GENETICALLY MODIFIED CROPS AND PRODUCT

DIFFERENTIATION: TRADE AND WELFARE EFFECTS

IN THE SOYBEAN COMPLEX

ANDREI SOBOLEVSKY, GIANCARLO MOSCHINI, AND HARVEY LAPAN

A partial equilibrium four-region world trade model for the soybean complex is developed in whichRoundup Ready (RR) products are weakly inferior substitutes to conventional ones, RR seeds arepriced at a premium, and costly segregation is necessary to separate conventional and biotech products.Solution of the calibrated model illustrates how incomplete adoption of RR technology arises inequilibrium. The United States, Argentina, Brazil, and the Rest of the World (ROW) all gain fromthe introduction of RR soybeans, although some groups may lose. The impacts of RR production orimport bans by the ROW or Brazil are analyzed. U.S. price support helps U.S. farmers, despite hurtingthe United States and has the potential to improve world efficiency.

Key words: biotechnology, differentiated demand, identity preservation, international trade, soybeans.

Biotechnology innovations in agriculture rep-resent a recent trend that is providing bothdazzling opportunities and unexpected chal-lenges. Genetically modified (GM) crops ac-count for a major share of U.S. cultivation ofsoybeans, maize, and cotton, and a few coun-tries, notably Argentina, Canada, and China,have followed the United States’ lead. It is es-timated that, in 2002, GM crops accounted for145 million acres worldwide (James). But fromthe beginning, the biotechnology revolutionin agriculture has been controversial (Nelson;Moschini; Pardey). Notably, consumer groupsand the public at large have raised, espe-cially in Europe, a vociferous opposition tothe introduction of GM products in the foodsystem. They have expressed concern aboutthe safety of GM food and about the envi-ronmental impact of GM crops, among otherthings, and have demanded that consumers begiven the “right to know” whether the foodthey buy contains GM products.1 As a re-sult, a number of countries are implementing

Andrei Sobolevsky (Ph.D.) is a finance specialist at Sprint, Over-land Park, Kansas; GianCarlo Moschini is professor and PioneerHi-Bred Endowed Chair in Science and Technology Policy, De-partment of Economics, Iowa State University; Harvey Lapanis University Professor, Department of Economics, Iowa StateUniversity.

The support of the U.S. Department of Agriculture, through aone-year cooperative agreement and through a National ResearchInitiative grant, is gratefully acknowledged.

1 A survey of a representative sample of 16,000 citizens of theEuropean Union confirms the existence of a potentially sizeablecustomer base with differentiated preferences (Eurobarometer).

mandatory labeling regulations aimed at pro-viding exactly that choice. Such national poli-cies necessarily interfere with trade (Sheldon).Indeed, the international trade implicationsof widely differing adoption of, and policyresponse to, GM crops are proving increas-ingly difficult to accommodate in an ever moreglobalized world characterized by commodity-based agricultural trade. Frustration with thecurrent situation is underscored by the offi-cial complaint to the World Trade Organi-zation that the United States (supported byseveral other countries) launched against theEuropean Union (EU) in May 2003.2 Suchunresolved issues call for a deeper economicanalysis of some trade-related consequencesof GM crop adoption, and this paper is an ef-fort in exactly that direction.

The GM crops that have been most suc-cessful embody a single-gene transformationthat makes the crop resistant either to a her-bicide (e.g., Roundup Ready [RR] soybeans)or to a particular pest (e.g., Bt maize). Assuch, these improved crops represent a typicalprocess innovation, increasing the efficiencyof production but not supplying any new at-tribute that consumers value per se. In fact,because some consumers object to GM food,

2 The complaint alleges not only that the EU policy is closingthat market for potential U.S. exports, but also that it affects thepolicies of other countries toward GM crop adoption because ofthe countries’ concerns over future market access.

Amer. J. Agr. Econ. 87(3) (August 2005): 621–644Copyright 2005 American Agricultural Economics Association

622 August 2005 Amer. J. Agr. Econ.

the introduction of GM crops actually is bring-ing to market a product that some considerinferior in quality to its traditional counter-part. This “induced” product differentiationis one of the hallmarks of the current mar-ket impact of GM products and has a num-ber of economic implications that need to beaddressed. In particular, costly identity preser-vation activities are necessary to ensure thatGM and non-GM products are segregatedalong the production, marketing, processing,and distribution chain of the food system(Bullock and Desquilbet). Some models re-cently have attempted to incorporate differen-tiated final product demands and supply-sideidentity preservation. Whereas these modelsvary in their approaches and the issues theyaddress, they share the common attribute ofbeing specified at a very aggregate level andof not modeling closely enough the character-istics of the innovation being analyzed (e.g.,Nielsen and Anderson; Nielsen, Thierfelder,and Robinson; Lence and Hayes). In partic-ular, the GM crops that we are interestedin have been developed by the private sec-tor and are protected by intellectual propertyrights (IPRs), which give innovators a limitedmonopoly power that affects the pricing ofGM seeds for farmers and cannot be ignoredin assessing the welfare effect of innovations(Moschini and Lapan). Studies that overcomesome of these limitations (Moschini, Lapan,and Sobolevsky; Falck-Zepeda, Traxler, andNelson) still do not address the issue of inducedproduct differentiation.

A few recent studies have addressed theimplication of product differentiation andidentity preservation. Desquilbet and Bullockstudy potential adoption of GM rapeseedwith non-GM market segregation in Europebased on a calibrated two-country model inwhich differentiated market supply and de-mand functions are built up from the individualagent level. Lapan and Moschini (2001, 2004)also build a two-country partial equilibriummodel of an agricultural industry to analyzesome implications of the introduction of GMproducts. They model the GM crop as “weaklyinferior” in quality to the non-GM one in theimporting country that does not produce theGM crop and consider that country’s ability toimpose policies that limit its exposure to GMproducts. Fulton and Giannakas model GMproduct introduction in a closed economy ina vertical product differentiation framework,with emphasis on the welfare impacts of al-ternative labeling regimes. Furtan, Gray, and

Holzman study the impact of the potential in-troduction of GM wheat, with emphasis on theirreversibility aspects of such GM crop release.Whereas these studies take the analysis in adesired direction, the treatment is mostly the-oretical and the need remains for quantitativeestimates concerning the impact of GM inno-vations, especially in an open-economy setting.

In this study, we present a model that closelyrepresents the product differentiation that isinduced by the GM crop innovation and explic-itly models the ensuing need for costly iden-tity preservation activities that are requiredto supply the preinnovation (non-GM) prod-uct. In particular, we show how the induceddifferentiated demand can be specified con-sistently so as to allow welfare analysis. Themodel is applied to the world market for soy-beans and soybean products (soybean oil andmeal).3 Specifically, we develop a four-regionworld trade model in which GM crops areproduced in a market with differentiated de-mands and segregation costs. Two of the prod-ucts (soybeans and soybean oil) are modeledas existing in two varieties: conventional andGM, the latter variety being produced us-ing herbicide-resistant RR technology.4 Thefour regions of the model are the UnitedStates, Argentina, Brazil, and the Rest of theWorld (hereafter ROW). The United Statesis the world’s largest soybean producer andexporter. The other main producing region isSouth America, where most of the cultivationtakes place in Argentina and Brazil (table 1).These two countries took different paths withrespect to adopting RR soybeans because ofdifferent government policies.5 Hence, consid-ering these two regions separately will allowsome interesting policy simulation analyses.6

3 Soybeans have been the most successful GM crop to date. In2002, GM soybeans accounted for 62% of the world area cultivatedto GM crops, and GM soybeans accounted for more than half ofworld soybean production (James).

4 Because meal is used as feed, and animals so fed need not carryany GM label, there appears to be no reason why demand for mealshould be differentiated.

5 Although Argentina was an early adopter and has the world’shighest adoption rates of RR soybeans, the use of GM soybeansin Brazil has not yet been permanently authorized. But followingwidespread growing of GM soybeans smuggled in from Argentina,Brazil has enacted a temporary and limited authorization for GMsoybeans. With the decree enacted in September 2003, farmers whohave GM soybean seeds are allowed to plant them for the 2003–4crop year, and market the harvest through the 2004 year. But seedcompanies are not yet allowed to legally sell RR soybean seeds inBrazil.

6 To represent South America in our two regions, in our modelthe Brazil region includes both Brazil and Paraguay, whereasthe Argentina region includes the residual South Americanproduction.

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 623

Table 1. Soybean Complex Production and Utilization, 1998–99 (mil. mt)

Soybeans Oil Meal

Area Net Direct Net Net(mil. Ha) Product Exports Use Crush Product Exports Product Exports

World 71.16 161.67 n.a. 23.58 135.70 24.56 n.a. 108.36 n.a.

United States 28.51 74.60 21.82 5.47 43.26 8.20 1.04 34.29 6.37South America 22.93 55.34 12.89 2.43 40.29 7.55 3.78 32.19 22.01

Argentina 8.17 20.00 2.70 0.66 16.80 3.16 3.08 13.69 13.22Brazil 12.90 31.30 8.27 1.52 21.60 4.04 1.22 17.01 9.98Paraguay 1.20 3.00 2.30 0.05 0.65 0.12 0.09 0.51 0.41

Rest of the World 19.72 31.73 −34.71 15.68 52.15 8.81 −4.82 41.88 −28.38European Union 0.52 1.53 −16.07 1.53 16.23 2.92 1.06 12.92 −14.91China 8.50 15.16 −3.66 7.32 12.61 2.05 −0.87 10.03 −1.39Japan 0.11 0.15 −4.81 1.28 3.70 – – – –Mexico 0.09 0.14 −3.76 0.03 3.95 – – – –Mid-East/North Africa – – – – – 0.26 −1.64 1.23 −3.70

Source: U.S. Department of Agriculture (2002).

Although, in principle, demands in allfour regions could be differentiated, for thepurpose of the analysis, only ROW is mod-eled with differentiated demands. This cap-tures the observation that most resistance toGM product innovation to date has mate-rialized overseas and provides the opportu-nity for our model to consider how differ-ing GM regulation and policies across coun-tries affect market performance. The modelallows for costly identity preservation, an en-dogenous adoption rate of the GM technol-ogy, and noncompetitively supplied GM seedby an innovator-monopolist residing in theUnited States. The model is calibrated suchthat, when solved under both spatial and ver-tical equilibrium conditions, it replicates ob-served data in a benchmark year. The model isthen solved under a number of scenarios thatallow us to study and quantify the effects ofGM crop adoption under the induced prod-uct differentiation hypothesis. In addition tostudying the impact of the cost of keeping tra-ditional and GM crop products segregated, inorder to meet the induced differentiated de-mand, questions addressed by this paper in-clude the direction of price changes and tradeflows in GM and non-GM markets, the effi-ciency gains from the GM crop innovation,and the distribution of welfare effects acrossregions and across agents (consumers, produc-ers, and the innovator-monopolist). Also, themodel is used to study the impact of policiesdirectly aimed at GM crops, such as GM pro-duction and/or import bans in regions (suchas the EU) with differentiated consumer de-mand, and GM production bans in exportingregions (such as Brazil) that want to preserve

privileged access to demand for the traditionalproduct in importing countries. Finally, the re-strictions on parameter values used at the cal-ibration stage also are investigated through anextensive sensitivity analysis.

