2. Theory model

Our theory framework is an adaptation of the Heckscher–Ohlin (HO) model. The HO model provides the foundation for a voluminous body of research on international trade. The HO theory says that countries produce and export goods that are intensive in their relatively abundant endowments, and import goods that are intensive in their relatively scarce endowments. The theory is grounded on the law of comparative advantage, which states that countries export goods that have lower relative costs and therefore prices, in autarky. In other words, country abundance or scarcity of endowments determines costs, which determines trade.

The Hechscher–Ohlin–Vanek (HOV) variation holds that countries export their relatively abundant endowments ‘embodied’ in their net exports to the world. For example, if a country were relatively abundant in land, the country would be a net exporter of goods that were produced using land intensively. Conceptually, the land endowment is embodied in the good since the endowment is immobile across countries, while the goods produced (using land) are mobile.Footnote ²

We begin with the HOV trade equation derived in the classic work of Leamer (Reference Leamer1984):

(1)$$(\underline{X} - \underline{M} ){\rm} = A^{-1}(\underline{V} - s\underline{V}_w)$$

where $(\underline{X} - \underline{M})$ is a n×1 vector of country net exports (exports minus imports) to the world; V is an m×1 column vector of country endowments; V _w is an m×1 column vector of world endowments; s is a scalar that equals a country's output share of world output (Y/Y _w); A ^–1 is an n×m inverted technology matrix of factor input requirements; and m and n are the number of endowments and industries, respectively.

Equation (1) says that a country's net exports in each of the n industries depends on the country's relative abundance of m endowments. A country is abundant in relative terms if the country has more of a given endowment j than its share of the world endowment (i.e., V _j– sV _jw). This relative abundance confers a comparative advantage to the country in industries that are intensive in the endowment.Footnote ³

The parameters comprising the inverted technology matrix (A ^–1) are Rybczynski coefficients.Footnote ⁴ These coefficients link a country's endowments to her net exports (and outputs) in each industry. They take positive (negative) values on endowments that provide comparative advantage (disadvantage). For example, a _ij is the coefficient corresponding with the ith industry and jth endowment of A ^–1. If a _ij > 0, then the jth endowment confers a comparative advantage in industry i; and if a _ij < 0 then the jth endowment confers a comparative disadvantage in industry i.

We adapt the HOV expression in several ways. First, we express equation (1) in industry form as

(2)$$\left( {X_i - M_i} \right){\rm} = A^i(\underline{V} - {\rm} s\underline{V} _w)$$

where (X _i– M_i) is the ith row of column vector (X– M), and Aⁱ is the ith row of matrix A ^–1.

Second, we divide all variables in equation (2) by country output (Y). This gives

(3)$$\left( {X_i - {\rm} M_i} \right){\rm} /Y = A^i(\underline{V} /Y - \underline{V} _w/Y_w)$$

where the term ‘s’ cancels out because s/Y = (Y/Y _w) / Y = 1/Y _w.

Third, we collect all terms that are constant across countries. In equation (3), V _w and Y _w are world aggregates and do not vary by country. The elements of the row vector Aⁱ are also constant across countries. Collecting these terms gives

(4)$$\left( {X_i - {\rm} M_i} \right){\rm} /Y = A^i(\underline{V} /Y){\rm} + C_i$$

where C _i = –A ⁱ(V_w/Y _w).

Fourth, we rewrite equation (4) using lower case letters to represent variables scaled by Y. This gives

(5)$$\left({x_i - m_i} \right) = C_i + A^{i} (\underline{v})$$

Fifth, we define endowments in the familiar way to include country aggregates of capital (K), labor (L), and land (T). Thus, the column vector (v) includes scaled expressions of these endowments (denoted k, l and t). Equation (5) can then be rewritten as

(6)$$\left({x_{i} - m_{i}} \right) = C_{i} + {a_{i1}}\; {k} + a_{i2} \;{l} + {a_{i3}} \;{t}$$

As the theory suggests, net exports are defined at the industry level, while endowments are aggregates. That is, a country's industry level trade depends on their aggregate endowments (not factor inputs). The Rybczynski coefficients (a _i1, a _i2, and a _i3) that link endowments to trade vary across industries. Positive (negative) coefficients correspond with endowments that confer comparative advantage (disadvantage) in the industry.

