A further investigation of the link between trade and income

Abstract

The link between openness and income has received increasing attention as countries try to justify their trade-promoting policies. Recent work of Frankel & Romer (1999) examines the effect of trade on income. We explore how the estimates of the trade effect change when we relax their assumption of heteroscedasticity in the bilateral trade equation they use to construct the instrument for the IV regression. Because the instrument is constructed through a nonlinear transformation, unequal disturbance variances imply inconsistency and not just inefficiency of the Frankel–Romer estimates. We find a smaller positive effect of trade than that found by Frankel & Romer.

Keywords:

• Trade
• income
• heteroscedasticity
• instrumental variable

Acknowledgments

The authors would like to thank David Romer and Jeffrey Frankel for allowing access to their data. All responsibility for errors remains with the authors.

Notes

¹See Levine & Renelt (1992); Rodriguez & Rodrik (2000).

²Exports create an outlet for excess production and generate income. Imports, especially the imports of capital and intermediate goods, improve the quality and productivity of the domestically produced goods and raise exports further. The causality may also move in the reverse direction, however, since higher domestic output encourages the production of exportable products as well as the consumption of foreign-made products, thereby raising trade volume.

³The measure of openness is controversial in the broader literature, as summarized by Rodriguez & Rodrik (2000) and by Hallack & Levinsohn (2004). Wacziarg (2001) places openness measures into three categories: outcome-based measures (volume of trade or its components), policy-based measures (tariff rates, non-tariff barriers, or tariff revenues), and deviation measures (deviations of observed trade volume from the predicted free-trade volume). While some studies use the trade-to-GDP ratio to represent trade, others create a trade policy variable as a proxy for trade openness. Average tariff rates, defined as the ratio of import duties to the import volume, are found to be negatively related to growth by Lee (1993), Harrison (1996), and Edwards (1998). Edwards (1998) also uses non-tariff barriers as a measure of trade restrictions and finds an insignificant relationship with growth. For a limited number of countries, Anderson & Neary (1992) construct a trade restrictiveness index that incorporates the effects of both tariffs and non-tariff barriers. One of the criticisms of this approach is that trade policies are multi-faceted. They are highly correlated with other macroeconomic policies. Rather than improving economic performance, trade policies reflect the lobbying efforts of special interest groups and alter the distribution of income. Existing studies have not controlled for policies related to growth (macro stability, small government consumption, and rule of law) when they use trade policy variables.

⁴Per capita income not only captures economic growth and national performance, most countries experiencing an increase in per capita income witness rapid technological improvement as well (Verdoorn's Law). We are also aware of the limitation of this income measure. Per capita income may not adequately capture economic well-being because it does not take into account advances in such factors as life expectancy, literacy, general health and inequality.

⁵Even though the great-circle distance, which does not vary over time, may well mis-estimate the true transportation costs, we use it as a proxy for distance to facilitate the comparison of the Frankel–Romer model and the revised model.

⁶The Gulf Cooperation Council is a Persian Gulf-based economic and political policy-coordinating forum with members including Kuwait, Saudi Arabia, Bahrain, Qatar, the United Arab Emirates, and Oman.

⁷Alternative definitions of the small country dummy variable with working populations less than 300,000 or 900,000 do not change the main results of this paper.

⁸ Figure 3 plots actual trade share against constructed trade share without controlling for the size of the country. Figure 4 shows the relationship between actual trade share and constructed trade share after controlling for population and area of countries.

⁹In Frankel & Romer (1999), the authors choose to adjust the standard errors of their estimates for the variation in the parameter estimates of the bilateral trade equation. As a result of this correction, the standard error of the Column 2 trade share in Frankel & Romer (1999) is 0.99 instead of 0.91 and 1.49 instead of 1.34 for the Column 5 trade share. The adjustment increases standard errors by approximately 10% but does not change the results of significance tests at standard significant levels. The adjustments we compute for the IV regressions below with a heteroscedastic first-stage are substantially less than 10% (see Zhang, 2003). In this paper, the standard errors that are presented have not been adjusted for variation in bilateral trade estimates.

¹⁰Disturbances in the second-stage per capita income equation are likely to be heteroscedastic as well. We take this into account by applying the Huber–White sandwich estimator of variance in place of the traditional variance calculation (see Huber, 1967; White, 1982).

¹¹To see the difference between oil-exporting countries and other countries, Figure 5 provides a scatter plot of the per capita GDP against the degree of openness for the 150 countries. Observations marked with a triangle ‘Δ’ are the oil-producing countries. Two conclusions can be drawn from Figure 5. First, the correlation between trade share and per capita income is positive for both types of countries. Second, at a given level of openness, oil-exporting countries have higher per capita income compared with the non-oil producing countries. The effect of trade on income is stronger for the oil-exporting countries.

¹²The coefficients for East European variables are insignificant and these variables are omitted from the final regressions.