trade costs and the gains from trade in crop agriculture

TRADE COSTS AND THE GAINS FROM TRADE IN CROP

AGRICULTURE

JEFFREY J. REIMER AND MAN LI

We develop a simulation model of world crop markets that is based upon Ricardian comparativeadvantage. We apply the model to twenty-three countries and provide measures of the degree ofglobalization in this sector, the gains from trade, and the elasticity of trade volumes to trade costs. Thedistribution of the gains from trade across countries is uneven due to important differences in opennessto imports, productivity, and other factors, some of which appear to be related to a country’s level ofdevelopment. Distance limits the extent by which changes in one country are transmitted to others.

Key words: geography, grains, trade costs, trade liberalization.

JEL codes: F18, Q17, Q54.

Although the theoretical case for the gainsfrom trade is well established, we know lit-tle about the empirical magnitude of the gainsfrom trade and the potential gains from tradeas trade costs fall, particularly for the worldcrop sector. In this study we develop a globalsimulation model to answer these questionswith respect to international trade in grains andoilseeds.

Examination of this sector is timely given therecent upheaval in commodity markets. During2007–2008, export quantitative restrictions orexport taxes were adopted in many exportingcountries, including Argentina, China, Egypt,India,Indonesia,Russia,Ukraine,andVietnam(Trostle 2008).This exacerbated the thinness ofa world crop market that is already highly insu-lated. Global average bound tariffs (the maxi-mum rate of tariff allowed by the World TradeOrganization for imports from any memberstate) are roughly double those in other sectorsof the economy (Effland et al. 2008). Prospectsfor further multilateral trade liberalization areuncertain (McClanahan 2008). Since countriesare often quite hesitant to make concessionsin exchange for trade liberalization in othercountries, it may help to illuminate basic factsabout this sector and get new perspectives on

Jeffrey J. Reimer and Man Li are an assistant professor and agraduate research assistant, respectively, in the Department ofAgricultural and Resource Economics, Oregon State University.The authors thank participants at various seminars and threeanonymous reviewers for insightful comments.

the size and distribution of the gains from trade,by country.

In this study we propose a new conceptualframework for international crop agriculturebased on the class of Ricardian trade mod-els developed by Eaton and Kortum (2002;hereafter EK) and adapted by Bernard et al.(2003) and Alvarez and Lucas (2005). Unlikethe textbook Ricardian model, in which twocountries each specialize in one of two goods,these authors model the goods sector as a con-tinuum, with multiple countries specializing insections of this continuum according to com-parative advantage. While the aforementionedstudies focus on labor usage in manufactur-ing, we adapt the model to land usage incrop agriculture. Specialization is determinedby productivity, land endowments, and thebilateral costs of trade. In many respects, themodel applies more naturally in this setting.Productivity is determined through a randomdraw from country-specific distributions,whichtranslates nicely to crop yields. Each coun-try has a chance of being a low-cost supplierdepending on whether it has a bumper cropor crop failure in a given year. Because ofthe availability of comparable internationalyield data, we can introduce some innova-tive ways of estimating parameters of themodel.

The resulting framework provides an alter-native to computable general equilibrium(CGE) models (e.g., Hertel 1997). The char-acterization of the global trading equilibrium

Amer. J. Agr. Econ. 1–16; doi: 10.1093/ajae/aaq046Received November 2008; accepted April 2010

© The Author (2010). Published by Oxford University Press on behalf of the Agricultural and Applied EconomicsAssociation. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org

American Journal of Agricultural Economics Advance Access published June 10, 2010 by guest on June 11, 2010

ajae.oxfordjournals.orgD

ownloaded from

2 Amer. J. Agr. Econ.

differs in important ways. For example, coun-tries specialize in a subset of homogeneouscrops as determined by their productivity dis-tributions and the costs of trading with foreignmarkets. By contrast,most CGE models invokespecialization through differentiation by coun-try, following Armington (1969). Relative tothat approach, this model has greater flexibilityin the extent to which trade patterns adjust inresponse to shocks to the system.