3. Empirical method and data

Our baseline statistical specification of theory equation (6) is

(7)$$\left( {x_i - m_i} \right) = \alpha _{i0} + {\alpha_{i1}}\; k + {\alpha_{i2}} \;l + {\alpha_{i3}} \;t + \varepsilon _i$$

where α _i0 = C _i; α _i1 = a _i1, α _i2 = a _i2, α _i3 = a _i3; and ε _i is a randomly distributed disturbance term. Equation (7) shows the relationship between aggregate endowments and industry level trade for one country at one point in time. Below we proceed by estimating equation (7) for a cross-section of countries j and/or for country cross-sections pooled across time t. For our baseline, we estimate equation (7) using country cross-sections for 1995 and 2010 for soybeans, maize, and cotton trade. This allows us to examine the sources of comparative advantage (across countries) in the industries at the two points in time representing pre- and post- GMOs.

Next, we extend the baseline in three ways. The first adaptation involves disaggregating the land endowment (t) into GMO and non-GMO land (t _g and t _n)

(8)$$\left( {x_i - m_i} \right) = {\alpha _{i0}} + {\alpha _{i1}} \;k + {\alpha _{i2}} \;l + {\alpha _{i3}} \;t_g + {\alpha _{i4}} \;{t_n} + \varepsilon _i$$

where the aggregate endowment equals the sum of its components (t = t _g + t _n); and α _i3 and α _i4 are decomposed Rybczynski coefficients. Recall, Rybczynski coefficients are derived from the inverted matrix of factor input requirements and the technologies that comprise these coefficients can be distinct. We expect α _i3 to take a positive sign indicating that a country's net exports in GI industries are intensive in GMO land. We expect α _i4 < α _i3 indicating that net exports in GI industries are less intensive in non-GMO land.Footnote ⁵ (Recall, GI industries include GMO and non-GMO content, and proportions vary by country.) We estimate equation (8) for 2010 to examine the trade effects of land allocation decisions across countries at one point in time.

Our second adaptation extends equation (8) to explore cross-time changes. In this variant, we pool the data across 1995 and 2010. We then define the land terms to reflect country endowments in the pre- and post-GMO periods as follows:

(9)$$\left( {x_i - m_i} \right) = {\rm \alpha}_{i0} + {\alpha _{i1}} \;k + {\alpha _{i2}} \;{l} + {\alpha_{i3}} \;t^{\prime}_g + {\alpha _{i4}} \;t^{\prime}_n + \varepsilon _i$$

where GMO land is zero (t′ _g = 0) for all countries in 1995 since this year precedes the advent of GMOs; and GMO land is greater than or equal to zero (t′ _g ⩾ 0) in 2010, depending on the country's land allocation; and t′ _n = t – t′ _g. In equation (9), the parameters have the same interpretations as in equation (8). However, by pooling the data we can now undertake a cross-country analysis of differences between two equilibrium points in time. We expect α _i3 to be statistically different from α _i4 if the reallocation of land from non-GMOs to GMOs affects comparative advantage between 1995 and 2010.

Our third adaptation augments equation (8) to include policy instruments as follows:

(10)$$\left( {x_i - m_i} \right) = \alpha _{i0} + {\alpha _{i1}} \;k + {\alpha _{i2}} \;l + {\alpha _{i3}} \;t + {\alpha _{i4}}\; {\rm PR} + \varepsilon _i$$

