalization have made slower progress in poverty reduction. Kis-Katos and Sparrow (2015) also examine the effects of trade liberalization in Indonesia. They find that districts with greater exposure to input tariff liberalization experience more rapid poverty reduction. This poverty-reducing impact of trade liberalization is also found in the context of Thailand’s accession to the World Trade Organization (WTO) (Durongkaveroj and Ryu 2019).

Overall, the evidence of the direct effect of openness on poverty is mixed. At a macro-level, a clear pattern emerges with the exception that there is no systematic impact of trade openness on the incomes of poor people beyond aggregate economic growth. At a micro-level, there seems to be a direct impact of liberalization on poor people. However, these studies do not pay attention to the possibility that their findings might have been conditioned by the well-known limitations of the trade-to-GDP ratio commonly used as the sole measure of trade openness.

The use of the trade-to-GDP ratio as an indicator of trade openness is highly debatable. Changes in the trade ratio can capture an increase in imports and/or exports driven by other factors such as economy size, geography, population, capital accumulation, technological advances, and a change in terms of trade, all of which have little to do with more liberal trade policies (O’Rourke and Williamson 2002, Berg and Krueger 2003, Williamson 2014). Dollar and Kraay (2004) argue that the change rather than the level of trade-to-GDP ratio is not contaminated by geography, nor by other unobserved country characteristics. However, this reasoning is valid only if all unobservable country characteristics remain unchanged over time (Birdsall and Hamoudi 2002). For instance, a change in the terms of trade can impact the trade share regardless of the openness of the trade regime. Moreover, since policymakers cannot control the level of trade driven by the ongoing process of global economic integration, trade share may have little to do with trade policy. In addition, there is strong empirical evidence that changes in the trade-to-GDP ratio are significantly driven by changes in GDP per se (Fujii 2019).

Finally, the trade share is susceptible to a country’s engagement in global production sharing, which is the cross-border dispersion of production processes within vertically integrated global industries (Athukorala 2014, Antràs 2016). This process involves spreading the total value addition of a given product among a number of countries. This implies that the value-added share of recorded exports from a given country tends to decline with the deepening of its involvement in global production networks. While GDP is measured in value-added terms, trade is measured in gross terms, thus resulting in inflated trade values relative to GDP. When the manufacturing sector of a country is well integrated within global production networks, the trade-to-GDP ratio can be artificially high even though export production involves adding small amounts of value to imported inputs (Krugman 1995).

Two other openness indicators, which are mostly used to supplement the standard trade-to-GDP ratio, are average applied tariff (simple average or trade weighted) and the Sachs–Warner (SW) index (Sachs and Warner 1995). There are a number of limitations in using tariff rates as an indicator of openness to trade. First, the official and effective tariff rates can be very different because imported inputs used in export production are duty-exempt in most countries (Pritchett and Sethi 1994). Second, in some cases, there are quantitative restrictions side-by-side with tariffs that naturally create an unexplained wedge between world market prices and the domestic prices of traded goods (Anderson and Neary 1992, Milner and Morrissey 1999).

The SW index is a binary indicator that helps distinguish between open and closed regimes. It has been designed to capture various policy measures impacting trade openness. In constructing this index, a country’s trade policy regime is treated as “open” based on five criteria: (i) an average tariff rate below 40%, (ii) nontariff barriers covering less than 40% of trade, (iii) a black-market exchange rate premium below 20% on average during the 1970s and 1980s, (iv) the absence of a socialist economic system, and (v) the absence of an extractive state monopoly in major export sectors (Sachs and Warner 1995). The SW index has also received several critiques. For example, Berg and Krueger (2003) assert that this dummy variable does not capture the different degrees of restrictiveness of trade regimes and, therefore, poses a limitation on the use of panel data analysis. In the same vein, Irwin (2019) points out that a dummy variable does not tell us about multiple periods of being open and closed.

III. A New Measure of Trade Openness

Mindful of the limitations of the traditional measure of trade openness, I constructed a new index to measure trade openness based on changes in the relative prices of traded goods.