In this way the framework is more likea spatial equilibrium model, which allocatestrade flows on the basis of lowest possibletransportation cost (Abbott, Paarlberg, andPatterson 1988). Unlike spatial equilibriummodels, however, our framework is based ona gravity model, which does a better job atreplicating and predicting trade flows (e.g.,Disdier, Fontagné, and Mimouni 2008; Grantand Lambert 2008). Our gravity model is dif-ferent from theirs, in turn, in that it is derivedfrom the trade model used for the simulationanalysis. It incorporates structural parametersfrom the yield distributions that govern spe-cialization. Furthermore, gravity studies tendto focus only on what gets traded. We allowthe size of the sector to be endogenous andaccount for the amount of trade relative tooverall consumption, i.e., the extent of “homebias” in consumption that is an important areaof research within the international economicsliterature.1

While our approach has many strengths, weare unable to account for every potential chan-nel by which changes in trade costs affect wel-fare. For example, we do not account for scaleeffects, varietal gains, or pass-through effects,all of which are associated with imperfect com-petition. We do not account for the dynamicgains from trade that may arise from new pat-terns of investment (Baldwin and Venables1995). We believe that our static, competitivemodel captures much of the gains from trade,however, since trade costs appear to be par-ticularly high for this sector, at least whencompared with what EK find for the manufac-turing sector. The credibility of our results isenhanced by the fact that we econometricallyestimate the parameters of the model and allowthe data an opportunity to accept or reject anumber of hypotheses.

1 The term“home bias”has become commonplace due to studiesby McCallum (1995), among others. He finds that trade betweentwo Canadian provinces is more than twenty times larger than tradebetween a Canadian province and a U.S. state, all else the same.

Our approach to measuring trade costsdiffers from studies that use index num-bers to measure trade restrictiveness for indi-vidual countries (e.g., Anderson and Neary1996; Niroomand and Nissan 1997; Lloyd andMacLaren 2002; Reimer and Kang 2010). Bycontrast, we are interested in the bilateral costsof trade and how a change of some sort inone country affects other countries. For exam-ple, if a large country experiences techno-logical change, the greatest effect in relativeterms may not be on itself, but on smallernations with whom it trades. This point maynot arise when trade restrictiveness indexesare calculated for individual countries, one at atime.

While our approach can be used to evalu-ate specific policy initiatives, in this study weconsider broad definitions of trade costs insix counterfactual scenarios to illustrate basicpoints about the model and the world trad-ing system. We calculate bilateral iceberg tradecosts using a structural gravity model, whichmeans that trade costs can include any fac-tor that restricts trade relative to what wouldoccur in a completely frictionless trading sys-tem (Anderson and van Wincoop 2004). Tradecosts can therefore include tariff and non-tariff policy barriers, freight costs, the timecost of shipping, information costs, contractenforcement costs,currency costs,and legal andregulatory costs.2

In the first two simulations, we compareobserved global trade volumes with those thatwould occur under zero trade cost and autar-kic equilibria. We find far less trade than onemight expect given the large degree of varia-tion in crop prices, land rental rates, and cropyields across countries (the coefficient of vari-ation in yields across countries averages 47%for our sample of crops). We show that theaggregate international crops market is quitethin, with observed trade volumes only one-fifteenth of that which would occur under cost-less trade.The volume of trade, in turn, is elasticto small changes in trade costs. A 1% reduc-tion in aggregate trade costs increases worldtrade volumes 2.0–2.5%. We compare ourresults for alternative parameterizations of themodel.

2 Iceberg trade costs are calculated as an ad valorem tax equiv-alent, implying that pricier goods are costlier to trade. In practice,trade costs are a mixture of per unit (specific) and ad valoremrates. The role of per unit trade costs may matter most at highlydisaggregated product levels (Irarrazabal, Moxnes, and Opromolla2010).

by guest on June 11, 2010ajae.oxfordjournals.org

nloaded from

Reimer and Li Trade Costs 3

In the third and fourth simulations,we distin-guish trade costs that are in principle reducible,such as tariffs, from those that are difficult toreduce, such as transportation costs. We findthat small reductions in reducible trade costscan lead to big welfare gains. This is mainlybecause consumer prices in most countries canpotentially fall a great deal from present levels.We show how the results are affected by vari-ations in how easy it is to bring new croplandinto production.