(11)$$\eqalign{\left( {x_i - m_i} \right) =& \;\alpha _{i0} + {\alpha _{i1}}\; k + {\alpha _{i2}}\; l + {\alpha _{i3}} \;t \cr & + \alpha _{i4} \;{\rm AP} + \alpha _{i5} \;{\rm RA} + \alpha _{i6} \;{\rm LP} + \alpha _{i7} \;{\rm TR} \cr &+ \alpha _{i8} \;{\rm CG} + \alpha _{i9} \;{\rm IA} + \varepsilon _i}$$

where PR is the policy regime: and AP (approval process), RA (risk assessment), LP (labeling policies), TR (traceability requirements), CG (coexistence guidelines), and IA (membership of international agreements) are the component policies. We estimate equations (10) and (11) for each industry in 2010. Recall, the HO model is a supply-side model predicting a country's net exports. Policies can be trade augmenting or reducing. Strong policies can increase supply-side costs and thus decrease net exports. In this case, we expect a negative parameter of the policy. Alternatively, strong policies can decrease supply-side costs and thus increase net exports. In this case, we expect a positive parameter of the policy. Thus, equations (10) and (11) allow us to empirically determine how policies affect comparative advantage.

Appendix A summarizes the expected signs of the estimates of equations (7)–(11).Footnote ⁶ Our econometric method involves estimating these equations using either cross-section data or pooled data for a large sample of countries in 1995 and/or 2010. We begin using OLS to generate the core results reported in Section 4. We then employ alternative approaches to assess the sensitivity of our findings in Section 5. Alternatives include: quantile regression methods to control for outliers such as early adopters of GMOs; two-stage least squares using instrumental variables to explore potential endogeneity of policies with respect to trade; fixed effects to assess the robustness of results with respect to cross-time factors; and the introduction of interaction terms to allow for factor-augmenting technologies.

The data include measures of trade, endowments, and policies. Appendix B details the data and sources and Appendix C reports select data.Footnote ⁷ Three measures are worthy of elaboration here. First, we measure trade as the dollar value of each country's net exports of soybeans, maize, and cotton. Recall, these trade data do not distinguish between GMO and non-GMO content because such data are not available. However, the industries we consider have the largest shares of GMO content. As such, our objective is to analyze trade in GMO intensive industries, while allowing countries to differ in their GMO vs. non-GMO makeup.

Second, we measure land endowments as the total hectares of arable land of each country. We then decompose this endowment into GMO and non-GMO components. We measure GMO land as hectares of biotech land area using data developed by James (Reference James2011). This land is designated as GMO (i.e., transgenic) when plants have a novel combination of genetic material obtained using biotechnology. Such crops can contain genes that have been artificially inserted, rather than being acquired through pollination (James Reference James2014). We construct the measure of non-GMO land as the difference between total land and GMO land. These measures are not industry specific because the HO model calls for endowments rather than factor inputs.

Third, we measure policies using the indexes of Vigani et al. (Reference Vigani, Raimondi and Olper2012). These data cover 60 countries and are based on GMO-related laws in 2007 or before, in the policy areas detailed in equation (11).Footnote ⁸ The composite index (representing the policy regime in equation (10)) is the sum of scores in each category normalized to the range [0,1]. Higher scores correspond with stronger policies. For example, countries with a composite index of one are ‘GMO free’ and countries with an index of zero have few regulations.

5. Sensitivity analysis

5.3 Technology differences

Next, we consider the robustness of findings with respect to technology differences across countries. The HO model assumes that technologies are identical across countries. However in our application, GMO and non-GMO technologies presumably differ, and countries differ in adoption of GMOs and thus differ in the technologies they tend to employ. The implication of technology differences within the HO model was first addressed in the seminal work of Trefler (Reference Trefler1993, Reference Trefler1995). Trefler (Reference Trefler1995) demonstrates that trade departs from HOV predictions due to ‘factor-augmenting’ technologies.