The Model

The soybean complex consists of three closelyrelated products: soybeans, soybean oil, andsoybean meal. Soybeans primarily are crushedto extract soybean oil and meal, which areactively traded internationally. The structuralmodel that we develop requires the specifica-tion of demand and supply functions for thethree products (possibly available in two va-rieties, GM and conventional) in each of thefour regions, as well as equilibrium conditions(market clearing for every product in every re-gion, spatial equilibrium across regions, andvertical equilibrium across segments of thesoybean complex).

Demand

The demand side of the model requiresspecifying two separate demands in the postin-novation period in at least one region, forconventional and GM varieties. Also, themodel must allow for the preinnovation de-mand with only the conventional variety andfor the postinnovation demand with only the(de facto) GM variety when no segregationtechnology is available. In addition, all thesedemands should arise from the same prefer-ence ordering, if welfare calculations are tobe meaningful. Furthermore, as emphasized inLapan and Moschini (2004), in our setting it is

624 August 2005 Amer. J. Agr. Econ.

important that demands satisfy the propertythat the GM product is a weakly inferior (notjust an imperfect) substitute for the traditionalone. The presumption here is that the GM soy-bean product does not have any additional at-tribute from the consumers’ point of view suchthat, ceteris paribus, all consumers will weaklyprefer the non-GM product. But whereas someconsumers may be willing to pay strictly posi-tive amounts to avoid the GM product, otherconsumers may be willing to pay very littleor may be indifferent between the two prod-ucts. Thus, the GM product will never com-mand a price that exceeds that of the non-GMproduct.

To implement the notion of weakly infe-rior substitutes, as in Fulton and Giannakas, agood starting point is the vertical product dif-ferentiation model with unit demand of Mussaand Rosen (see also Tirole, chap. 7), wherebyone can postulate a population of consumerswith heterogeneous preferences concerningGM and non-GM goods. Specifically, to gen-eralize this framework to the case where con-sumers choose both the type of good and thequantity to consume, let preferences for con-sumers of type � be represented by the quasi-linear utility function,

U = u(q0 + �q1) + y(1)

where u(·) is increasing and strictly concave,q0 and q1 denote physical consumption by theconsumer of the non-GM and GM product, re-spectively, and y denotes the consumption of anumeraire good. The condition limx→0 u′(x) =∞ ensures that the consumer will buy eitherthe non-GM or the GM variety, and it is fur-ther assumed that income is sufficiently high sothat an interior solution holds. The parameter� ∈ [0, 1] reflects the “weak inferiority” of theGM variety.

Given this structure, the demand by a con-sumer of type � depends upon the relativeprices of each variety. In particular, a consumerof type � will buy the GM variety if and onlyif p1 ≤ �p0.7 Thus, from equation (1) the indi-vidual demand curves can be written as:

q0 = d(p0) and q1 = 0 for � < �

q0 = 0 and q1 = 1�

d(p1/�) for � ≥ �

7 The consumer is actually indifferent between the two varietiesif an equality holds, but in such a case we may as well assume thatthe new variety is purchased.

where the individual demand function d(·) sat-isfies d−1(·) = u′(·), and � ≡ min{(p1/p0), 1}.Aggregate market demand functions can thenbe defined as:

Q0(p0, p1) =∫ �

0d(p0) dF(�)

Q1(p0, p1) =∫ 1

�

1�

d(p1/�) dF(�)

where F(�) denotes the distribution functionof consumer types.

To make this framework operational, oneneeds to specify the utility function u(·) andthe distribution function of consumer typesF(�).8 Note that the model-relevant proper-ties that arise from the above specification areas follows: (i) ∂Q0(p0, p1)/∂p1 = ∂Q1(p0, p1)/∂p0 ≥ 0 (i.e., goods are substitutes), and (ii)Q1(p0, p1) = 0 if p1 > p0 (i.e., the GM productis viewed as weakly inferior substitute for thetraditional one). Because exact aggregation ispossible with quasilinear preferences, we canalternatively think of Q0(p0, p1) and Q1(p0, p1)as arising from the choices of a representa-tive consumer who demands both varieties,while maintaining the property that the GMproduct is a weakly inferior substitute for thetraditional one. Following this approach, weadopt a linear specification for Q0(p0, p1) andQ1(p0, p1) and ensure the weak inferiorityproperty by carefully defining the domain ofthe functions. Specifically, the demand func-tions for conventional and RR products arewritten as:

Q0 = a0 − b0 p0 + cp1

Q1 = a1 + cp0 − b1 p1

}if p0 > p1(2)

Q0 ∈ [a0 − (b0 − c)p,

(a0 + a1) − (b0 + b1 − 2c)p]

Q1 ∈ [0, a1 − (b1 − c)p]

if p0 = p1 ≡ p

(3)

Q0 = (a0 + a1) − (b0 + b1 − 2c)p0

Q1 = 0

}

if p0 < p1

(4)

where all parameters are strictly positive.

8 For example, Fulton and Giannakas assume that consumersare restricted to buy either one unit or zero unit of the goods inquestion, that the utility function is linear, and that consumer typesare uniformly distributed.

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 625

Several observations are in order. First, thesymmetry condition is maintained, such thatthis demand system is integrable into well-defined (quasi-linear) preferences, a conditionthat will become important when making wel-fare evaluations. Next, the total demand thatis implied by this structure is

QT = (a0 + a1) − (b0 − c)p0 − (b1 − c)p1.

(5)

Given the condition b0 > c and b1 > c,9 whichwe shall assume, total demand is decreasing ineither price. Also note that, at p0 = p1, equa-tion (2) gives Q1 = a1 − (b1 − c)p0 (in the do-main p0 ≤ a1 /(b1 − c)). Thus, this specificationis consistent with a positive mass of consumersbeing perfectly indifferent between good 0 andgood 1 at p1 = p0 (but with p0 < p1, demandfor Q1 vanishes).

The specification in equation (4), which ap-plies in the domain p0 < p1, also representsmarket demand before the introduction of RRproducts. That is, the before-innovation situa-tion is equivalent to the new technology be-ing available but only at a prohibitive price,that is, above p0. When the new technology isadopted, but the RR and conventional vari-eties are not separated in the supply chain, theeffective demand for the conventional prod-uct is assumed to be zero. In other words,the assumption is that consumers treat comin-gled product as equivalent to GM product.10

Hence, this case is equivalent to the situationwhere the price of the GM-free product is pro-hibitively high, that is, above the “choke” pricep0 ≡ (a0 + cp1)/b0. Therefore, the postinno-vation demand without identity preservationis written as:

Q0 = 0

Q1 = a1 + ca0b0

−(

b1 − c2

b0

)p1

}

if p0 ≥ p0.

(6)

Note that the conditions b0 > c and b1 > censure that this demand is downward sloping.The complete specification of the demand sys-tem (2)–(4) for all prices in R

2+ is represented

in figure 1. Finally, for regions and/or prod-

9 This condition, which implies that own-price effects dominatescross-price effects, is a bit more restrictive than requiring the Slut-sky matrix to be negative semidefinite. See Lapan and Moschini(2004) for more details.

10 This is consistent with the current European Union proposalto set tolerance limits for non-GM products at the very low 0.9%level.

ucts with undifferentiated demand, the spec-ification used is also linear and written as:

QU(p) = a − bp.(7)

Supply

We use an extended version of the parsimo-nious specification for soybean production andsupply developed in Moschini, Lapan, andSobolevsky, which accounts for the main fea-tures of soybean production practices, reflectsthe nature of biotechnology innovation in thesoybean industry, and is suitable for calibra-tion purposes. In its original form, this specifi-cation assumes homogeneous soybean farmerswho have the choice of growing conventionalor RR soybeans or both,11 who do not segre-gate the two varieties during the productionprocess and therefore receive the same pricefor either variety. The aggregate soybean sup-ply function is written as YB = Ly, where YB istotal production consisting of a mix of conven-tional and RR soybeans, L is land allocated tosoybeans (which depends on the profitabilityof this crop), and y denotes yield (productionper hectare).12 But with differentiated prod-ucts and identity preservation costs, farmersmay obtain different prices in equilibrium. Toaccount for this, we need to represent explicitlythe cost of segregating conventional and GMproducts.

Separation of non-GM soybeans and soy-bean products requires extensive segregationactivities to ensure “identity preservation”(Lin, Chambers, and Harwood; Bullock andDesquilbet). This activity includes separationof non-GM beans at all levels of productionand along the supply chain (from plantingthrough harvest, storage, and transportation)and testing for GM content at various points inthe marketing system. These costs are modeledby a constant unit segregation cost � > 0 whichapplies if, and only if, the region in questionproduces both varieties. This unit segregationcost arises between the production level (at thefarm gate) and the point of domestic user de-mand (or, equivalently, the exporting point forgoods to be shipped to foreign markets). The

11 Perhaps a more natural assumption would be that farmers areheterogeneous with respect to the profitability of the new tech-nology, as in Lapan and Moschini (2004), which can explain in-complete adoption. But the approach taken here also allows forincomplete adoption, which arises because the two types of goodsare imperfect substitutes in demand, and it is easier to calibrate forthe purpose of empirical analysis.

12 The region subscript is omitted here and elsewhere in this sec-tion for notational simplicity.

626 August 2005 Amer. J. Agr. Econ.

Figure 1. Parametric domains for the differentiated demand system

parameter � thus represents a wedge betweenthe producer and the home demand price or,if the product is not consumed at home, theimporting region’s demand price minus trans-portation costs.

With segregation costs, the profit functionsper hectare for each variety of soybeans in eachregion are written as:

�0 = A + G

1 + �

(p0

B − �)1+� − �w(8)

�1 = A + � + (1 + �)G

1 + �

(p1

B

)1+�

− �w(1 + �)

(9)

where p0B is the (demand) price of con-

ventional soybeans (so that the farm-levelprice in the conventional soybean market

is p0B − �) and p1

B is the market price ofRR soybeans. By Hotelling’s lemma, thisspecification implies that the yield func-tions are y0 = (p0

B − �)�G for the conven-tional technology and y1 = (1 + �)×(p1

B)�G for the RR technology. The interpre-tation of the parameters involved is as follows:� is the (constant) amount of seed per unit ofland, w is the price of soybean seed, � is themarkup (which reflects the technology fee) onthe RR seed price charged by the innovator-monopolist who developed the RR technol-ogy, � is the elasticity of yield with respectto the soybean price, � is the (additive) co-efficient of unit profit increase due to the RRtechnology, � is the coefficient of yield changedue to the RR technology, and A and G areparameters subsuming all other input prices(presumed constant).

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 627

The relationship between �0 and �1 deter-mines which technology is adopted by farm-ers. Specifically, the equilibrium in which bothsoybean varieties are produced requires thatfarmers are indifferent between the two tech-nologies, that is, �0 = �1. The total supply ofland to the soybean industry in each regionis written as a function of land rents, L(�),where � = max{�0, �1}. Specifically, the landsupply function is written in the constant elas-ticity form L(�) = �, where is the elas-ticity of land supply with respect to soybeanprofit per hectare and is a scale parame-ter. The region’s adoption rate � ∈ [0, 1] or,equivalently, the land allocation between con-ventional and RR soybeans, is endogenouslydetermined in equilibrium. But for a given � ,RR and conventional soybeans will have �Land (1 − �)L hectares of land allocated tothem, respectively, and thus aggregate supplyof each soybean variety in each region can bewritten in equilibrium as:

Y 0B =

[A + G

1 + �

(p0

B − �)1+� − �w

]

× (1 − �)(

p0B − �

)�G

(10)

Y 1B =

[A + � + (1 + �)G

1 + �

(p1

B

)1+�

− �w(1 + �)]

�(1 + �)(

p1B

)�G.