The final two simulations examine howchanges in supply in one country affect thewelfare of other countries. Distance in partic-ular can greatly reduce the extent by which anevent in one country is transmitted to others.The distribution of the gains from trade alsodepends on characteristics such as a country’scompetitiveness in the crop sector and its tradecosts with major trading partners. Many of thephenomena that we observe appear to have arelationship with national per capita income.The countries that are most productive at pro-ducing grains and oilseeds and that export ahigher than average share of production aremore likely to be at the high end of the globalincome distribution. The countries most opento imports are also more likely to be at the highend of the global income distribution.

Conceptual Framework

In the classical Ricardian model, goods arecompetitively produced with a single fac-tor of production using constant-returns-to-scale technologies. Dornbusch, Fisher, andSamuelson (1977) extend the model to multiplegoods,showing that the goods can be ranked ona continuum according to the relative amountof labor required to produce them. The twocountries generally have a comparative advan-tage on opposite ends of this continuum. Thisapproach does not work for more than twocountries, however. In a pathbreaking study,EK extend the model to more than two coun-tries while at the same time accounting forbilateral trade. Productivity is determined bya draw from a probability distribution, witheach country having some chance of produc-ing at a lower cost than any other country.This probabilistic representation assigns com-parative advantage in the context of manygoods and countries and determines the frac-tion of a country’s goods that get importedand exported. Since there are bilateral trade

costs, countries may produce goods for whichthey are otherwise not the world’s lowest-costproducer.

In our application of this approach, we iden-tify land instead of labor as the key factor ofproduction. Therefore, the wage in the stan-dard framework becomes the land rental rate.3This gives a meaningful interpretation of therandom productivity shocks. In our version,they arise from the weather-induced random-ness of agricultural production, as well as rela-tively permanent differences in weather, soilquality, or technology across countries. Withland as the principal factor of production, pro-ductivity is defined as crop output per area ofland (yield). Since yield data are readily avail-able,we can directly estimate the parameters ofthis distribution. We discuss other differencesbelow.

There are N countries indexed alternativelyby i and n. Land used for crop production isdenoted Li. The yield of crop j in country i iszi(j) and the rental rate of cropland is wi (inreality, this corresponds to the entire bundle ofresources associated with land). With constantreturns to scale, the cost of producing j in i iswi/zi(j).

To model bilateral trade, the export coun-try is denoted i and the import country isdenoted n,with i = n when a country buys fromhome. Trade costs follow the iceberg assump-tion, implying that delivery of one unit tocountry n requires dni units produced in i.4 Thecrop sector is modeled as a continuum indexedon the unit interval j ∈ [0, 1].The representativebuyer in country n has symmetric preferencesover the different crops, with utility given by:

(1) Un =[∫ 1

0qn(j)(σ−1)/σ dj

]σ/(σ−1)

where qn(j) is the quantity purchased and σ > 0is the elasticity of substitution among crops.Country n’s representative buyer maximizesequation (1) subject to spending constraint Xn.In a perfectly competitive market, the price

3 This idea has been productively employed by Donaldson(2009), although he does not use yield data to estimate parameters,as we do.

4 A trade cost is anything that restricts imports or exports of agood. If strict quarantine measures restrict grain imports, for exam-ple, then this will be captured as a trade cost. However, if freightcosts fall, yet imports cannot increase due to very strict quarantinemeasures, the fall in freight costs would not be counted as a fall intrade costs.

nloaded from

that n pays for crop j from country i is:

(2) pni(j) = dniwi

zi(j).