The idea for constructing this index comes from the work of Jeffrey Williamson and his research associates (O’Rourke and Williamson 1999, 2002; Williamson 2000, 2002, 2014). As they point out, price convergence is a better indicator of openness than the trade-to-GDP ratio. It is important to note that most of these historical studies have focused on trade in primary products (e.g., sugar, spice, and coffee). However, as noted by O’Rourke and Williamson (1994, p. 899), the concept of price convergence generally applies to tradable manufactured goods, not just primary products. Rodrik (2002, p. 10) also gives credence to the case for using price convergence as a superior measure alternative to the standard trade-to-GDP ratio: “From an economic standpoint, what matters most is not the volume of trade as much as the degree of price convergence across national markers.”

The concept of convergence of prices of traded goods in the process of global economic integration is closely related to the law of one price (LOP), which postulates that, in the absence of transport costs and trade restrictions, each traded good is uniformly priced throughout the world by perfect commodity arbitrage (Isard 1977, p. 942). Despite mixed evidence, the key inference from the empirical literature is that the “relative” version of the LOP (changes in relative prices) holds even though its absolute version (absolute price difference) does not hold. As convincingly argued in these studies, if international markets are integrated, the rate of change in prices at home and abroad should converge. At a given point in time, prices of a given product can of course be different across countries due to differences in consumer purchasing power (which depends on the stage of economic advancement), transportation costs, and other fixed costs. However, over time, openness to trade should manifest in a convergence of changes in relative prices of traded goods. In other words, even though price levels are naturally different, the rates of change in prices are, on average, synchronized among countries (Engel and Rogers 2001; Cecchetti, Mark, and Sonora 2002; Hufbauer, Wada, and Warren 2002; Goldberg and Verboven 2005). Thus, an index that captures the convergence of prices of traded goods across countries is a superior measure of openness to trade compared to the standard trade-to-GDP ratio. It captures the impact of both tariff and nontariff restrictions and behind-the-border barriers impacting a country’s engagement in foreign trade. At the same time, unlike the trade-to-GDP ratio, this index is less susceptible to other nontrade-related factors, in particular, country size and participation in global production networks.

In the market integration literature, there has been an attempt to examine whether prices in different markets move together and if the price differential is driven by transfer costs (Baulch 1997, Keller and Shiue 2007). While these studies focus on testing the comovement of prices on which the LOP is based, the PCI measures the overall trade openness of a given economy by examining changes in its national price compared to the world price. Due to methodological choices and the available data, the results from testing commodity market integration using agricultural prices are rather mixed (Federico 2012).

In this study, I construct a PCI that captures changes over time in the price of traded goods in a given economy relative to that of the world price. To construct the index, manufacturing price is measured by the implicit price deflator (with 1970 as the base year) derived from the national accounts of individual economies, while treating the implicit price deflator for the US as the proxy indicator of the world price. Individual economy price indices are adjusted for changes in the exchange rate with the US dollar and then expressed as a ratio of the US price index to obtain the relative manufacturing price indices. The PCI is then constructed as the absolute deviation of relative price from the base value ($1970=100$). See the Appendix for details on how to construct this index.³

I use the manufacturing price index to measure the traded goods price due to the relatively high degree of tradability of manufactured goods. The GDP deflator is not appropriate because it captures both tradable and nontradable prices. Agricultural products are traded goods, but some agricultural products are quasi-nontradables (e.g., vegetables and some other food items). More importantly, agricultural prices are influenced by changes in global commodity price cycles. The US manufacturing price is taken as the reference price because the US is the largest trading country in the world with a highly open trade regime during the study period, particularly for manufactured goods.

Data for manufacturing value-added deflators for all economies other than the PRC were obtained from the Food and Agriculture Organization database.⁴ Data for the PRC were compiled from the data extracted from the World Bank’s World Development Indicators Database.⁵ Note that only data for industry (mining, construction, utilities, and manufacturing) are available for the PRC during the entire study period. However, a comparison done for a recent period (2000–2015) for which disaggregated data are available suggests that the manufacturing deflator closely follows the patterns of the deflator for industrial production.

Figures 1 and 2 depict the trade-to-GDP ratio and the PCI for four economies—the PRC, India, Indonesia, and the Republic of Korea—over the period 1970–2017. These four economies have experienced trade regime policy shifts during the period of study.