(11)

U.S. Price Support Policies

The supply equations (10) and (11) were ob-tained under the assumption of no govern-ment intervention. But in the soybean sector,a major support program has been available toU.S. producers since 1996 through nonrecoursemarketing assistance loans and loan deficiencypayments (LDPs) (U.S. Department of Agri-culture 1998). Essentially, LDPs establish aneffective floor for the soybean price at the farmlevel. It turns out that, whereas the 1996 and1997 soybean crops did not benefit from LDPs,soybean prices got as low as $150/mt in the fol-lowing years, well below the national averageloan rate of $193/mt that remained fixed at thatlevel until 2002. Only in the summer of 2002did soybean prices recover to exceed the loanrate. Thus, LDPs have played a significant rolein the U.S. soybean industry in recent years andmay continue to do so again in the future.

A number of studies, summarized in Alstonand Martin, explain how price-distorting poli-

cies may affect the size and distribution of wel-fare changes because of innovation. In such asetting there is even the possibility of immis-erizing technical change, as in Bhagwati, whodemonstrates that growth may be welfare re-ducing because of various trade policy distor-tions and terms-of-trade effects. Thus, in thepolicy analysis that we present, it is impor-tant to account for the effects of price supportthrough LDPs. A relevant feature of the U.S.price support program in our setting is that itdoes not distinguish between conventional andRR soybeans (i.e., it provides the same floorprice for either variety). To integrate such ef-fects into our model, let pLDP denote the aver-age price offered by price support programs,such that in the supply equations (10) and(11) the farm-level conventional soybean price(p0

B − �) is replaced by max {pLDP, p0B − �} and

the GM soybean price p1B is replaced by max

{pLDP, p1B}.

Trade and Market Equilibrium

In our model, the world is divided into fourregions: the United States (U), Brazil (Z),Argentina (A), and the ROW (R). Such re-gional division of the world allows the model tospecifically describe individual economic char-acteristics of the main players in the soybeancomplex and to emphasize the existing differ-ences among them. The model allows us tostudy whether different regions are affecteddifferently by the introduction of RR technol-ogy and to model region-specific policy actionsof interest and estimate their economic impacton each region separately.

Trade takes place at all levels of the soybeanscomplex: in soybeans (B), soybean oil (O), andsoybean meal (M). Any region can be involvedin trading any product of any variety, and thereare no a priori restrictions on the direction oftrade. The spatial relationship among prices indifferent regions is established using constantprice differentials defined for each pair of re-gions for each product, each variety, and eachpossible direction of trade flow. These spatialprice differentials essentially represent trans-portation costs but may also incorporate theeffects of the existing import policies (Meilkeand Swidinsky).

Equilibrium Conditions

We assume that crushing one unit of soybeansproduces �O units of oil and �M units of meal

628 August 2005 Amer. J. Agr. Econ.

and that unit crushing costs (crushing margins)are constant and equal to mi (where the sub-script i indexes the region). Then, the spatialmarket equilibrium conditions for the three-good, four-region model previously outlinedare as follows:∑

i=U,A,Z,R

Q0B,i

(p0

B,i , p1B,i

)

+ 1�O

( ∑i=U,A,Z,R

Q0O,i

(p0

O,i , p0O,i

))

=∑

i=U,A,Z,R

Y 0B,i

(p0

B,i , �i)

(12)

Q0B,i

(p0

B,i , p1B,i

) + 1�O

Q0O,i

(p0

O,i , p1O,i

)= Y 0

B,i

(p0

B,i , �i), i ∈ I0

(13)

∑i=U,A,Z,R

Q1B,i

(p0

B,i , p1B,i

)

+ 1�O

∑i=U,A,Z,R

Q1O,i

(p0

O,i , p1O,i

)=

∑i=U,A,Z,R

Y 1B,i

(p1

B,i , �i)

(14)

Q1B,i

(p0

B,i , p1B,i

) + 1�O

Q1O,i

(p0

O,i , p1O,i

)= Y 1

B,i

(p1

B,i , �i), i ∈ I1

(15)

1�M

( ∑i=U,A,Z,R

QM,i (pM,i )

)

= 1�O

( ∑i=U,A,Z,R

Q0O,i

(p0

O,i , p1O,i

)

+∑

i=U,A,Z,R

Q1O,i

(p0

O,i , p1O,i

))

(16)

p0B,i + mi = �M pM,i + �O p0

O,i , i ∈ I ′0(17)

p1B,i + mi = �M pM,i + �O p1

O,i , i ∈ I ′1(18)

�0i

(p0

B,i

) = �1i

(p1

B,i

)if �i ∈ (0, 1)

�0i

(p0

B,i

) ≥ �1i

(p1

B,i

)if �i = 0

�0i

(p0

B,i

) ≤ �1i

(p1

B,i

)if �i = 1

i = U, A, Z, R,

(19)

∣∣p0B,i − p0

B, j

∣∣ ≤ t0B,i j ,

i, j = U, A, Z, R, i �= j

(20)

∣∣p1B,i − p1

B, j

∣∣ ≤ t1B,i j ,

i, j = U, A, Z, R, i �= j

(21)

∣∣p0O,i − p0

O, j

∣∣ ≤ t0O,i j ,

i, j = U, A, Z, R, i �= j

(22)

∣∣p1O,i − p1

O, j

∣∣ ≤ t1O,i j ,

i, j = U, A, Z, R, i �= j

(23)

|pM,i − pM, j | ≤ tM,i j ,

i, j = U, A, Z, R, i �= j.

(24)

Equations (12) and (14) are market clear-ing equations requiring that the total worldsoybean demand for direct use and processingequals world supply in each variety. Market-clearing conditions for regions that do nottrade soybeans and oil, if such regions exist,are represented by equation (13) (for conven-tional products) and equation (15) (for RRproducts). Thus, the subset I0 ⊂ {U, A, Z, R}contains the indices of nontrading regions forconventional products, and the subset I1 ⊂{U, A, Z, R} contains the indices of nontradingregions for RR products. Also, given in equa-tion (12), the number of elements in I0 shouldnot exceed three and the same applies to I1.Equation (16) ensures that the soybean equiv-alents of oil and meal demands are the same,in aggregate. Equations (17) and (18) ensurethat soybean processors of either variety re-ceive a constant crushing margin mi to covertheir costs. Because of the existence of spatialprice linkages among trading regions, each ofthese equations should be applied only to a sin-gle trading partner and all nontrading regions(if such regions exist in equilibrium). Thus, I′

0is the set of indices of one trading region and allnontrading regions for conventional products,and I′

1 is the set of indices of one trading regionand all nontrading regions for RR products.

Equation (19) describes the farmers’ incen-tive compatibility constraints. Production ofboth conventional and RR soybeans in thesame region takes place only when the respec-tive unit profits are the same, that is, whenfarmers are indifferent about which variety toproduce; otherwise only the more profitablevariety is produced. Equations (20)–(24) de-fine the spatial configuration of prices. Thefour-region spatial model in equilibrium willhave a maximum of three trade flows in each

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 629

product variety. In the case of the soybeancomplex and the chosen regional division ofthe world, there are three specific trade flowsthat are most likely to prevail in any conceiv-able equilibrium. Currently, trade takes placebetween the United States and the ROW, be-tween Brazil and the ROW, and between Ar-gentina and the ROW, but whether that willhold with differentiated markets is to be de-termined by equilibrium. Price differentials(transportation costs), assumed symmetric foreach pair of regions, are denoted by tk

m,ij.13

Whenever trade between two regions in a par-ticular product variety exists, the correspond-ing inequality becomes an equality.

As mentioned earlier, the model assumesthat the soybean and soybean oil demandsin the ROW are the only differentiated de-mands in the system, while U.S., Argentine,and Brazilian consumers remain indifferent towhat variety of soybeans, oil, or meal they con-sume. In a nontrivial differentiated equilib-rium with no production or import bans (i.e.,the one in which both varieties are producedand consumed), it follows that the demands inequations (12)–(24) must satisfy

Q0B,i

(p0

B,i , p1B,i

) = 0, i = U, A, Z

Q0O,i

(p0

O,i , p1O,i

) = 0, i = U, A, Z

Q1B,i

(p0

B,i , p1B,i

) ≡ QUB,i

(p1

B,i

), i = U, A, Z

Q1O,i

(p0

O,i , p1O,i

) ≡ QUO,i

(p1

O,i

), i = U, A, Z

QM,i(

pM,i) ≡ QU

M,i (pM,i ), i = U, A, Z, R.

(25)

Were we to assume that all four regions havedifferentiated demands in soybeans and soy-bean oil, only the last of the five identities inequation (25) would apply.

The existence and uniqueness of equilibriumis guaranteed by the standard shape of de-mand and supply functions (Samuelson). Butbecause we are assuming that a region pro-ducing only conventional soybeans pays nosegregation cost, we are introducing a discon-tinuity that can affect the uniqueness propertyof equilibrium. A limitation of the equilibriumsystem (12)–(24) is that it does not allow recov-ery of all individual trade flows, that is, distinctexports/imports of soybeans, soybean oil, andsoybean meal. This feature ultimately is dueto the assumption of constant-returns-to-scale

13 See the section on calibration and table 2 for more details.

(and no capacity constraints) for the crushingtechnology in all regions of the world, whichmakes the interregional distribution of crushundetermined in equilibrium. The meaningfultrade flow that can be recovered from equi-librium is the factor content of trade, in theform of the excess supply of soybeans (in eachvariety) remaining after subtracting domesticsoybean demand and the soybean equivalentof domestic oil demand from the domestic sup-ply of beans

ES jB,i = Y j

B,i − Q jB,i − 1

�OQ j

O,i ,

i = U, A, Z, R, j = 0, 1.

(26)

We can call ESjB,i the soybean-equivalent net

exports. However, because trade in soybeanproducts does not necessarily follow the fixedproportions of the crushing technology, thismeasure must be supplemented by a residual,which here is defined as the residual meal netexport:

ESM,i = 1�O

(Q0

O,i + Q1O,i

)�M − QM,i ,

i = U, A, Z, R.

(27)

Solution Algorithm

The task is that of solving a spatial four-region,three-good equilibrium model. The solutionof spatial equilibrium models can be tracedback to Samuelson, who showed that in thepartial-equilibrium (one commodity) context,the problem of finding a competitive equilib-rium among spatially separated markets couldbe converted into a maximum problem. Hesuggested that this problem could be solvedby trial and error or by a systematic proce-dure of varying export shipments in the di-rection of increasing social welfare. Takayamaand Judge extended Samuelson’s work to amultiple-commodity competitive equilibriumand, under the additional assumption of lin-ear regional demand and supply functions,reduced spatial equilibrium to a quadraticprogramming problem solvable with avail-able simplex methods. Attempts to extend thisframework to nonlinear demand and supplyspecifications have been less successful, as dis-cussed in Takayama and Labys.