Since users in country n seek to buy crop j fromthe cheapest source, they pay:

pn(j) = min{pn1(j), pn2(j), pn3(j),

. . . , pnN(j)}(3)

where N is the total number of countries.We now let yields be the random variable

Zi(j) in place of the constant zi(j). Since pricedepends on Zi(j) it is also a random variable,denoted Pni(j). Therefore, n chooses the min-imum from a sequence of random variablesinvolving Zi(j). In this particular context, theFréchet Type II extreme value distribution isan appropriate distribution for Zi(j):

(4) Fi(z) = Pr[Zi ≤ z] = exp(−Tiz−θ

)where Ti > 0, θ > 1, and z > 0.5 Higher Timeans higher average crop yields in i. Lower θmeans that yield distributions are broader anda country’s relative strengths and weaknessesin the productivity of different crops are morepronounced. As strengths and weaknessesbecome magnified, comparative advantageexerts a greater force, and high-productivitycrops will be exported, while low productivitycrops will be imported.

Due to the continuum assumption in con-junction with identical cost and demand struc-tures, the index for crops (j) can be dropped,and we focus on the crop sector in the aggre-gate. As shown in EK, the probability thatcountry i supplies country n at the lowestprice is:

Pr[Pni(j) ≤ min {Pns(j); s �= i}](5)

= Ti(widni)−θ∑N

i=1 Ti(widni)−θ.

equation (5) says that n’s probability of buy-ing from i is increased by higher average yieldsin i (Ti), lowered by trade costs between nand i (dni), and lowered by land rental ratesin i (wi). equation (5) can also be related tothe share of n’s spending on crops from i.

5 See Billingsley (1986) for a discussion of this distribution.

Let Xni be n’s spending on crops from coun-try i, with i = n when a country buys fromhome. Summing over all sources of supplygives:

∑Ni=1(Xni/Xn) = 1. Due to the continuum

assumption, the share of n’s spending on cropsfrom i is equal to equation (5), which impliesthat:

(6)Xni

Xn= Ti(widni)

−θ∑Ni=1 Ti(widni)−θ

equation (6) relates trade shares back to theyield parameters (Ti and θ), bilateral tradecosts (dni), and land rental rates (wi). Theprice index for country n can be derivedusing the moment-generating function for theextreme value distribution (EK). The result is:

Pn =[�

(θ + 1 − σ

)]1/1−σ

N∑i=1

Ti(widni)−θ

]−1/θ

where � is the Gamma function used to expresscertain types of definite integrals.6 Price indexPn relates the overall prices paid in country nback to the yield distributions, trade costs, andland rental rates.

The total domestic product derived fromcropland equals the sum of country i’s world-wide crop sales: wiLi = ∑N

n=1 Xni. Alterna-tively, this is land value added in the cropsector. There is also a second, land-based non-crop agricultural sector, denoted YO

i . It is anuméraire good and remains fixed in all coun-terfactual simulations. Our model thereforecaptures all land value added, similar to howEK capture all labor value added in theirmodel. Overall crop spending in country nis given by Xn = α(wnLn + YO

n ), where α iscrops’ fixed share of total spending. Using thisand equation (6), the cropland market clearingcondition can be calculated to be:

wiLi =N∑

(Ti(widni)

−θ∑Ni=1 Ti(widni)−θ

× [α(wnLn + YO

6 A useful reference is Johnson and Kotz (1970). The derivationof all equations in the model is available upon request from theauthors.

nloaded from

The structure of equation (8) allows for twopolar cases regarding land mobility across sec-tors. When land is immobile across sectors,equations (6), (7), and (8) are solved simul-taneously for trade shares, crop prices, andland rental rates. Land rental rates, total agri-cultural income, and crop spending all adjust.When land is mobile across sectors, however,equation (8) is solved for equilibrium croplandvalues only after equations (6) and (7) havebeen solved for trade shares and prices. Landrental rates, total agricultural income, and cropspending are all fixed.

