The estimation process is conducted in several steps. Specifically, we estimate Eq. 2 using different estimators to account for cross-sectional dependence and endogeneity for the panel of 82 countries from 1960 to 2017. The natural start of the estimation process is to estimate Eq. 2 through ordinary least squares (OLS). This regression controls for country-fixed effects and time-fixed effects. The results from such a regression serve as baseline results. Then, we estimate the regression through CCEMG to address the issue of cross-sectional dependence and through GMM to deal with endogeneity. After this, CCEMG-GMM is estimated to address cross-sectional dependence and endogeneity jointly. In addition, several diagnostic tests are performed to test the statistical health of the regressions.
The findings for the full sample are presented in Table 2. The first three columns (i.e., columns 1, 2, and 3) of Table 2 depict the results of the pooled ordinary least square (POLS) estimator. The results are based on a priori expectations. However, we do not discuss the result of the POLS due to the bias inherited from endogeneity and cross-sectional dependence. The lower panel of columns 4, 5, and 6 of Table 2 portray the cross-sectional dependence (CD) statistics and p-values. Based on the p-values, the null hypothesis is rejected regarding the cross-sectional independence among the panels. This implies that POLS estimates provide biased results. Therefore, we use CCEMG as an alternative. The upper panel of columns 4, 5, and 6 of Table 2 shows the CCMEG estimates.
The first CCMEG regression, which regresses trade, k, fd, cpi, edu, and gov on log difference of per capita GDP, is estimated. Denoted by trade (the natural log of trade-to-GDP ratio) is the trade dependency ratio (column 4, Table 2). This is the sum of the exports and imports of goods and services measured as a share of GDP. The value of , which is the coefficient of trade openness, is positive and significant. This implies that trade volumes enter significantly positively in the growth regression.
The economic implication of this finding is that an increase in international trade in terms of trade volumes may lead to an increase in the economic growth of a country. This finding is in line with, for example, Chang et al. (2009), Rao et al. (2011), Cooray et al. (2014), Cooray et al. (2017), and Zahonogo (2017). However, the findings presented in other studies (e.g., O’Rourke, 2000; Gries et al., 2009) are not consistent with our findings. However, a country’s trade dependency ratio and trade volume do not reflect its political stance. It is postulated that several economic developments (e.g., exchange rate dynamics, technological regress) and other macroeconomic fluctuations may be reflected in trade volumes. Therefore, we look at other proxies for trade openness that indicate a country’s stance in terms of trade policies. Furthermore, as mentioned earlier, the impacts of trade liberalization and trade policies are more controversial than the influence of trade volume in terms of economic impact.
We think that tariff rates are a direct indicator of trade openness and serve as an accurate policy-related variable. However, different studies on the trade-growth nexus define and measure tariff rates in different ways. Two noteworthy studies are mentioned presently. First, effective tariff rate (i.e., the ratio of tariff revenue to total import revenue) is used by Kanbur and Zhang (2005), while average tariff rate is considered by Dobson and Ramlogan (2009). Ma and Dei (2009) note that these measures of tariff rates could have different impacts on the economic outcomes of a country. Therefore, it is worthwhile to use multiple measures of tariff rates in an analysis. Doing so will provide more accurate information regarding the impact of trade liberalization on economic growth than using one measure alone.
Thus, we use multiple measures of tariff rates but do not see any ambiguous findings in the cross-country analysis. Therefore, we present our results in terms of average tariff rates to make comparisons easier between our results and those of notable previous studies (see columns 2, 5, 8, and 11, Table 2).
The average tariff rate enters significantly negatively in the growth regression. This trend implies that the trade policy of reducing restrictions has a positive impact on economic growth. The statistically significant negative sign implies that the economic growth of developing countries could be enhanced by reducing the average tariff rate. Specifically, economic growth could increase by 0.8010 percent due to a 1-percent decrease in tariff rates. Our results corroborate the findings of Wacziarg and Welch (2008), Kneller et al. (2008), and Gries et al. (2009).
The use of tariffs as a proxy of trade openness has been criticized on various grounds despite having the appropriate trade policy dimensions. For example, some aggregation biases exist, especially in the case of average tariff rates. Furthermore, there are significant gaps between statutory tariff rates and the collected tariffs, especially in developing countries. In addition, Dollar and Kraay (2004) note that there is little correlation between the measures of trade volume and tariff rates. Therefore, trade volumes and trade policy measures might yield misleading results in terms of economic growth.