In view of the above, we elected to solve di-rectly the system of nonlinear equations (12)–(24) defining the spatial equilibrium conditionsby using numerical techniques implemented

630 August 2005 Amer. J. Agr. Econ.

by user-written programs coded in GAUSS.14

Obviously, the equations defining the system tobe solved must be binding, but in our case thenumber of binding equations in (12)–(24) is notdetermined a priori. There are two sources ofambiguity: the number of trade flows in eachcommodity and the possible specialization inproduction of a particular soybean variety ineach region.15 Our algorithm looks for an equi-librium by repeatedly solving the fluctuating-in-size binding portion of the system (12)–(24)over all of the following combinations: (a) eachregion specializes in conventional soybeans, inRR soybeans, or does not specialize; (b) thereis no trade in RR beans/oil; (c) there is onlyone RR trade flow involving a pair of regions,in either direction, for all possible region pairs;(d) there are two RR trade flows, in all possiblecombinations of directions, excluding (for ar-bitrage reasons) cases when the same region isboth exporter and importer of the same prod-uct(s); (e) there are three RR trade flows, in allpossible combinations of directions, excluding(for arbitrage reasons) cases when the same re-gion is both the exporter and importer of thesame product(s). The solution—when found—is checked against the remaining nonbindingequations of the system (12)–(24).

Calibration

The parameters of the model are calibrated,such as to replicate prices and quantities in thesoybean complex, for the crop year 1998–99,the most recent complete year when the anal-ysis was undertaken. Production and utiliza-tion data are given in table 1. Additional dataon the history of world adoption rates for RRsoybeans, as well as prices for various soybeanproducts in the main world markets, are re-ported in Sobolevsky, Moschini, and Lapan.Based on that, the benchmark U.S. prices forsoybeans, oil, and meal were set to $176, $441,and $145 per mt, respectively. In the UnitedStates in 1998–99, the soybean price at the pro-ducer level differed from $176/mt because ofLDPs. The spatial price differentials (transfercosts) were set at the levels used in Moschini,Lapan, and Sobolevsky, extended to accountfor South America being broken down into

14 This software package is equipped with the eqSolve procedurethat solves N × N systems of nonlinear equations by inverting thesystems’ Jacobian, while iterating until convergence.

15 For example, when differentiated markets exist only in theROW, the size of the binding portion of the system in equations(12)–(24) can be anywhere from N = 5 to N = 21.

two regions.16 See table 2 for individual trans-portation costs.

Calibration of demand parameters requiresassigning values to the parameters (a0, a1, b0,b1, c) so as to retrieve the benchmark quan-tity and price data. As is typical in this setting,a range of parameter values is admissible, de-pending on elasticity assumptions. But here,elasticity assumptions are difficult becausethe benchmark equilibrium is a pooled one(without segregation), whereas the demandsystem we wish to calibrate distinguishes be-tween GM and non-GM goods. To proceed, wefollow the strategy whereby the five parame-ters of interest are identified by the (observed)benchmark price and (pooled) quantity de-manded ( p and Q), by the own-price elastic-ity of undifferentiated demand εUU, by theown-price elasticity of conventional demandε00, by the fraction ∈ (0, 1) of demand thatis “indifferent” between GM and non-GM atprices p0 = p1 = p, and by the fraction k ≥ 1by which total demand would increase if thenew product were to become available at pricep1 = p. More specifically, because no signifi-cant segregation took place in the referenceyear 1998–99, we can assume, as discussed ear-lier, that in this reference year Q0 = 0 and Q1 =a1 + c p0 − b1 p1. Hence, for the observed totalquantity demanded Q and price p, it must bethat

Q = a1 + ca0

b0−

(b1 − c2

b0

)p.(28)

If p0 were to fall from the choke level p0

to p0 = p1 = p, total demand (the sum of dif-ferentiated demands) increases such that wewrite

a0 + a1 − (b0 + b1 − 2c) p = k Q(29)

but a fraction of the total demand is indifferentat those prices, such that we write

a1 − (b1 − c) p

(a0 + a1) − (b0 + b1 − 2c) p= .(30)

The own-price elasticity of (undifferentiated)demand at price p satisfies

16 Data presented by Schnepf, Dohlman, and Bolling support the$30/mt soybean transportation cost estimate between the UnitedStates and the ROW and at least a $10/mt U.S. transportation costadvantage over Argentina and Brazil due to distance and higherinsurance costs.

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 631

Table 2. Model’s Parameters and Their Baseline Values

Values

Parameter Description U.S. Brazil Argentina ROW

εUUB Own-price nonsegregated bean

demand elasticity−0.4 −0.4 −0.4 −0.4

εUUO Own-price nonsegregated oil

demand elasticity−0.4 −0.4 −0.4 −0.4

εUUM Own-price nonsegregated meal

demand elasticity−0.4 −0.4 −0.4 −0.4

ε00B Own-price conventional bean

demand elasticity−4.5

ε00O Own-price conventional oil demand

elasticitya−4.5

kB Total bean demand increase due toprice decreasea

1.05

kO Total oil demand increase due toprice decreasea

1.05

B Share of indifferent bean demand intotala

0.5

O Share of indifferent oil demand intotala

0.5

� Elasticity of land supply w.r.t.soybean price

0.8 1.0 0.8 0.6

� Elasticity of yield w.r.t. soybean price 0.05 0.05 0.05 0.05�� Unit seed cost ($) 45.0 40.0 40.0 40.0�� Producer unit profit change due to

RR technology (ceteris paribus)15.0 15.0 15.0 10.0

r Producer rent share in average profit 0.4 0.4 0.4 0.4� Markup on RR seed price 0.4 0.2 0.2 0.2� Coefficient of yield increase due to

RR technology0.0 0.0 0.0 0.0

pLDP Soybean farmer LDP/loan price($/mt)

193.0

� Segregation cost ($/mt) 0.0 0.0 0.0 0.06.6 6.6 6.6 6.6

13.2 13.2 13.2 13.219.8 19.8 19.8 19.8

t km,ij Transportation costs ($/mt)b

U.S. – (30, 60, 30) (30, 60, 30) (30, 60, 30)Brazil (30, 60, 30) – (27, 47, 27) (40, 70, 40)Argentina (30, 60, 30) (27, 47, 27) – (40, 70, 40)ROW (30, 60, 30) (40, 70, 40) (40, 70, 40) –

aSee text for details.bEach triplet represents the transportation cost between the two source/destination regions for beans, oil and meal, respectively ($/mt).

εUU = −(

b1 − c2

b0

)p

Q(31)

and the own-price elasticity of conventionaldemand elasticity at p0 = p1 = p satisfies

ε00 = −b0p

a0 − (b0 − c) p.(32)

To solve equations (28)–(32) for the param-eters of interest, we set = 0.5, k = 1.05,ε00 = −4.5, and, as in Moschini, Lapan, andSobolevsky, εUU = −0.4.

Calibrated supply parameters are obtainedusing specifications (18)–(23) together withspecific assumptions as in Moschini, Lapan,and Sobolevsky. Specifically, as in that study,the unit seed costs �� are set at {45, 40, 40,40},17 values consistent with the data reportedby the U.S. Government Accounting Officeand Schnepf, Dohlman, and Bolling. The RR

17 Here and elsewhere in the text, the elements of the four-dimensional vectors refer to the U.S., Brazil, Argentina, and theROW, respectively.

632 August 2005 Amer. J. Agr. Econ.

seed markups are set to � = {0.4, 0.2, 0.2,0.2}.18 Note that the assumption of a con-stant markup is made to keep the analysisto a tractable scope. A more complete modelwould endogenize the pricing decision of theinnovator (Lapan and Moschini 2004). TheU.S. value of 0.4 reflects the observed tech-nology fee charged by major seed companies(about $6 per bag). Furthermore, the assump-tion that seed price markups are lower out-side the United States, as noted by a reviewer,is quite consistent with the presumption thatan innovator-monopolist would rationally tryto segment markets. Such an attempt at third-degree price discrimination would tailor theseed price markup to local conditions, and anumber of reasons suggest that the improvedseed demand may be more elastic in non–U.S.regions. In particular, the strength of IPRs iscrucial in determining the improved seed de-mand elasticity. For example, in Argentina RRseed prices have declined substantially, afterbeing set initially at levels comparable to thoseseen in the United States, a fact ultimately dueto the weaker IPRs for this crop in that country(U.S. Government Accounting Office; Gold-smith, Ramos, and Steiger).19 Because IPRprotection is unlikely to be any better in Brazilor in the ROW, for the remaining three regionsthe RR seed markup is assumed to be one-halfof the U.S. value, and thus we set � = 0.2.

To calibrate the parameter � (i.e., the ad-ditive efficiency gain of the GM variety), wenote that �� ≡ �1 − �0 = � − ��� (underthe baseline assumption of no segregation and� = 0). Based on U.S. data in 2000, we esti-mate that the cost savings of using RR technol-ogy lies between $8.90 and $22.49 per hectareand therefore for this country, we conserva-tively set �� = 15. We also assume that, ifBrazil and Argentina were to face the sameRR seed markup (� = 0.4) and the same soy-bean seed price (�w = 45), for these countrieswe would also have �� = 15. For the ROW, onthe other hand, the assumption is that it wouldhave �� = 10 under the same seed price pa-rameters as the United States (the lower unit

18 By setting a constant markup, we are modeling a somewhat pas-sive seed supplier. This is done to keep the analysis to a tractablescope. As pointed out by a reviewer, a more complete model wouldendogenize this parameter and account for the possibility that theGM seed supplier may attempt to segment markets by third-degreeprice discrimination. See Lapan and Moschini (2004) for some the-oretical results with an optimizing innovator-monopolist.

19 For related issues on international IPR enforcement in agricul-tural biotechnology, see Gaisford et al. and Giannakas.

profit increase justified by the lower yields; seetable 1). The actual calibration of the � pa-rameter in each region, of course, adjusts theabove �� assumptions by the previously dis-cussed differences in seed prices. To calibrate, it is useful to relate this parameter to themore standard notion of elasticity of land sup-ply with respect to soybean prices. Specifically, = r� , where � is elasticity of land supply withrespect to soybean prices and r ≡ �/(pBy) isthe farmer’s share (rent) of unit revenue. Theelasticity � is set to 0.8 in the United Statesand 0.6 in the ROW, based on the review ofthe literature discussed in Moschini, Lapan,and Sobolevsky. The value of � = 1 .0 usedin Moschini, Lapan, and Sobolevsky for SouthAmerica here applies to Brazil, whereas we set� = 0.6 for Argentina.20

The technical coefficients �M and �O are setto their world average values for the 1998–99 crop year; that is, �M = 0.7985 and �O =0.1810. As for segregation costs, we rely onLin, Chambers, and Harwood, who extendthe segregation cost estimates available forspecialty crops grown in the United States(Bender et al.) to non-GM soybeans. Theyproject that for U.S. grain handlers, segregat-ing non-GM soybeans may cost from $6.60to $19.80/mt (depending on whether han-dling process patterns for high-oil corn or theones for sulfonylurea-tolerant soybeans wereused).21 Because segregation costs are crucialto our investigation, the model is solved forboth the minimum and maximum values ofthe range identified by Lin, Chambers, andHarwood, as well as the midpoint of that rangeand costless segregation. That is, for all re-gions, we alternatively set � = 0, � = 6.6,� = 13.2, and � = 19.8. Finally, we need toaccount for the fact that LDPs were effectivein the benchmark year 1998–99. Thus, the U.S.producer price is set at $193/mt in 1998–99, andin scenarios in which the U.S. price support pro-gram is assumed to remain in force, we also willhave pLDP = 193. The summary of all param-eters and their values used for model calibra-tion purposes and for the baseline solution ofthe world soybean complex partial equilibriumdefined by equations (12)–(25) is provided intable 2.

20 Brazil has vast areas of undeveloped arable land in its Center-West (the “cerrado”) region that can serve, and have served, asengines of soybean production growth (Schnepf, Dohlman, andBolling). In Argentina, as in the United States, growth in soybeanareas can be achieved only by substitution for other crops.