Equations (6), (7), and (8) are comprised inour two-sector model of the agricultural sector,but with all detail and focus on the crop sector.Data to implement the model come from theGlobalTradeAnalysis Project (GTAP) and theUnited Nations Food and Agricultural Organi-zation (FAO). The model is calibrated to thefollowing categories in the GTAP database:paddy rice, wheat, cereal grains not elsewhereclassified, and oil seeds. GTAP factor mar-ket data show that land not used in cropsectors tends to be used in other food and agri-cultural sectors (Dimaranan and McDougall2007). According to our data, the share that thecrop sector has of economy-wide GDP is 2.2%in the average country. The share of the cropsector in land value added is 23.4% for the aver-age country, and ranges from a low of 12.3%(Greece) to a high of 44.1% (Argentina).

Unlike EK, who model intermediate inputs,we assume that all crop production is used infinal consumption.7 Of course, in reality somecrops are grown for seed. However, the useof crops as input to own production is minorcompared with inputs to manufactures. Forexample, according to GTAP data, the sharethat rice has in the cost of its own produc-tion is only 0.07, while the corresponding inputshares for electronic equipment and chemicalproducts are 0.28 and 0.31, respectively.

A major benefit of excluding intermediateinputs is that the model is much simplifiedand we can develop new ways of estimatingcertain parameters. For example, we can usethe model itself to solve for base land rentalrates (wi). This frees us to directly estimateparameters such as Ti and θ instead of inferringthem through indirect means. Before we turn tothese issues, however, we consider estimationof trade costs.

7 This is equivalent to assuming that β equals 1 in EK’s model.

Estimation of Trade Costs

According to the GTAP data,average spendingon domestic crops as a share of total spend-ing on crops is 88.6%.8 This is clear evidenceof so-called home bias in consumption in thetrade literature. Partly since we account for this,our gravity approach to trade costs ends upbeing somewhat different than other gravitystudies. Using equation (6), a trade equation,we follow EK and normalize (Xni/Xn) by thehome sales of a buyer (Xnn/Xn) to get:

(9)Xni

Xnn= Ti(widni)

Tnw−θn

)−θ

d−θni .

Now take the log:

(10) ln

Tn− θ ln

wn− θ ln dni.

To make this more useful, we introduce anexpression, Si ≡ ln Ti − θ ln wi, that is produc-tivity adjusted for costs. It could be considereda measure of competitiveness. We substitute Siinto equation (10) to get:

(11) ln

)= −θ ln dni + Si − Sn.

When we go to estimate equation (11), theSi are captured by way of dummies. Since wecannot observe dni, we estimate this using vari-ables typically employed in gravity equations.Distance is accounted for by using six dummyvariables representing different intervals ofgreat-circle distance between capitals. Theassociated coefficients are dk (k = 1, . . . , 6),where d1 is associated with a distance of 375miles or less, d2 is associated with a distance of375 to 750 miles, and so on (a spline approach).We also account for whether two countriesshare a border (b), share membership in atrade agreement (eh), and have a commonlanguage (l). Finally, we include an overall des-tination effect (mn) that proxies for opennessto imports, i.e., trade costs that are more likelyto be controllable (as opposed to the geo-graphic trade costs with more permanence).Substituting these in for ln dni in equation (11)

8 This is the share if the potential source countries include onlythe twenty-three countries of our sample. The share is 85.5% if thepotential source countries include all countries of the world.