Keeping these problems in view, we incorporate a third proxy measure of trade openness, (i.e., the globalization index of Dreher et al. (2008)). This index gives proper weights to trade volumes and tariff rates, as well as to other flow and restriction measures of trade openness. Therefore, it should be considered a more comprehensive measure than trade volume and average tariff rate. The information provided in columns 3, 6, 9, and 12 of Table 2 shows that the measure of globalization enters significantly positively in the growth regression. This confirms our finding that trade openness has a positive impact on economic growth in the cross-country analysis.
As mentioned earlier, there are other variables besides trade openness that may affect economic growth. Such variables are incorporated in our study so that the relationship between trade openness and economic growth will not be spurious. For this purpose, we select a few variables (e.g., capital series, financial development, inflation, human capital, and governance), as has been done in most of the literature on endogenous growth. The construction of these variables is mentioned in the appendix.
Table 2 shows that physical capital has a positive impact on economic growth for all three specifications. This outcome is in line with a large portion of the literature that states that investment activities and physical capital stocks have a positive impact on economic growth. Human capital and physical capital are of great importance in the discussion of endogenous growth theories. We also incorporate a measure of human capital, which enters significantly positively in the growth regression. This finding supports the most highly accepted endogenous growth theory.
Recently, financial sector development has received special attention in two dimensions. First, it has been discovered that a well-functioning financial system could enhance the economic growth of countries (Jalil et al., 2010) and that trade-growth could flourish via various channels of financial sector development (Baltagi et al., 2009). Therefore, a measure of financial sector development must be incorporated into a growth regression. By including such a measure, we show that financial sector development has a positive impact on economic growth in the cross-country analysis. Inflation is another variable that should be discussed in the context of the trade-growth nexus. The findings of the current paper suggest that inflation hurts economic growth. However, these parameters are statistically insignificant in almost all cases.
Second, the quality of institutions is an important explanatory variable discussed in the recent literature. Acemoglu et al. (2001), among others, provide evidence that the quality of institutions explains the differences among countries in terms of economic growth. However, there is no consensus in the literature as to how the quality of institutions should be measured. For example, it can be measured through law and order, the quality of formal institutions, the corruption and accountability of public officials, and many other factors. Therefore, we develop an index (denoted GOV) by using a principal component analysis (see the appendix). The regression result is in line with those of distinguished studies. That is, it is found that the quality of institutions has a positive impact on economic growth. However, we contradict the findings of Rodrik et al. (2004) and Rigobon and Rodrik (2005), who state that trade openness enters negatively in the growth regression when it is restricted by institutions and geography.
Further, we distribute the sample of countries into five subsamples (see Appendix A). We replicate the exercise on each of these subsamples with alternative measures of trade openness. That is, we estimate CCEMG and GMM for each of the five subsamples. Interestingly, trade openness does not alter the sign in these subsamples. We address two methodological issues – cross-sectional dependence and endogeneity. These can be addressed individually using a CCMEG estimator and a GMM estimator. So far, we are convinced that the positive impact of trade openness on GDP growth is not derived from cross-sectional dependence or endogeneity. However, jointly addressing these issues could be enlightening.
For this purpose, following Chudik and Pesaran (2015) and Neal (2015), we use CCEMG-GMM by using the lags of endogenous variables. The estimation results are shown in Table 2 (for the full sample) and Table 3 (for the region-wise analysis). The signs and significance levels are not changed considerably by combining the CCEMG and GMM. This implies that our main results are not affected by endogeneity or cross-sectional dependence. As such, empirical studies that have shown that trade openness has either a negative impact or no impact on economic activities should be reconsidered.
We have estimated the impact of trade openness on economic growth through four estimators. The estimates are similar in terms of their economic interpretations (i.e., the signs of the coefficients). However, interestingly, the sizes of the coefficients of the CCEMG-GMM estimations are much smaller than those of any other estimator. The coefficients range from 0.2634 to 0.3071 for CCEMG-GMM. Therefore, we suspect that the CCEMG-GMM estimates might be affected by endogeneity and cross-sectional dependence.
Interestingly, the findings derived from the CCEMG-GMM are very close to the findings of Lewer and Berg (2003), who suggest that a 1-percent improvement in trade openness results in an increase in economic growth of around 0.2 percent.
Our estimated models pass various diagnostic tests (see the lower panels of Tables 2 and 3). We restrict the estimation by up to two lags. We perform the Sargan test and autocorrelation test to verify the validity of the instruments used in our study. The p-values of the tests show that the instruments are valid, and the p-values of Hansen and difference-in-Hansen tests signify the appropriateness of the instruments. Specifically, the p-values of significantly less than one imply that there is no issue of instrument proliferation.