21 Estimates presented in Bullock and Desquilbet are within thisrange.

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 633

Table 3. Equilibrium Solution with No U.S. Price Support, Various Scenarios

Bean Price Oil Price Soybean Supply Export BEaMeal Export

Region � Conv. RR Conv. RR Price Conv. RR Conv. RR Mealb

PreinnovationU.S. 0.00 181.9 480.2 143.6 70.1 26.9 2.3BR 0.00 171.9 470.2 133.6 35.6 18.8 5.1AR 0.00 171.9 470.2 133.6 21.1 15.3 0.9ROW 0.00 211.9 540.2 173.6 32.3 −60.9 −8.3

No segregationU.S. 1.00 174.5 444.8 142.3 69.3 24.8 3.2BR 1.00 164.5 434.8 132.3 35.9 18.6 5.5AR 1.00 164.5 434.8 132.3 21.2 15.2 1.0ROW 1.00 204.5 504.8 172.3 32.6 −58.6 −9.7

High segregation cost = $19.8/mtU.S. 0.95 200.4 174.8 586.7 445.5 142.5 3.7 65.8 3.7 21.3 3.2BR 1.00 164.8 435.5 132.5 0.0 35.9 0.0 18.6 5.5AR 1.00 164.8 435.5 132.5 0.0 21.3 0.0 15.3 1.0ROW 1.00 230.4 204.8 616.4 505.5 172.5 0.0 32.6 −3.7 −55.2 −9.7

Medium segregation cost = $13.2/mtU.S. 0.90 194.0 175.0 551.7 447.0 142.4 7.0 62.5 7.0 18.1 3.1BR 1.00 165.0 437.0 132.4 0.0 36.0 0.0 18.7 5.5AR 1.00 165.0 437.0 132.4 0.0 21.3 0.0 15.3 1.0ROW 1.00 224.0 205.0 611.7 507.0 172.4 0.0 32.6 −7.0 −52.0 −9.7

Low segregation cost = $6.6/mtU.S. 0.70 187.9 175.5 522.8 454.5 141.4 20.9 48.8 20.9 4.6 2.9BR 1.00 165.5 444.5 131.4 0.0 36.1 0.0 18.9 5.4AR 1.00 165.5 444.5 131.4 0.0 21.3 0.0 15.4 1.0ROW 1.00 217.9 205.5 582.8 514.5 171.4 0.0 32.7 −20.9 −38.9 −9.2

Zero segregation cost = 0U.S. 0.62 183.6 177.9 502.9 471.1 140.5 27.0 43.6 27.0 0.0 2.3BR 0.99 173.6 164.2 492.9 440.7 130.5 0.3 35.5 0.3 18.3 5.4AR 1.00 164.2 440.7 130.5 0.0 21.2 0.0 15.2 1.0ROW 1.00 213.6 204.2 562.9 510.7 170.5 0.0 32.5 −27.3 −33.5 −8.7

Note: Prices are in U.S.$/mt, quantities are in mil mt.aBE = Bean Equivalent (see equation (26)).bMeal exports, additional to those imbedded in previous two columns (see equation (27)).

Results

The model is solved for several parameter val-ues and policy scenarios. We study the implica-tions of introducing the RR technology in thesoybean complex, look at the effects of the U.S.domestic price support policy, and analyze theimpacts of policy aimed at the new GM prod-uct in the ROW and/or in Brazil. One resultcommon to all scenarios is that the directionof trade flows, when flows are nonzero, doesnot change in any equilibrium from what isobserved in the preinnovation market. Tradein all products and in all varieties flows fromthe United States, Argentina, and Brazil to theROW except for special cases when particularregions find themselves in autarky in a partic-ular product variety.

Consider first the case of no U.S. price sup-port. Equilibrium adoption rates, prices, soy-bean supply, and exports for various scenariosare reported in table 3. Our basic benchmarkis the “preinnovation” scenario. Because themodel was calibrated on 1998–99 data, whenLDPs were in fact effective and when consider-able adoption of RR soybeans had already oc-curred, the simulated equilibrium prices underthe “preinnovation” scenario are higher thanthose observed in that year. Next, considerthe “no-segregation” scenario, which wouldattain if segregation costs were prohibitivelyhigh. In such a case, there is no effective de-mand for the conventional product and thusthe world moves to complete adoption of thenew (more efficient) production technology.U.S. soybean prices fall by 4%, oil by 7%, and

634 August 2005 Amer. J. Agr. Econ.

meal by 1%, and prices in all other regionsdecline as well from the preinnovation bench-mark. U.S. soybean supply falls because theregion’s new technology cost savings are thesmallest among the four regions, owing to en-forcement of IPRs, and are not high enough tooffset the price decline. Other regions’ suppliesgrow. Consumption increases in all regions butthe ROW, where GM-conscious consumers cutdown on the consumption of inferior RR soy-beans and soybean oil.

The “zero-segregation-cost” scenario, re-ported at the bottom of table 3, is the otherpolar case and it is useful because it iso-lates the RR technology impacts from thosecaused by segregation costs. Here the UnitedStates grows most of the conventional soy-beans (Brazil adds a tiny amount). The fact thatthe ROW does not, despite it having the GM-conscious consumers, is because in our modelU.S. producers experience a smaller cost reduc-tion owing to the RR technology, as previouslydiscussed. Thus, “comparative advantage” inthis model is very much affected by unevenIPR protection across countries. Without anysegregation costs, the world share of the con-ventional soybean market reaches 17%. TheU.S. adoption rate is a low 62% and the re-gion finds itself in an autarky equilibrium inthe RR market, exporting only the conven-tional variety to the ROW. As a result, RRprices in the other regions fall compared tothe low-segregation-cost scenario (see below)under the pressure of weakened RR importdemand from the ROW. In regions that growboth varieties, there is a “price premium” forconventional soybean producers, which is nec-essary to offset the higher production costs forthis crop vis-a-vis RR soybeans. In the UnitedStates such a premium is $5.7/mt (about 3.2%of the RR soybean price).

The fact that it actually costs to segregateaffects the result in a predictable way. Thehigh level of segregation costs that we consid-ered, $19.8/mt or 11% of the price received byU.S. farmers growing conventional soybeans,is almost enough to drive out conventionalsoybean production. A small amount of con-ventional soybeans is produced in this sce-nario (2% of world output), all in the UnitedStates, and the welfare gains relative to the no-segregation scenario are minimal. The UnitedStates is the only region producing both vari-eties (its adoption rate in equilibrium is 95%),while all other regions specialize in produc-tion of RR soybeans. As segregation costs de-cline to $13.2/mt or to $6.6/mt, the effects are

shared between the conventional variety’s con-sumers and producers because of the fact thatdemands are not completely inelastic and be-cause conventional consumer prices fall andconventional producer prices increase. Thisbenefits ROW consumers and U.S. produc-ers, whose share of conventional soybean pro-duction increases to 30% when segregationcosts are low. The United States remains theonly producer of the conventional variety, withthe worldwide share of the conventional soy-bean market growing to 13%. As more pro-duction shifts toward conventional soybeans,the world’s RR supply decreases, causing RRprices to increase.

The next set of results, shown in table 4, re-peats the analysis given the presence of theU.S. LDP price support program. The U.S. pro-ducer price floor of $193/mt is effective bothin the preinnovation and postinnovation sit-uations, with the latter entailing a support ofabout 10% over the market price. The pres-ence of this price support program in theUnited States has a dramatic impact on the spa-tial allocation of soybean production, such thatthe United States does not produce the con-ventional variety at all. This is because LDPsequate farmer prices for conventional and RRsoybeans and create a permanent incentive tospecialize in the (more efficient) RR variety.Brazil emerges as the only producer and ex-porter of conventional products to the ROWin all scenarios but the zero-segregation-costone, in which Argentina also dedicates 50% ofits total production to conventional soybeans.The U.S. producer price support program de-presses prices everywhere, such that in all re-gions except the United States such reducedprices tend to offset the supply expansion ef-fect of the more efficient RR technology. Therelative effect of various levels of segregationcosts is similar to the case of no price supportreported in table 3.

Welfare Effects

For each scenario, we are particularly inter-ested in computing the welfare change effectsfor each region and for distinct groups withinregions. The benchmark for all welfare calcu-lations is the preinnovation scenario in whichthe RR soybean is not yet available (� i = 0, i =U, A, Z, R), such that demands are describedby equations (4) and (7), while supplies are de-scribed by equation (10) with � = � = 0. If p0

j,iis the equilibrium undifferentiated preinnova-tion price for product j in region i, and p0

j,i and

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 635

Table 4. Equilibrium Solution with U.S. LDP Price Support Program, Various Scenarios

Bean Price Oil Price Soybean Supply Export BEaMeal Export

Region � Conv. RR Conv. RR Price Conv. RR Conv. RR Mealb

PreinnovationU.S. 0.00 176.6 468.7 139.5 74.0 30.3 2.3BR 0.00 166.6 458.7 129.5 34.4 17.5 5.2AR 0.00 166.6 458.7 129.5 20.5 14.6 0.9ROW 0.00 206.6 528.7 169.5 31.8 −62.3 −8.4

No segregationU.S. 1.00 165.6 425.4 135.5 75.7 30.3 3.2BR 1.00 155.6 415.4 125.5 34.0 16.4 5.6AR 1.00 155.6 415.4 125.5 20.3 14.2 1.0ROW 1.00 195.6 485.4 165.5 31.7 −60.9 −9.9

High segregation cost = $19.8/mtU.S. 1.00 166.0 426.3 135.8 0.0 75.7 0.0 30.4 3.2BR 0.91 185.3 156.0 578.0 416.3 125.8 3.1 31.0 3.1 13.3 5.6AR 1.00 156.0 416.3 125.8 0.0 20.3 0.0 14.2 1.0ROW 1.00 225.3 196.0 599.0 486.3 165.8 0.0 31.8 −3.1 −57.9 −9.9

Medium segregation cost = $13.2/mtU.S. 1.00 166.2 426.6 135.9 0.0 75.7 0.0 30.4 3.2BR 0.87 178.8 156.2 541.9 416.6 125.9 4.3 29.8 4.3 12.2 5.6AR 1.00 156.2 416.6 125.9 0.0 20.3 0.0 14.2 1.0ROW 1.00 218.8 196.2 599.3 486.6 165.9 0.0 31.8 −4.3 −56.8 −9.9

Low segregation cost = $6.6/mtU.S. 1.00 166.7 432.0 135.4 0.0 75.7 0.0 30.6 3.0BR 0.60 172.8 156.7 510.8 422.0 125.4 13.9 20.4 13.9 2.9 5.5AR 1.00 156.7 422.0 125.4 0.0 20.4 0.0 14.3 1.0ROW 1.00 212.8 196.7 580.8 492.0 165.4 0.0 31.8 −13.9 −47.8 −9.6

Zero segregation cost = 0U.S. 1.00 167.4 439.5 134.5 0.0 75.7 0.0 30.9 2.8BR 0.51 167.0 157.6 482.9 430.6 124.5 17.1 17.4 17.1 0.0 5.4AR 0.50 167.0 157.4 482.9 429.5 124.5 10.3 10.2 10.3 4.2 1.0ROW 1.00 207.0 197.4 552.9 499.5 164.5 0.0 31.9 −27.4 −35.0 −9.1

Note: Prices are in U.S.$/mt, quantities are in mil mt.aBE = Bean equivalent (see equation (26)).bMeal exports, additional to those imbedded in previous two columns (see equation (27)).

p1j,i are equilibrium prices of conventional and

RR varieties in the postinnovation differenti-ated market, then, setting the reservation pricep1

j,i ≡ p0j,i , the change in consumer surplus is

defined as follows:

�CS j,i =∫ p1

j,i

p1j,i

Q1j,i

(p0

j,i , p1j,i

)dp1

j,i

+∫ p0

j,i

p0j,i

Q0j,i

(p0

j,i , p1j,i

)dp0

j,i .