nloaded from

Table 1. Bilateral Trade Equation

Description Coefficient Estimate p-value Coefficient Estimate p-value

Dist [0,375] −θd1 −5.52 <0.01Dist [375,750] −θd2 −5.86 <0.01Dist [750,1500] −θd3 −7.03 <0.01Dist [1500,3000] −θd4 −8.20 <0.01Dist [3000,6000] −θd5 −9.96 <0.01Dist [6000,max] −θd6 −10.26 <0.01Border −θb 0.38 0.38Language −θ l 0.98 <0.01NAFTA −θe1 1.48 0.27EU −θe2 1.41 0.02Mercosur −θe3 −0.81 0.36Argentina S1 3.93 <0.01 −θm1 2.70 <0.01Australia S2 1.82 <0.01 −θm2 2.27 <0.01Brazil S3 3.23 <0.01 −θm3 2.63 <0.01Bulgaria S4 −1.22 <0.01 −θm4 −4.05 <0.01China S5 2.51 <0.01 −θm5 1.73 <0.01Ethiopia S6 0.90 0.01 −θm6 1.87 <0.01France S7 1.92 <0.01 −θm7 3.23 <0.01Greece S8 −2.62 <0.01 −θm8 −0.99 0.07Hungary S9 −1.54 <0.01 −θm9 −1.18 0.03Italy S10 −1.30 <0.01 −θm10 0.03 0.96Japan S11 −2.21 <0.01 −θm11 −0.86 0.10Mexico S12 −0.39 0.28 −θm12 −0.70 0.20Morocco S13 −0.65 0.07 −θm13 0.59 0.27Peru S14 −3.22 <0.01 −θm14 −3.55 <0.01Romania S15 −1.35 <0.01 −θm15 −2.29 <0.01Russia S16 0.00 1.00 −θm16 −0.75 0.16South Africa S17 0.47 0.19 −θm17 1.70 <0.01Spain S18 −1.24 <0.01 −θm18 1.11 0.04Turkey S19 0.02 0.95 −θm19 0.85 0.11Ukraine S20 0.51 0.16 −θm20 −2.52 <0.01United States S21 5.42 <0.01 −θm21 5.88 <0.01Uruguay S22 −1.92 <0.01 −θm22 −3.42 <0.01Zimbabwe S23 −3.06 <0.01 −θm23 −4.29 <0.01

Note: Estimated by feasible generalized least squares with 506 observations. Adjusted R2 is 0.70.

gives:

)= Si − Sn − θmn − θdk − θb(12)

− θ l − θeh + θξni.

The dummy variable associated with eacheffect in equation (12) is suppressed for nota-tional simplicity. The error term is ξni = ξ 2

ni +ξ 1

ni. ξ 2ni affects two-way international trade

and has variance σ 22 , with ξ 2

ni = ξ 2in. ξ 1

ni affectsone-way international trade and has varianceσ 2

1 . Under this error structure, diagonal ele-ments of the variance-covariance matrix areE(ξniξni) = σ 2

1 + σ 22 , while certain off-diagonal

elements are E(ξniξin) = σ 22 . This allows for

“reciprocity” in geographic barriers, i.e., forthe possibility that the disturbance concerning

shipments from n to i is positively correlated tothe disturbance concerning shipments from i ton.To avoid the dummy variable trap,we impose∑

Si = 0,∑

mn = 0, and no overall intercept.In our estimation, we work with a cross

section for 2001, since that is the only yearfor which complete data (including fully recon-ciled trade flows) are available for our analysisas a whole. Data on 2001 bilateral crop pur-chases (Xni) for twenty-three countries arefrom the GTAP database (Dimaranan andMcDougall 2007). The GTAP data have thekey advantage of being fully reconciled acrosseach exporter i and importer n. Among thesecountries, imports from the other twenty-twocountries as a share of total imports are 76%on average.

Table 1 reports the results of estimatingequation (12) with generalized least squares for

nloaded from

506 observations.The fit is good,as the adjustedR2 is 0.70 and most of the coefficients arestatistically nonzero at the 1% level.The coeffi-cients in the upper part of table 1 indicate howdistance, border, language, and trade agree-ments affect trade volumes. The coefficientson the distance effects are particularly large.For example, the coefficient on distances lessthan 375 miles,−5.52, is larger in absolute valuethan any nondistance coefficient. The negativesigns and successively larger magnitudes on thedistance dummies suggest that freight costs,and possibly other aspects of transport costs,may be a particularly important impedimentto trade in crop markets.9

The coefficients on border, language, theNorth American Free Trade Agreement(NAFTA), and the European Union are allpositive, which implies that these reduce tradecosts, as expected. In looking at the country-specific effects, the countries most open toimports are the United States and France, with−θmn estimates of 5.88 and 3.23, respectively.The countries least open to imports are Zim-babwe and Bulgaria, with estimates of −4.29and −4.05, respectively.