It is clear from Figure 1 that, regardless of differences in respective policy changes, the trade-to-GDP ratio increased for all four economies during the past few decades. This increasing trend did not reverse even during the 1997/98 Asian financial crisis. Declining trade shares after the global financial crisis were due to the slowdown in world trade, not changes in economies’ trade policies. Using the traditional measure of openness, before 2000, Indonesia and the Republic of Korea were relatively open compared to the PRC and India. After that, the Republic of Korea’s degree of openness exceeded that of the other three economies. However, there were more variations in relative prices, and some episodes of these movements are associated with policy changes.

Despite the liberalization reforms initiated in 1978, the PRC was still considered a “closed economic system” until the late 1990s (Wacziarg and Welch 2008). Following its accession to the WTO in 2001, the PRC established a relatively open trade regime, resulting in significant reductions in tariffs, the gradual elimination of quotas and licensing, and a commitment to international standards in the protection of intellectual property. The PRC’s trade-to-GDP ratio has increased gradually over time, with a sharp increase in the trade share since the early 2000s. However, the relative price movement shown in Figure 2 indicates that the PRC’s trade regime was relatively closed throughout the 1980s and 1990s. This is consistent with evidence that trading rights, import licensing, canalization, and exclusive import rights were liberalized only in the late 1990s (Panagariya 2019). After accession to the WTO, the PRC’s price movement gradually became more in line with that of the US.

India seems to share a similar trend with the PRC. India gradually opened its economy to trade and investment after 1991, which was followed by some minor liberalization efforts during the 1980s (Pursell 1992, Panagariya 2005). This is illustrated by a relatively high degree of openness during the 1980s as shown in Figure 2. However, average manufacturing price movement in the 1990s suggests that protection in India remained high. Bown and Tovar (2011) suggest that India offset the effect of reduced tariffs through the use of antidumping and safeguard protections, especially after the late 1990s. A slight increase in the relative price movement from 2000 to 2010 indicates that India was more open during this period. However, from 2011 to 2016, such price movement diverged from that of the US again. During this period, the government of Prime Minister Narendra Modi launched the “Make in India” program (in 2014), which was accompanied by targeted tariff protections and government subsidies to specific industries (Athukorala 2020). Overall, India’s economy remained less open when compared with the Republic of Korea and the PRC.

Indonesia became relatively open from about the early 1980s, with some episodes of protectionism (Fane and Condon 1996, Marks and Rahardja 2012). Yet, as shown in Figure 2, the relative price movement suggests that Indonesia experienced policy reversals. During the 1970s and the early 1980s, Indonesia followed some forms of import-substitution industrialization with the use of tariffs, export bans, and import licensing (Pangestu, Rahardja, and Ing 2015). From the late 1980s to the mid-1990s, when Indonesia implemented deregulation and export promotion, relative price movement was relatively stable. Price divergence took place again after the 1997/98 Asian financial crisis. However, Indonesia has seen the return of protectionism in recent years, especially in the form of nontariff barriers (Soesastro and Basri 2005, Basri and Patunru 2012, Patunru 2018). This can be clearly observed in the divergence of Indonesia’s relative price movement since 2000.

As shown by both the trade-to-GDP ratio and the PCI, the Republic of Korea remained relatively open throughout the period of study. Even though its trade-to-GDP ratio rose steadily, the relative price movement indicates that there were some fluctuations in its trade regime. During the 1960s, the expansion in labor-intensive exports contributed to rapid economic growth. Nonetheless, the Republic of Korea launched a targeted promotion of heavy and chemical industries in 1973, with such firms enjoying protection through high tariffs. Several incentives were also provided to firms in heavy and chemical industries such as directed bank credit at low (on average, negative) real interest rates and special tax treatment and trade policy concessions (Graham 2003, Adelman 2007). While the trade-to-GDP ratio during this period increased, the Republic of Korea’s average manufacturing price in the 1970s diverged from the world price. The relatively more liberal trade policy stance of the government in the 1980s is reflected in more convergence in price changes as the economy returned to a neutral regime (Panagariya 2019, p. 229). An example of liberalization effort can be seen in the establishment of the Tariff Reform Committee in 1983. The Republic of Korea’s trade policy regime has since remained relatively open, albeit with some divergences in price movement during the 1997/98 Asian financial crisis and the global financial crisis.