(33)

Consumer surplus changes in undifferentiatedmarkets are computed in the standard way:

�CS j,i =∫ p1

j,i

p1j,i

QUj,i (p) dp.(34)

A change in producer surplus between prein-novation and differentiated market scenariosis

�PSi =∫ ˜�i

ˆ�i

Li (v) dv(35)

where Li is the land allocation function, ˆ�iis the preinnovation equilibrium average unitprofit, and ˜�i is the differentiated market equi-librium average unit profit. The innovator-monopolist’s profit is computed as

�M =∑

i=U,S,R

�i Li ( ˜�i )�i �wi(36)

where �i is the equilibrium rate of adoption inregion i. Finally, the total change in welfare isdefined as:

636 August 2005 Amer. J. Agr. Econ.

�WU =∑

j=B,O,M

�CSj,U

+ �P SU + �M + �Gov

�Wi =∑

j=B,O,M

�CS j,i

+ �P Si , i = A, Z, R

(37)

where �Gov denotes the change in U.S. tax-payer surplus due to changes in LDP outlays.

Table 5 reports the estimated welfare effectsfor various scenarios, for both the case of noU.S. price support and the case when LDPsare effective. Considering first the case of noU.S. price support, it is apparent that there isa sizeable welfare gain that would arise from aworldwide adoption of the more efficient RRtechnology, even when no segregation takesplace. The estimated gain of $1.564 billion

Table 5. Estimated Welfare Effects of RR Soybean Innovation, Various Scenarios (U.S.$ mil)

No U.S. Price Support With U.S. LDP Price Support Program

�W �W �W fromRegion �CS �PS ��M Total �CS �PS ��M �Gov Total no LDP

No segregationU.S. 323 −117 831 1,037 478 429 859 −860 907 −123BR 120 72 192 169 −51 117 −75AR 43 47 89 62 −27 35 −54ROW 125 121 247 472 7 479 233World 611 123 831 1,564 1,181 358 859 −860 1,538 −26

High segregation cost = $19.8/mtU.S. 310 −95 807 1,021 461 429 850 −830 910 −111BR 116 83 199 163 −38 125 −74AR 41 53 94 60 −19 41 −53ROW 131 132 263 460 20 480 217World 597 173 807 1,577 1,144 392 850 −830 1,556 −21

Medium segregation cost = $13.2/mtU.S. 301 −83 784 1,003 455 429 846 −819 911 −92BR 112 90 202 161 −33 129 −73AR 40 57 97 59 −16 43 −54ROW 145 138 283 470 25 494 211World 598 201 784 1,584 1,144 405 846 −819 1,577 −7

Low segregation cost = $6.6/mtU.S. 275 −46 690 919 428 429 816 −777 896 −24BR 97 109 206 149 −14 134 −71AR 36 69 104 55 −4 51 −53ROW 198 155 353 474 42 516 163World 606 286 690 1,582 1,106 452 816 −777 1,597 15

Zero segregation cost = 0U.S. 169 120 651 940 396 429 772 −727 870 −70BR 116 61 177 129 15 144 −32AR 43 40 83 50 9 59 −24ROW 399 111 511 552 63 615 104World 727 332 651 1,710 1,127 517 772 −727 1,688 −22

is 25% lower than the worldwide gain esti-mated in Moschini, Lapan, and Sobolevsky,mostly because in their model they did notallow for preference differentiation. In thisno-segregation scenario, consumers capture39% of the welfare gain, while the innovator-monopolist captures another 53%. Note thatconsumers in the ROW gain, despite the factthat 50% of the market has preference for theconventional variety, which is unavailable inthis scenario. U.S. farmers lose from the intro-duction of RR soybean in all scenarios exceptthat of zero segregation costs. The fact thatfarmers elsewhere gain is due to their access-ing the RR technology at a lower cost than theU.S. farmers because of lower IPR standardsin these other regions. The further increase inwelfare that takes place when segregation ispossible at zero costs illustrates the benefitsof allowing differentiated consumers (in the

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 637

ROW) to exercise the choice to consume con-ventional products instead of RR products.

The existence of positive segregation costs,however, takes away most of the additionalwelfare gain arising because of unconstrainedpreferences. As noted earlier, the high segre-gation cost considered ($19.8/mt) is nearly pro-hibitive, such that the model outcomes for thisscenario are similar to the no-segregation sce-nario. But even the low-segregation cost sce-nario entails an overall welfare gain from RRinnovation similar to the no-segregation sce-nario. Consumers in the ROW, however, ben-efit from feasible segregation in all scenarios.It is also interesting to note that the profit ofthe RR seed supplier is positively related tothe level of segregation costs because highersegregation costs lead to higher RR adoptionrates in equilibrium.

Table 5 illustrates that the presence of LDPsin the United States entails large welfare re-distribution effects and a small efficiency loss(relative to no LDPs). Worldwide welfaregains are similar to those in the no-LDP sce-nario, which means that the possibility of im-miserizing growth, discussed earlier, does notoccur. The U.S. price support program, not sur-prisingly, has a large positive impact on the wel-fare of U.S. farmers, ensuring that they benefitsubstantially from the introduction of the newtechnology. This price distortion, however, de-presses RR prices worldwide to the degree thatfarmers in Brazil and Argentina lose when-ever segregation costs are positive and are ableto gain only in the zero-segregation-cost casewhen 50% of their production is in the higher-priced conventional market. The overall effectof the U.S. price support program is to reducewelfare in all regions except the ROW, a netimporting region that therefore benefits fromthe price decline. Interestingly, the U.S. pricesupport actually improves world welfare at thelow ($6.6/mt) level of segregation costs. Thisis a classic second-best result. The existenceof market power for the innovator-monopolistmeans that the adoption rate of the innova-tion, ex post, is lower than socially optimal.Because the U.S. price support increases adop-tion of RR seeds, it counters that effect, butthis useful impact disappears as the segrega-tion cost rises (because conventional produc-tion is driven out of the market anyway).

Effects of Policies for GM Products

In table 6, we report the impact of variouspolicies that countries can consider (and are

considering) to deal with the introduction ofGM products. Here we restrict ourselves tothe case of only the medium-level segrega-tion costs ($13.2/mt) and to the case wherethere is no price support program in the UnitedStates. The first set of results in table 6 looksat what is currently the case for the EU andseveral Asian countries that are part of theROW region—a ban on local production ofRR soybeans and products. As the last col-umn in table 6 shows, the ban benefits theROW (as well as Argentina and Brazil), butit hurts the United States as well as world wel-fare. This case is characterized by complete re-gional specialization in production. The ROWmeets its domestic demand for conventionalsoybean products, while the United States,Brazil, and Argentina produce and export onlythe RR variety. De facto segregation costs arezero in equilibrium because no actual segre-gation needs to take place. In comparison tothe unregulated production scenario, the banimproves the ROW welfare by $35 millionbecause of the positive change in consumersurplus that arises from lower conventionalproduct prices (driven down by the effectivelyzero segregation costs) under the ban, whichmore than offsets the corresponding negativechange in producer surplus.22 The ban reducesU.S. welfare by $81 million, primarily becauseof forgone innovator-monopolist profit.

As noted earlier, Brazil has for manyyears declined to approve the growing ofRR soybeans but, in response to the mas-sive clandestine growing of GM soybeans, inSeptember 2003 it enacted a temporary andlimited exemption for the 2003–4 crop year.Brazil’s attempt at implementing a local pro-duction ban for GM products has been ra-tionalized by Brazil’s alleged interest inavoiding segregation costs in order to gaina competitive advantage selling conventionalsoybeans and soybean products to Europeand the ROW. The second set of results intable 6 shows that the ban on RR produc-tion in Brazil actually does not benefit the re-gion overall, although it benefits that coun-try’s farmers. Again, the ban results in thecomplete regional specialization in produc-tion, with the United States, Argentina, andthe ROW producing and trading only theRR variety. Because Brazil specializes in pro-ducing conventional beans, it does not incur

22 But this ROW welfare gain disappears at lower levels of seg-regation costs.

638 August 2005 Amer. J. Agr. Econ.

Table 6. Economic Impacts of Government Policies (No-LDP Case with $13.2/mt SegregationCost): Changes from Preinnovation Equilibrium and from No-Ban Scenario

Bean Supply Export BEa�CS �PS �W Export �W from

Region � Total Total ��M Total Conv. RR Conv. RR Mealb No Ban

RR production ban in ROWU.S. 1.00 239 9 675 922 0.0 70.0 0.0 26.0 2.6 −81BR 1.00 81 137 218 0.0 36.2 0.0 19.2 5.3 16AR 1.00 30 85 116 0.0 21.4 0.0 15.5 1.0 19ROW 0.00 277 41 318 32.4 0.0 0.0 −60.7 −8.9 35World 626 272 675 1,573 −11

RR production ban in BrazilU.S. 1.00 326 −124 712 914 0.0 69.3 0.0 24.8 3.1 −89BR 0.00 −94 188 93 36.6 0.0 20.4 0.0 4.7 −109c

AR 1.00 43 45 87 0.0 21.2 0.0 15.2 1.0 −10ROW 1.00 291 118 409 0.0 32.6 −20.4 −40.0 −8.8 127World 565 226 712 1,504 −81

RR production ban in ROW and BrazilU.S. 1.00 113 215 564 891 0.0 71.1 0.0 27.6 2.3 −112BR 0.00 35 −96 −61 35.0 0.0 18.1 0.0 5.2 −262AR 1.00 14 148 162 0.0 21.7 0.0 15.9 0.9 65ROW 0.00 271 −87 183 32.0 0.0 −18.1 −43.5 −8.4 −99World 432 180 564 1,176 −409

RR production and import bans in ROWU.S. 0.65 353 −149 391 594 24.1 45.1 24.1 0.0 3.8 −408BR 0.51 208 −76 132 17.2 17.8 17.2 0.0 6.0 −70AR 0.30 72 −45 27 14.6 6.2 14.6 0.0 1.2 −70ROW 0.00 −1,021 363 −658 33.3 0.0 −56.0 0.0 −11.0 −941World −389 93 391 95 −1,489

RR production bans in ROW and Brazil and import ban in ROWU.S. 0.67 498 −371 337 464 22.3 45.7 22.3 0.0 4.2 −539BR 0.00 −124 284 160 37.1 0.0 21.0 0.0 4.7 −42AR 0.31 92 −112 −20 14.2 6.2 14.2 0.0 1.3 −117ROW 0.00 −727 256 −471 33.0 0.0 −57.4 0.0 −10.2 −754World −261 57 337 133 −1,451

Note: Monetary variables are measured in mil U.S.$; quantities are in mil mt.aSee footnote a, table 3.bSee footnote b, table 3.cComprised of CS change of −206 and PS change of 98.

segregation costs and therefore prices receivedby Brazilian farmers increase relative to thepreinnovation benchmark. These higher pricesbenefit the region’s farmers but hurt its con-sumers, who consume the domestically grownand crushed conventional products despitehaving no differentiated tastes. Comparingwelfare changes between the ban and no-banscenarios, we see that whereas Brazilian farm-ers gain from the ban by switching to higher-priced conventional soybeans, the same switchin consumption due to the noncompetitivepricing from potential RR imports hurts theregion more and results in a net $109 millionloss of welfare.