The lower-left portion of table 1 reportsour estimates of competitiveness (S), thatis, productivity adjusted for input costs. TheUnited States is the most competitive country(5.42), followed by Argentina, another impor-tant exporter (3.93). Peru and Zimbabwe arethe least competitive, at −3.22 and −3.06,respectively.

Some country-specific results are plotted infigures 1 and 2. Figure 1 plots estimated open-ness to imports (−θmn) against per capita GDP,and figure 2 plots competitiveness against percapita GDP. Rich countries appear to be some-what more open to imports, which could reflecta variety of factors: better infrastructure andrisk-handling institutions or lower tariffs andnontariff barriers. Rich countries also appearto be more competitive, reflecting factors suchas better access to new technologies (figure 2).On the other hand, both of these relationshipsare somewhat dependent on extreme points,such as Zimbabwe on the low end, and theUnited States on the high end. Ideally wewould have a larger sample to verify the abovepoints, but we have already maximized sample

9 Other aspects of transport costs might include insurance, hold-ing costs for goods in transit, inventory cost due to buffering thevariability of delivery dates, and preparation costs associated withshipment size (Anderson and van Wincoop 2004).

6 7 8 9 10 11 12

(–qm

Per capita GDP, 2001, natural log

Zimbabwe

Figure 1. Scatterplot of estimated opennessand per capita GDP

Zimbabwe

6 7 8 9 10 11 12

Figure 2. Scatterplot of estimated competi-tiveness and per capita GDP

size with respect to constraints imposed byother parts of our empirical approach.

Estimation of Remaining Parameters

The above regression provides us with baselineestimates of the dθ

ni parameters. We developthree alternative approaches to estimating theθ and Ti parameters. The first two approachesinvolve directly estimating the parameters ofequation (4), a yield equation, using yield dataon multiple crops.10 Approach 1 is a general-ized method of moments (GMM) technique.

10 An anonymous reviewer points out that estimating theseparameters from yield data is strictly valid only under autarky. Oth-erwise, the model predicts that each country obtains each crop fromonly one source, and we would not observe the yield in countriesthat import that crop. Given that our estimates suggest that theworld is not far from autarky, however, our approach is probablya good approximation. Note that Donaldson (2009), by contrast,assumes that each crop comes in a unit continuum of varieties, and

nloaded from

Since there are (2N) first- and second-orderraw moments but only (N + 1) unknownparameters (since θ is not indexed by i), theGMM model is overidentified. To deal withthis. we replace the second-order raw momentswith a single second-order central momentequation (a full description is available in a sup-plementary appendix online). Approach 2 is amaximum likelihood estimation (MLE) tech-nique. An empirical likelihood function basedupon equation (4) is:

L(θ , Ti| zij

) =N∏

J∏j=1

θTiz−θ−1ij(13)

× exp(−Tiz−θij ).

Approach 3 differs from the above twoapproaches because estimation of θ is doneseparately from Ti, and equations (6) and (7)are used instead of equation (4). To determineθ ,we follow EK (p. 1753),who link the model todata on prices by first dividing equation (6) bythe analogous expression for the share of coun-try i producers at home and then substitutingin equation (7):

(14)Xni/Xn

Xii/Xi=

(pidni

)−θ

If we take the logarithm of both sides, it isstraightforward to estimate θ using ordinaryleast squares (OLS). EK recommend calculat-ing the logarithm of the right-hand side as:

(pidni

)= max 2 {ln pn(j) − ln pi(j)}(15)

−∑J

j=1{ln pn(j) − ln pi(j)}J

We now consider data requirements.Approaches 1 and 2 are estimated using FAOyield data for twenty-three countries and sixcrops produced in each of these countries(barley, maize, oats, rice, soybeans, and wheat).This set of countries and crops was chosen tomaximize the total number of observationsin the sample. To make the yields of differentcrops comparable, we normalize country i’syield of crop j by j’s worldwide average yield,

thus each region would be expected to produce at least some vari-eties of each crop. He is then unable to use the yield data to estimatethe parameters as done here.

using national acreages as weights. This givesless weight to niche producers of a givencrop. We end up with six comparable obser-vations on yields for each of the twenty-threecountries.11 In Approach 3, the dependentvariable of equation (14) is constructed usingGTAP data on bilateral crop purchases. Theright-hand side of equation (14) is constructedusing FAO data on producer prices for the sixcrops listed above.