We consider the role of such technologies in the spirit of Trefler (Reference Trefler1995). We derive a modified HOV expression where technologies can differ across GMO and non-GMO countries (see Appendix G). The new specifications of equations (10) and (11) are:

(12)$$\eqalign{\left( {x_i - m_i} \right) = &\;\beta _{i0} + {\beta _{i1}} \;G + {\beta _{i2}} \;k + {\beta _{i3}} \;l + {\beta _{i4}} \;t \cr & + {\beta _{i5}} \;{G}\,\cdot \,k + {\beta _{i6}} \;G\,\cdot \,l + {\beta _{i7}} \;G\,\cdot \,t + {\beta _{i8}} \;{\rm PR} + \varepsilon _i}$$

(13)$$\eqalign{\left({x_i - m_i} \right) = &\;{\beta _{i0}} + {\beta _{i1}} \;G + {\beta _{i2}} \;k + {\beta _{i3}} \;l + {\beta _{i4}} \;t \cr & + {\beta _{i5}} \;G \cdot k + {\beta _{i6}} \;G \cdot l + {\beta _{i7}} \;G \cdot t \cr & + {\beta _{i8}} \;{\rm AP} + {\beta _{i9}} \;{\rm RA} + {\beta _{i10}} \;{\rm LP} + {\beta _{i11}} \;{\rm TR} + {\beta _{i12}} \;{\rm CG} + {\beta _{i13}} \;{\rm IA} + \varepsilon _i}$$

where G is a dummy variable that equals 1 if the country adopts GMOs and 0 otherwise; ε _i is a randomly distributed error term; and all other variables are as previously defined.

In equations (12) and (13), the coefficients are effective Rybczynski coefficients (ERCs) defined as the product of the standard Rybczynski coefficients (α) and factor-augmenting technologies (π). In other words, β _i = α _iπ. We expect ERCs to vary across GMO and non-GMO countries if factor-augmenting technologies differ across these groups. The interaction terms in equations (12) and (13) allow us to examine differences in ERC's for GMO countries relative to non-GMO countries. Our a priori expectation is that ERCs will be larger in absolute value for GMO countries if GMO technologies are factor augmenting.

Appendix H reports cross-country estimates of equations (12) and (13) for 2010. Findings show patterns consistent with Table 3, where land confers a comparative advantage and labor confers a comparative disadvantage. However, we now see that land confers a particularly strong comparative advantage in GMO countries (for soybeans and cotton) and labor confers a particularly strong comparative disadvantage in non-GMO countries (for maize and cotton). This suggests that GMO technologies are land and labor augmenting. Further, we see that capital is a source of comparative disadvantage for non-GMO countries (in soybean and maize), but is a source of comparative advantage for GMO countries (in maize). This suggests that GMO technologies are capital augmenting and strong enough to reverse net trade (for maize). This result differs from our previous finding that capital has a negligible trade effect. The new results are more consistent with the literature, which shows GMO crops to be capital intensive.Footnote ²⁵

Finally, Appendix H shows that policies still have little effect on comparative advantage, after controlling for endowments and technologies. One exception is that traceability requirements have a positive trade effect for cotton, consistent with the results in Table 3. The second exception is that coexistence guidelines have a positive trade effect for maize, a result not shown in Table 3. This later finding likely reflects the influence of countries such as Argentina and France who are exceptionally large net exporters of maize and have strong coexistence guidelines (see Appendix C).Footnote ²⁶ However, all of the remaining estimates on policy instruments are insignificant. Thus, allowing for factor-augmenting technologies does not alter our previous conclusion that these policies have a negligible supply-side effect on trade.

It is plausible that the policy results are weak because the critical policy decision is whether to allow GMO technologies in production or not, and this ‘adoption decision’ may overshadow the regulation of GMOs once in the market. Further, it is plausible that the (zero-one) technology adoption decision is related to the (continuous) land allocation decision, since countries differ in adoption of GMOs and consequently differ in the technologies they tend to employ. Thus, our finding that technologies play a role in augmenting land productivity likely reflects the effects of the technology adoption decision (a type of policy).

To consider this issue further, we first examine correlations between the GMO dummy (i.e., adoption decision), GMO land (i.e., allocation decision), and GMO policies (i.e., policy instruments). We find these correlations are modest and take the expected signs.Footnote ²⁷ The strongest correlation is between GMO land and the GMO dummy (0.34), suggesting a positive relationship between the adoption and land allocation decisions. We then re-estimate the regressions where land is disaggregated into its GMO and non-GMO components (i.e., Table 1), but now include the GMO dummy to control for the adoption decision. Appendix I reports the findings. As shown, the core results are insensitive to the inclusion of the dummy. That is, the land allocation decision has strong explanatory power, even after controlling for the GMO adoption decision. One possible explanation is that land allocations are based (in part) on matching GMO traits with exogenously determined land and weather attributes. For example, adoption of salt or drought tolerant crops addresses specific environmental features. These exogenous features likely determine the magnitude of land allocations following an adoption decision.