If the ROW and Brazil ban RR productionsimultaneously, our third set of results in ta-

ble 6 suggests that welfare will fall for the tworegions and for the world in general. Brazilexports the conventional variety to the ROW,and the United States and Argentina produceonly RR products for domestic consumptionand export. With two regions growing conven-tional soybeans, the size of the conventionalsoybean sector proves to be quite large in equi-librium. As a result, equilibrium is character-ized by equal conventional and RR soybeanand oil prices in the ROW, with 17% of theindifferent demand actually satisfied by con-ventional soybeans and soybean oil. In gen-eral, all prices in this equilibrium are lowerthan their preinnovation benchmark counter-parts, implying that consumers gain from theRR technology in all regions and producers

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 639

in Brazil and the ROW lose. A welfare com-parison between the ban and no-ban scenariosshows that forced abundance of the conven-tional variety and a relative scarcity of theRR product imply that equilibrium conven-tional prices in the ban scenario are lower thantheir counterparts in the unregulated scenario,whereas RR prices are higher. As a result, pro-ducers in Brazil and the ROW and all but ROWconsumers lose, and Argentina is the only re-gion that benefits from this RR production banin Brazil and the ROW.

Depending on the severity of GM aver-sion in the EU and other countries manifestedin their official government regulations, theROW may choose to ban any presence of cropsand food products with biotech content on itsterritory. In other words, the ROW may banboth RR production and imports, which wouldhave dramatic consequences for productionpatterns in exporting regions as some of themwould have to scale back on their adoptionof RR technology. First, considering the casewhen Brazil does not ban RR production, wesee from table 6 that having no export des-tination for the RR soybeans and products,the United States, Argentina, and Brazil eachproduce both varieties—RR for domestic con-sumption and conventional for export to theROW. The adoption rate for RR technology inthe United States is 65%, in Brazil 51%, and inArgentina 30%. ROW consumers experiencevery large losses in excess of $1 billion becauseof the unavailability of the cheaper RR vari-ety. This fact drives the overall welfare lossfor the ROW as a result of the introduction ofRR technology. Other regions gain despite thewelfare losses by producers, and the world’swelfare improves, albeit slightly. Adding anRR production ban in Brazil changes the char-acteristics of the equilibrium only to the ex-tent that Brazil experiences a loss of consumersurplus because of consumption of more ex-pensive conventional products and an increasein the producer surplus due to specialization

Table 7. Base and Alternative Values of Parameters Used in Sensitivity Analysis

Parameter Base Value Alternative Value 1 Alternative Value 2

εUU {−0.4, −0.4, −0.4, −0.4} Base value × 12 Base value × 2

� {0.8, 1.0, 0.8, 0.6} Base value × 12 Base value × 2

� {0, 0, 0, 0} Base value + 0.02 – 0.5 Base value × 2

3 Base values × 1 13

k 1.05 Base value − 0.025 Base value + 0.025ε00 −4.5 Base value × 2

3 Base values × 1 13

(see the last set of results in table 6). How-ever, unlike the ROW, Brazil’s overall welfareimproves as compared to the preinnovationbenchmark. Welfare comparisons between theunregulated and ban scenarios show that all re-gions lose overall as a result of the combinedproduction and import ban in the ROW nomatter whether Brazil introduces the RR pro-duction ban or not. The only benefiting par-ties are consumers in unregulated regions andROW producers. But note that, given an RRproduction and import ban in ROW, the intro-duction of a production ban in Brazil actuallyimproves world welfare, again an instance of asecond-best result.

Sensitivity Analysis

All parameters of the model were subjected toextensive sensitivity analysis, some of which isreported here. The model’s sensitivity to thesegregation cost parameter � was discussedin the previous section, where we consideredthree alternative values for this parameter. Weanalyzed the sensitivity to the share of indif-ferent demand ; the coefficient of total de-mand increase due to conventional and RRprice equalization k; the own-price elasticity ofconventional demand ε00; the own-price elas-ticities of demand for nonsegregated soybeans,soybean oil, and soybean meal εUU; the elastic-ity of land supply with respect to soybean price� ; and the coefficient of yield increase due tothe RR technology �. Parameter values usedin the analysis are provided in table 7.

Sensitivity results for the aforementionedparameters are illustrated in table 8. The firstset of results shows the effects of halvingand doubling the base values of elasticities of(total) demand for nonsegregated soybeans,soybean oil, and soybean meal. Comparedto the base-values scenario, the only notice-able change is in the distribution of welfaregains between consumers and producers: a

640 August 2005 Amer. J. Agr. Econ.

Tabl

e8.

Mod

el’s

Sens

itiv

ity

toK

eyP

aram

eter

s:W

elfa

reC

hang

esfr

omth

eP

rein

nova

tion

Equ

ilibr

ium

inN

o-L

DP,

No-

Ban

Scen

ario

s(U

nles

sN

oted

Oth

erw

ise)

,w

ith

$13.

2/m

tSeg

rega

tion

Cos

t

Alt

erna

tive

Val

ue1

Bas

eV

alue

Alt

erna

tive

Val

ue2

�C

S�

PS

�W

�C

S�

PS

�W

�C

S�

PS

�W

Reg

ion

�To

tal

Tota

l�

�M

Tota

l�

Tota

lTo

tal

��

MTo

tal

�To

tal

Tota

l�

�M

Tota

l

Par

amet

erεU

U

U.S

.0.

9035

9−1

6078

698

50.

9030

1−8

378

41,

003

0.90

280

−52

785

1,01

3B

R1.

0014

850

198

1.00

112

9020

21.

0010

610

621

1A

R1.

0049

3483

1.00

4057

971.

0037

6710

4R

OW

1.00

231

102

333

1.00

145

138

283

1.00

9115

224

3W

orld

787

2678

61,

599

598

201

784

1,58

451

427

378

51,

571

Par

amet

er�

U.S

.0.

9028

1−4

979

21,

024

0.90

301

−83

784

1,00

30.

8932

0−1

0977

298

3B

R1.

0010

610

521

11.

0011

290

202

1.00

118

7719

5A

R1.

0037

6710

51.

0040

5797

1.00

4248

90R

OW

1.00

9415

224

61.

0014

513

828

31.

0019

112

631

7W

orld

519

275

792

1,58

659

820

178

41,

584

670

143

772

1,58

5P

aram

eter

�U

.S.

0.88

411

−288

770

893

0.90

301

−83

784

1,00

3B

R1.

0014

6−6

140

1.00

112

9020

2A

R1.

0053

356

1.00

4057

97R

OW

1.00

409

5145

91.

0014

513

828

3W

orld

1,01

9−2

4077

01,

548

598

201

784

1,58

4P

aram

eter

kU

.S.

0.90

298

−76

784

1,00

60.

9030

1−8

378

41,

003

0.90

335

−133

785

987

BR

1.00

112

9320

51.

0011

290

202

1.00

131

6419

5A

R1.

0039

5999

1.00

4057

971.

0045

4287

RO

W1.

0013

214

027

21.

0014

513

828

31.

0020

711

432

1W

orld

581

216

784

1,58

159

820

178

41,

584

717

8778

51,

590

Par

amet

er

U.S

.0.

8733

5−1

3276

897

10.

9030

1−8

378

41,

003

0.93

300

−81

801

1,02

0B

R1.

0013

265

197

1.00

112

9020

21.

0011

191

202

AR

1.00

4543

881.

0040

5797

1.00

4058

97R

OW

1.00

2111

513

61.

0014

513

828

31.

0031

713

845

6W

orld

533

9076

81,

391

598

201

784

1,58

476

820

680

11,

775

Par

amet

er

:RR

prod

ucti

onba

nin

the

RO

WU

.S.

1.00

301

−85

673

889

1.00

239

967

592

21.

0018

494

679

957

BR

1.00

114

8920

31.

0081

137

218

1.00

5718

123

8A

R1.

0040

5797

1.00

3085

116

1.00

2311

113

4R

OW

0.00

145

121

267

0.00

277

4131

80.

0044

2−1

4230

0W

orld

600

182

673

1,45

562

627

267

51,

573

706

244

679

1,62

9

Not

e:M

onet

ary

vari

able

sar

em

easu

red

inm

ilU

.S.$

;qua

ntit

ies

are

inm

ilm

t.

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 641

lower elasticity increases consumer benefitsand reduces producer gains, while a higherone has the opposite effect. The innovator-monopolist’s profit remains essentially insen-sitive to variations in εUU. In fact, these resultshold equally for free trade and ban scenarios,and the overall region-level results appear ro-bust. The second set of results in table 8 sum-marizes the adoption and welfare results whenthe elasticity of land supply with respect to soy-bean prices � is halved or doubled. Doubling� works just the opposite of doubling εUU,and the same can be said about halving � ver-sus halving εUU. The innovator-monopolist’sprofit is robust, deviating within 1% of the basevalue. Again, none of the qualitative results ofthe model changes.

The model’s results appear to be more sen-sitive to the change in the RR technology-induced yield increase parameter �. A positiveyield gain is equivalent to an outward sup-ply shift relative to the base-values scenario.Therefore, in the free-trade equilibrium with� = 0.02, all prices are lower, which leads tothe reallocation of welfare gains between con-sumers and producers; while both producersand consumers benefit at the world level fromthe new technology in the base-values sce-nario, producers at the world level lose andconsumers gain when � = 0.02. At the re-gion level, Brazilian and U.S. farmers lose byadopting the RR technology, although overallregion-level results are robust to the increasein the yield parameter.

Parameter measures the relative size ofdemand that is indifferent between conven-tional and RR varieties at specific prices. Nat-urally, lower increases the relative share ofthe worldwide conventional demand and re-duces the share of demand for the RR vari-ety, causing higher conventional and lower RRequilibrium prices relative to the base-valuesscenario. Higher works in the opposite di-rection. The same can be said about parameterε00; given that the total soybean and soybeanoil demands are inelastic, making conventionaldemands less own-price elastic translates intolower cross-price elasticity, which means lessflexibility in the demand system to shift fromconsuming the conventional to the RR vari-ety.23 This relationship makes the model andour conclusions in ban scenarios somewhatsensitive to the values of and ε00 (suggestingthat more work to shed empirical light on these

23 Because the impacts of changes in or ε00 are very similar, intable 8 we omit the explicit sensitivity results for ε00.

parameters may be useful). Table 8 shows thatwhen = 2/3, the ROW does not benefit fromits RR production ban relative to the no-banscenario. In addition (not shown in the table),the ROW benefits from the simultaneous RRproduction bans at home and in Brazil whenε00 = −3.0 or = 1/3, and Brazilian farmerslose under the production ban at home relativeto the no-ban scenario when either parametertakes on a high (absolute) value.

Lastly, the changes in the value of parameterk have some minor quantitative and no qualita-tive effects on the results of the model. A lowerk acts as an inward demand shift that lowers allprices (except for meal) in all equilibria, whilea higher k acts as an outward demand shift thatleads to an increase in soybean and soybean oilprices.