Table A1 in the Appendix reports the coefficient of variation of the PCI for all economies covered in this study. The coefficient of variation declined for almost all economies during the study period, illustrating the increasing integration of the world economy over time. Of course, the PCI is not a perfect indicator of economic openness. Given the enormous heterogeneity of manufacturing trade and other economy-specific fixed factors such as geographic distance and economy size, it is impossible to assume perfect convergence of manufacturing prices even in the absence of trade restrictions. Moreover, since the PCI is constructed from the manufacturing price deflator, this index captures the average price movement over time. Of course, the ideal choice would be a comparison of price movement for a product that is homogenous across all economies, but there is no suitable product for such a comparison. However, allowing for these complications, we can reasonably assume that trade restrictions play a role in the movement of relative prices of manufactured goods among economies.

IV. Methodology

A. The Model

This section uses the new measure of trade openness to examine how trade openness affects poverty. Following previous studies on the trade–growth nexus (Ravallion and Chen 1997; Dollar and Kraay 2004; Santos, Dabus, and Delbianco 2019), the empirical model is specified as follows :

where $\mathrm{\text{POV}}$ is the poverty headcount ratio and the subscripts i and t refer to economy and year, respectively. The explanatory variables are listed as follows, with the postulated sign of the regression coefficient for the explanatory variables in parentheses:

In equation (1), ${\beta }_1$ is commonly known as the growth elasticity of poverty. This elasticity expresses how much the poverty incidence declines in response to economic growth. The sign of ${\beta }_1$ is expected to be negative, meaning that, all things being equal, an increase in GDP per capita should reduce the incidence of poverty. Also, the sign of ${\beta }_2$ is expected to be negative. This implies that an increase in the degree of openness should result in less poverty. However, the estimated coefficient may be indifferent from zero if the effect of openness is manifested through growth—that is, an indirect effect of openness on poverty. The model is first estimated with the trade-to-GDP ratio to measure openness as the benchmark and then with the PCI.

The rate of inflation (INF) is included to capture macroeconomic stability. It is important to explain inter-economy differences in the poverty rate since real wages among poor people tend to be smaller in economies with higher inflation. The sign of the coefficient on inflation is anticipated to be positive. Total government expenditure (GE) is included to control for the effect of government programs. As these programs should directly benefit the poor, the expected sign of this coefficient is negative. Finally, political and economic institutions are important to overall economic performance because they can foster conditions that are conducive to competitive markets. The sign of the coefficient on regime repressiveness (RR) is likely to be positive, implying that the higher the repressiveness, the higher the poverty rate. This is because in an economy where political rights and civil liberties are low, the economic incentives and conditions tend to favor only a small group of people (i.e., the rich and elite) instead of the poor. The interests of economically disadvantaged groups may not be prioritized in the decision-making process. However, it is also possible for a nondemocratic regime to implement policies to alleviate poverty. The expected sign of the coefficient is, therefore, ambiguous.

In addition, I investigate whether the relationship between poverty and growth is conditioned by trade openness by introducing an interaction of the openness measure with the log-level of real GDP per capita. The estimating equation is as follows:

The impact of economic growth on poverty is given by the partial derivative of P in equation (2) with respect to Y, ${\beta }_1+{\beta }_3{\mathrm{\text{OPEN}}}_{it}$. To test this hypothesis, the statistical significance of ${\beta }_3$ is examined. If the sign of ${\beta }_3$ is negative and statistically significant, it indicates that the impact of growth on poverty reduction is greater among relatively open economies. Moreover, ${\beta }_2$ captures the direct impact of openness on poverty, while ${\beta }_1$ captures the indirect impact since its impact operates through growth. The total impact of openness on poverty is, therefore, ${\beta }_2+{\beta }_3$, conditional on economic growth. Note that the sign of ${\beta }_2$ itself is ambiguous. This coefficient can be positive or negative depending on the nature of trade openness. Even if ${\beta }_2$ is positive, it does not imply that openness increases poverty. Whether its contribution is positive or not depends on the size of the coefficient of the interaction term of openness and growth. The expected signs of other explanatory variables are identical to equation (1).

B. Data

The model is estimated for a sample of 123 economies, and separately for all developing economies and developing economies in Asia only. The data have been compiled from various sources listed in Table A2 in the Appendix.

Developing economies are defined based on the United Nations Standard Country Classification. This group includes four economies that achieved high-income status in the 1990s based on the World Bank’s income-based classification (Hong Kong, China; the Republic of Korea; Singapore; and Taipei,China). The experience of these economies reaching high-income status is central to the debate on the openness–poverty nexus.