In summary, we conclude that the policy dimension that matters most is the decision of whether to adopt GMOs or not, rather than the regulation of GMOs once in the market. We also conclude that there is a modest correlation between GMO adoption and land allocations, suggesting these decisions are to some extent interrelated.

6. Concluding remarks

This paper provides empirical evidence on the supply-side determinants of trade in the GMO-intensive industries of soybeans, maize, and cotton during the time periods before and after the advent of GMOs. Five conclusions emerge. First, we found that land endowments have a positive supply-side effect on trade, and labor endowments have a negative trade effect. Second, we found that the allocation of land to GMO crops has a positive effect on trade. For example, a one million hectare allocation of land to GMO crops results in a $537.79 million increase in net soybean exports (on average across countries). The economic significance of this effect is particularly large for emerging GMO countries. Third, we found that land is a source of comparative advantage and labor is a source of comparative disadvantage both before and after the advent of GMOs. We also found that the reallocation of land to GMOs substantially increased the net exports of countries between 1995 and 2010. In other words, our first two conclusions hold both across countries at one point in time, and also across time from 1995 to 2010. Fourth, we found that the GMO policy regime and individual policies of countries appear to have a negligible effect on trade (on average across countries).

These finding are robust to a variety of considerations. First, using a counter-factual approach, we examined whether our findings would reverse for a non-GMO industry such as wheat. As expected, we found that non-GMO land serves as a determinant of trade for wheat, while GMO land does not. Second, using a lagged explanatory variable approach, we examined the potential endogeneity of countries’ land allocation decisions with respect to their trade. We found no evidence of this form of endogeneity. To the contrary, we found that past land allocations serve as a strong determinant of future comparative advantage. Third, using quantile regression methods, we analyzed the influence of outlier countries (such as early adopters of GMOs). We found that after accounting for outliers, land reallocations to GMOs continue to have a significant effect on comparative advantage. Fourth, we considered the potential endogeneity of policies with respect to trade. Using a lagged variable approach and two IV approaches, we found scant evidence of policy endogeneity. Fifth, using a political economy approach, we examined whether policies influence land allocation decisions between GMOs and non-GMOs. We found no evidence that policies affect land allocations (on average across countries). Finally, we examined the robustness of results when we allow for technology differences across GMO and non-GMO countries. We found that our earlier results are robust; however, we also found that GMO technologies are factor augmenting in some cases. For example, we found that GMO technologies tend to be land and labor augmenting; and are also capital augmenting and strong enough to reverse net trade (for maize).

Our findings point to a puzzle: Why do land allocations affect comparative advantage, while policies have a negligible effect on land allocations and on comparative advantage? One explanation is that the global policy setting in 2010 was evolving with considerable uncertainty. As a by-product, there may be a substantial lag in the effects of policies on land allocations and net exports. Yet another explanation is that policy stringency can both increase and decrease supply-side costs (and thus have a negligible net effect). Weak GMO policies may reduce supply-side costs because policy-compliance costs are lower. At the same time, weak GMO policies may increase supply-side costs because of the costs associated with creating transparency around product characteristics are higher. For example, producers may incur costs associated with voluntary labeling even in the absence of mandatory labeling regulations. Further, it is plausible that GMO policies effect supply-side costs for both GMO and non-GMO producers in the same direction. For example, strong coexistence guidelines could increase supply-side costs for both GMO and non-GMO producers. Finally, it is plausible that the policy results are weak because the critical policy decision is whether to allow GMO technologies or not, and this decision overshadows the regulation of GMOs once in the market. These issues point to the need for further research on the interactions of policies and the cost effects of alternative combinations. Such research may become feasible as measures of GMO policies and GMO trade are further developed and refined.