Conclusion

The distinctive feature of our new partial-equilibrium, four-region world trade modelfor the soybean complex is that consumersin the importing region view GM soybeans,and products derived from them, as weaklyinferior substitutes for their conventionalcounterparts. Thus, this article shows thatdifferentiated preferences can be introducedinto a conventional spatial equilibrium modelin a simple yet consistent fashion that permitsstandard welfare calculations. Our modelprovides a close representation of the worldsoybean market by also accounting for the factthat the RR seed is protected by IPRs and soldworldwide by a U.S. firm at a premium andthat producers have to employ a costly seg-regation technology to separate conventionaland biotech products in the supply chain. Themodel is disaggregated just enough to captureindividual behavior of the industry’s mainplayers and allows us to analyze the impact oftheir (possibly different) policies toward GMproducts.

Our analysis shows that in the world withno feasible segregation technology, the long-run equilibrium state of the world after thecost-saving RR technology is introduced isthat of complete worldwide adoption. Thisequilibrium is characterized by lower pricesfor soybeans and soybean products, a con-tinued U.S. lead in world soybean exports,and welfare gains to all regions and all eco-nomic agents (producers, consumers, and theinnovator-monopolist selling RR seed), exceptU.S. farmers. When segregation technology is

642 August 2005 Amer. J. Agr. Econ.

available at a positive cost, absent any gov-ernment production and trade regulations, theUnited States emerges as the only region pro-ducing both RR and conventional soybeans;all other regions specialize in RR production.The introduction of the RR technology leadsto reduced prices for RR products, lower pricesfor producers of the conventional variety, andhigher consumption prices of conventionalproducts. Lower segregation costs reduce theprices of conventional products, increase theprices received by farmers who grow the con-ventional variety, and are associated with moreland allocated to growing conventional soy-beans, which hurts the profits received bythe innovator-monopolist. This result impliesa conflict of interest between the RR-inputsupplier and farmers who benefit from lowersegregation costs. The world in general bene-fits from using the segregation technology atany feasible cost level, as GM-conscious con-sumers realize their right to choose.

The analysis shows that an output subsidyreceived by U.S. farmers, although clearly ben-eficial for them and the region’s consumers, isnevertheless welfare reducing to the UnitedStates as a whole because of the high cost ofthe subsidy. The only region that gains in thissituation is the ROW, but the world in generalcan potentially benefit from this policy as thesubsidy works to correct a less-than-optimaladoption of the RR technology caused by thedistorted RR seed prices established by themonopoly.

When the ROW and Brazil impose produc-tion bans on RR products, the ROW has a clearpotential to benefit from such a ban relativeto the no-ban scenario if segregation costs arenot too low, while in Brazil only farmers wouldbenefit from such regulation. These results are,however, somewhat sensitive to the underlyingassumptions. In particular, this gain is morelikely to hold the higher the size of the con-ventional market and/or the lower the elastic-ity of conventional demands (which indirectlyaffects the market share of interest). Also, itis possible that the ROW can gain relative tothe no-ban scenario when RR production bansare implemented in the two regions simulta-neously, although this result is not observed atbase parameter values. But our analysis alsoshows that, whenever beneficial to the ROW,production bans reduce U.S. welfare, whichjustifies the region’s concerned position withregard to anti-GM regulation. Also, Brazil asa whole never benefits from a unilateral pro-duction ban. The last important result of thispaper is the robust welfare losses to all regions

when an import ban on RR products is intro-duced in the ROW. Overall, all conclusions ofthe model, except for those mentioned above,prove to be robust to some variation in criticalparameter values.

An emerging literature is attempting tocharacterize the economic impact of GMproducts in agriculture. We believe that thisarticle provides an important addition byvalidating some of the theoretical resultshitherto derived in very stylized models, byshowing how the critical inferior-substituteproperty of GM products can be empiricallyimplemented in a simple yet coherent fashion,by explicitly modeling the IPRs dimensions ofthe problem, by investigating the interactionof new technology adoption with existing pricesupports programs, and by evaluating somecritical GM-related policies being consideredby some importing and exporting countries.In particular, this article is the first to providequantitative estimates of the welfare effectsof the large-scale adoption of an existingGM crop in a differentiated-product, open-economy context. As such, the results of thisarticle provide a range of important insightsinto the channels through which the benefits(and unintended costs) of the current gener-ation of GM products arise, and explain thepossible implications of existing and pendingpolicies pursued by the countries across theworld in response to the challenges posed bythe dawn of biotechnology in agriculture.

[Received June 2003;accepted October 2004.]

References

Alston, J.M., and W.J. Martin. “Reversal of Fortune:Immiserizing Technical Change in Agricul-ture.” American Journal of Agricultural Eco-nomics 77(1995):251–59.

Bender, K., L. Hill, B. Wenzel, and R. Hornbaker.“Alternative Market Channels for SpecialtyCorn and Soybeans.” Dept. Agr. Econ. AE-4726, University of Illinois, 1999.

Bhagwati, J. “Distortions and Immiserizing Growth:A Generalization.” Review of Economic Stud-ies 35(1968):481–85.

Bullock, D., and M. Desquilbet. “The Economics ofNon-GMO Segregation and Identity Preserva-tion.” Food Policy 27(2002):81–99.

Desquilbet, M., and D. Bullock. “Who Pays theCosts of Non-GMO Segregation and IdentityPreservation?” Paper presented at the 2001IATRC meeting, Tucson, AZ, 14 December2001.

Sobolevsky, Moschini, and Lapan GM Crops, Product Differentiation, and Trade 643

Eurobarometer. “Europeans, Science and Technol-ogy.” Eurobarometer Standard Report 55.2.December 2001. Available at http://europe.eu.int/comm/research/press/2001/pr0612en-report.pdf (accessed November 2002).

Falck-Zepeda, J.B., G. Traxler, and R.G. Nelson.“Surplus Distribution from the Introduction ofa Biotechnology Innovation.” American Jour-nal of Agricultural Economics 82(2000):360–69.

Fulton, M., and K. Giannakas. “Inserting GM Prod-ucts into the Food Chain: The Market and Wel-fare Effects of Different Labeling and Regula-tory Regimes.” American Journal of Agricul-tural Economics 86(2004):360–69.

Furtan, W.H., R.S. Gray, and J.J. Holzman. “TheOptimal Time to License a Biotech ‘Lemon’.”Contemporary Economic Policy 21(2003):433–44.

Gaisford, J.D., R. Tarvydas, J.E. Hobbs, and W.A.Kerr. “Biotechnology Piracy: Rethinking theInternational Protection of Intellectual Prop-erty.” Canadian Journal of Agricultural Eco-nomics 50(2002):15–34.

Giannakas, K. “Infringement of Intellectual Prop-erty Rights: Causes and Consequences.”American Journal of Agricultural Economics84(2002):482–94.

Goldsmith, P., G. Ramos, and C. Steiger. “A Tale ofTwo Businesses: Property Rights and the Mar-keting of Agricultural Biotechnology.” ChoicesThird Quarter (2003):25–30.

James, C. Global Review of Commercialized Trans-genic Crops: 2002 (Preview). ISAAA Brief No.27, Ithaca, NY, 2002.

Lapan, H., and G. Moschini. “GMO Labeling andTrade: Consumer Protection or (Just) Protec-tionism?” Paper presented at the 2001 IATRCmeeting, Tucson, AZ, 14 December 2001.

——. “Innovation and Trade with Endogenous Mar-ket Failure: The Case of Genetically ModifiedProducts.” American Journal of AgriculturalEconomics 86(2004):634–48.

Lence, S., and D. Hayes. “Response to an Asymmet-ric Demand for Attributes: An Application tothe Market for Genetically Modified Crops.”Working paper 01-MWP 5, Center for Agri-cultural and Rural Development, Iowa StateUniversity, 2001.

Lin, W., W. Chambers, and J. Harwood. “Biotech-nology: U.S. Grain Handlers Look Ahead.”Agricultural Outlook. Washington DC: U.S.Department of Agriculture, 2000, pp. 29–34.

Meilke, K.D., and Michael Swidinsky. “An Evalu-ation of Oilseed Trade Liberalization.” Agri-culture and Agri-Food Canada, Report, TradeResearch Series, Ottawa, Canada, 1998.

Moschini, G. “Biotech—Who Wins? EconomicBenefits and Costs of Biotechnology Innova-tions in Agriculture.” Estey Centre Journal ofInternational Law and Trade Policy 2(2001):93–117.

Moschini, G., and H. Lapan. “Intellectual PropertyRights and the Welfare Effects of AgriculturalR&D.” American Journal of Agricultural Eco-nomics 79(1997):1229–42.

Moschini, G., H. Lapan, and A. Sobolevsky.“Roundup Ready Soybeans and Welfare Ef-fects in the Soybean Complex.” Agribusiness16(2000):33–55.

Mussa, M., and S. Rosen, “Monopoly and Prod-uct Quality.” Journal of Economics Theory18(1978):301–17.

Nelson, G.C., ed. Genetically Modified Organisms inAgriculture: Economics and Politics. London:Academic Press, 2001.

Nielsen, C., and K. Anderson. “GMOs, Trade Policy,and Welfare in Rich and Poor Countries.” InK. Maskus and J. Wilson, eds. Quantifying theImpact of Technical Barriers to Trade: Can It BeDone?, Chapter 6, Ann Arbor, MI: Universityof Michigan Press, 2001.

Nielsen, C., K. Thierfelder, and S. Robinson. “Con-sumer Attitudes towards Genetically ModifiedFoods: The Modelling of Preference Changes.”Working paper 1/2001, Danish Institute ofAgricultural and Fisheries Economics, Copen-hagen, 2001.

Pardey, P.G., ed. The Future of Food—Biotechnology Markets and Policies in anInternational Setting. Washington DC: IFPRI,2001.

Samuelson, P.A. “Spatial Price Equilibrium andLinear Programming.” American EconomicReview 42(1952):283–303.

Schnepf, R., E. Dohlman, and C. Bolling. “Agricul-ture in Brazil and Argentina: Developmentsand Prospects for Major Field Crops.” Wash-ington DC: U.S. Department of Agriculture,Agriculture and Trade Rep. WRS-01-3, 2001.

Sheldon, I. “Regulation of Biotechnology: Will WeEver ‘Freely’ Trade GMOs?” European Reviewof Agricultural Economics 29(2002):155–76.

Sobolevsky, A., G. Moschini, and H. Lapan. “Ge-netically Modified Crop Innovations and Prod-uct Differentiation: Trade and Welfare Effectsin the Soybean Complex.” Working paper 02-WP 319, Center for Agricultural and Rural De-velopment, Iowa State University, 2002.

Takayama, T., and G.G. Judge. Spatial and Tempo-ral Price and Allocation Models. Amsterdam:North-Holland, 1971.

Takayama, T., and W.C. Labys. “Spatial EquilibriumAnalysis.” In I.E.S. Mills and P. Nijkamp, eds.

644 August 2005 Amer. J. Agr. Econ.

Handbook of Regional and Urban Economics,Vol. 1. Amsterdam: Elsevier Science Publish-ers, 1986.

Tirole, J. The Theory of Industrial Organization.Cambridge, MA: The MIT Press, 1988.

U.S. Department of Agriculture. NonrecourseMarketing Assistance Loans and Loan Defi-ciency Payments. Washington DC: Farm Ser-vice Agency, 1998.

——. Production, Supply and Distribution (PS&D)Database: By Country, Commodity Supply, andUse Time Series. Washington DC: EconomicResearch Service, 2002.

U.S. Government Accounting Office. Informa-tion on Prices of Genetically Modified Seedsin the United States and Argentina. ReportGAO/RCED/NSIAD-00-55, Washington DC,January 2000.