One of the difficulties in studying poverty across economies over a long period of time is that data on poverty (and income distribution) are scant, resulting in a highly unbalanced and irregularly spaced panel of observations. Data on poverty before 2000 cover only a few developing economies (e.g., Argentina, Indonesia, and Thailand). A common way to handle this data issue is to create a “spell” (interval) defined by the periods of time spanning two successive data series, which is based on household surveys. I follow closely the methodology used in the literature on poverty (Adams 2004, Dollar and Kraay 2004, Loayza and Raddatz 2010). To create a spell for each economy, I begin with the first available observation (poverty rate) and then move forward in time until the next observation on poverty exists. Since we are focusing on growth over the medium to long run, the condition needed to create each spell is that the length of each spell is at least 5 years. The last interval moves until the final observation of poverty incidence. This means that I drop economies that have only one observation of poverty (e.g., Japan, Tuvalu, and Zimbabwe). Also, the adjacent annual observations are dropped.

For example, in the case of Thailand, the first change in the poverty rate is between 1981 and 1988. This spell is followed by the change between 1988 and 1992. The poverty rate in 1990 is disregarded because the gap with the previous observation is less than 5 years. This applies to other explanatory variables over the same period. Since the length of the spells differ, I then annualize the changes in poverty and other explanatory variables in order to make spells of different lengths comparable. Table A3 in the Appendix reports economy coverage. Table 1 presents the summary statistics of the variables. The inverse of PCI is used in the econometric analysis to make it comparable with the trade-to-GDP ratio.

C. Estimation Method

Following the previous studies (Ravallion and Chen 1997, Dollar and Kraay 2004), the model is estimated using the fixed effects estimator. This estimator takes into account time-invariant economy characteristics that can influence both poverty and growth (i.e., institution, geography, and colonial history). An important concern is the possibility that the variance of the errors is not constant across observations. I address this issue by using heteroscedasticity-robust standard errors clustered at the economy level, allowing for errors to be correlated across spells within economies. I cluster the standard error by economy because observations of a given economy (spells) are more likely to be correlated within each spell over time, not across economies.

Another potential problem relates to the omitted variable bias. One of the crucial factors that affect both poverty and growth is, of course, institutional quality. The role of government, especially in poor economies, should not be neglected. Yet, data on institutions have only become available in recent years and are time-variant. In this study, I use several variables from different sources to capture those effects in order to minimize the omitted variable bias. In addition, the use of a conceptually superior measure of openness should hopefully reduce the measurement error of trade openness.

V. Results

The results are presented in Tables A4 and A5 in the Appendix. Table A4 reports the results of equations (1) and (2) with the trade-to-GDP ratio as the measure of trade openness. Table A5 presents the regression results with the PCI. In each table, the results are presented for the total sample, and separately for developed and developing economies.

As shown in Figure 3 and Table A4, the coefficient on economic growth, commonly known as the growth elasticity of poverty, is negative and statistically significant at the 1% level in all specifications. The point estimates for the growth elasticity of poverty for the full sample range between −1.85 and −2.06. Thus, a 10% increase in growth is associated with approximately a 20% reduction in the proportion of people living in poverty (below $1.90 per person per day). I obtain a lower growth elasticity of poverty for the sample of developing economies, as reported in columns 4–9. The estimates for the sample of developing economies range from −1.33 to −1.65.

The coefficient on trade-to-GDP ratio is not statistically significant even at the 10% level, indicating that there is no direct impact of trade openness on poverty. The coefficient on the interaction term between growth and the trade-to-GDP ratio is also not statistically different from zero. The results hold for all three economy subsamples. Thus, the findings are consistent with other studies that the effect of trade openness on poverty operates solely through economic growth (Roemer and Gugerty 1997; Dollar and Kraay 2002, 2004).

To comment on the results based on the PCI, the growth elasticity of poverty ranges between −1.84 and 2.00 (Table A5, columns 1–3). This indicates that the poverty rate declines by around 20% with a 10% increase in economic growth. The results withstand the inclusion of inflation, government spending, and regime repressiveness. The results also hold when high-income economies are excluded from the sample (columns 4–9). There is no evidence of heteroscedasticity in terms of the Breusch-Pagan test and the Cook Weisberg and White’s Generate test. This suggests that the error variances are all equal. In addition, Ramsey’s RESET Test is employed to test general functional form misspecification. The results suggest the absence of this problem.

As shown in Figure 4 and Table A5, the coefficient on the interaction between growth and the PCI is negative and statistically significant. The result indicates that, for a given rate of economic growth, an increase in this openness index by 10 percentage points is associated with the further reduction of poverty by 2.9%. Thus, the findings provide strong support for the theoretical postulate that the impact of economic growth on poverty is enhanced by trade openness. Also, the coefficient of the PCI is negative and statistically significant at the 1% level. This suggests that an increase of 10 percentage points in this index is associated with a decrease in the poverty rate of 0.11%, all things being equal. Therefore, there is a direct impact of openness on poverty reduction, even after controlling for the growth effect.

Robustness Checks

To check the robustness of the results, I estimated two alternative specifications of the model. I only report alternative estimates with the PCI as the trade openness variable. The coefficients on the trade-to-GDP ratio and its interaction term with economic growth are not statistically significant in any specification.

The relationship between poverty and economic growth can be sensitive to the poverty line used to measure the poverty rate because different poverty lines detect changes in different segments of the distribution of incomes (Ravallion 2016, Fosu 2017). To address this concern, the model is reestimated using the poverty rate calculated based on a poverty line of $3.20 per day and the respective national poverty line. According to Table A6 in the Appendix, the results are largely consistent with previous findings, although the coefficient on growth is slightly smaller (columns 1–3). The results also hold when poverty is measured using the national poverty lines of each individual economy (columns 4–6).

The model estimated for data averaged over 5-year periods and using the fixed effects estimator are reported in Table A7 in the Appendix. The results are largely consistent with previous findings. The results suggest that a 10% increase in real GDP per capita is associated with approximately a 20% decrease in the poverty rate. The poverty-reducing impact of economic growth is larger for economies with a more open trade regime.

VI. Conclusion

Over the past few decades, developing countries have increasingly engaged in the world economy. This phenomenon has been accompanied by rapid economic growth, structural transformation, and poverty reduction. This study examines the relationship between economic growth, poverty reduction, and trade openness by using a new measure of trade openness, dubbed the PCI, and a multieconomy panel dataset from various sources covering 123 economies from 1970 to 2017.

The results based on the traditional measure of trade openness are consistent with the results of previous studies. However, when the new measure of trade openness (i.e., the PCI) is used, it is found that there is a systematic relationship between openness and the incidence of poverty. In addition, the results suggest that the relationship between growth and poverty is conditioned by the degree of trade openness—that is, the poverty-reducing impact of a given rate of growth is greater for economies with more open trade regimes. This finding provides support to the prediction of the factor proportion theory of international trade.

The new openness measure developed in this study is conceptually preferable to the traditional trade-to-GDP ratio, as it better tracks trade policy regime shifts within economies over time. However, it has its own limitations dictated by the availability of data. The results from this study call for further attempts to develop better indicators of trade openness in order to broaden our understanding of the poverty impacts of openness in this era of economic globalization. Further research could also extend the analysis by focusing on case studies of individual countries to supplement existing multicountry econometric studies.

ORCID

Wannaphong Durongkaveroj  https://orcid.org/0000-0002-1792-8177

Notes

¹ The most conspicuous examples are the PRC and India. Both countries have experienced rapid growth and poverty reduction in recent decades. However, the poverty rate has fallen much faster in the PRC compared to India. There is ample evidence that the PRC is more open to foreign trade and investment compared to India (Ghosh 2010, Ravallion 2011, Bhagwati and Panagariya 2013, Joshi 2017, Panagariya 2019).

² For a survey on this literature, see Bhagwati and Srinivasan (2002); Winters, McCulloch, and McKay (2004); Winters and Martuscelli (2014); and Panagariya (2019).

³ To view all appendixes, please refer to the supplemental materials that are available at: https://www.worldscientific.com/doi/suppl/10.1142/S0116110524400092.

⁴ Food and Agriculture Organization. FAOSTAT. Prices Data. https://www.fao.org/faostat/en/#data (accessed 10 March 2020).

⁵ World Bank. World Development Indicators. https://databank.worldbank.org/source/world-development-indicators (accessed 10 March 2020).

Appendix

To view all appendixes, please refer to the supplemental materials that are available at: https://www.worldscientific.com/doi/suppl/10.1142/S0116110524400092.