We first discuss our estimates of θ , whichmeasures the breadth of the yield distribu-tions and the potential role for comparativeadvantage. The approach 1 (GMM) estimateof θ is 2.83 with standard error of 0.0001. Theapproach 2 (MLE) estimate of θ is 2.52 withstandard error of 0.25. The approach 3 esti-mate of θ is 4.96 with standard error 1.37.These estimates are on the low end of the3.60–12.86 range of estimates that EK reportfor the manufactures sector. However, this isconsistent with their observation that a lowerθ should be expected, since productivity inagriculture is more heterogeneous across coun-tries than productivity in manufacturing (p.1768). There are large differences in temper-ature, precipitation, growing season, and soiltype across the world. Applied agriculturalresearch therefore tends to be specific to par-ticular regions. Furthermore, agricultural inno-vations with broad applicability are not alwaysadopted around the world due to other con-straints, such as imperfectly functioning inputmarkets.

We now turn to discuss our estimates of Ti,which reflect the average level of yields in i.The approach 1 (GMM) estimates range from0.01 for Morocco to 2.24 for France and arereported in table 2 along with estimator stan-dard errors. Note that high estimates of Tido not automatically imply greater competi-tiveness in international markets, since landrental rates may be higher in those coun-tries. The approach 2 (MLE) estimates of Tiare quite similar and have a 0.93 correla-tion with the GMM estimates. The approach3 (OLS) estimates of Ti are taken to bethe average yield by country, normalized byweighted average world yield. Estimates are

11 Note that while we estimate individual crop yield distributionsfor Ethiopia and Ukraine, the GTAP database combines these intoa regional composite: XSS (“Rest of Sub-Saharan Africa”) andXSU (“Rest of former Soviet Union”), respectively (Dimarananand McDougall 2007). For simplicity, we refer to Ethiopia andUkraine in the analysis, but in reality the results below refer tothese regional composites.

nloaded from

Table 2. Key Parameters

Approach 1: θ = 2.83 Approach 3: θ = 4.96

Country Ti(std. err.) wi Li Ti(std. dev.) wi Li

Argentina 0.48 (0.02) 0.19 40,188 0.95 (0.09) 0.45 17,273Australia 0.43 (0.02) 0.39 5,613 0.95 (0.33) 0.69 3,188Brazil 0.24 (0.01) 0.19 41,047 0.76 (0.21) 0.49 16,126Bulgaria 0.27 (0.01) 0.97 7,425 0.85 (0.38) 1.24 5,827China 0.69 (0.03) 0.36 126,262 1.12 (0.35) 0.62 74,055Ethiopia 0.08 (0.00) 0.30 31,239 0.59 (0.40) 0.75 12,625France 2.24 (0.10) 0.67 6,353 1.76 (0.69) 0.76 5,636Greece 0.73 (0.03) 2.25 286 1.14 (0.39) 1.74 370Hungary 0.70 (0.03) 1.52 535 1.15 (0.40) 1.40 578Italy 1.14 (0.05) 1.66 2,790 1.32 (0.31) 1.37 3,372Japan 0.51 (0.02) 1.72 9,691 1.03 (0.35) 1.57 10,615Mexico 0.46 (0.02) 0.87 5,280 1.04 (0.52) 1.09 4,218Morocco 0.01 (0.00) 0.28 8,693 0.38 (0.28) 0.94 2,583Peru 0.12 (0.01) 1.48 697 0.61 (0.25) 1.73 594Romania 0.19 (0.01) 0.89 3,022 0.79 (0.39) 1.25 2,161Russia 0.13 (0.01) 0.48 10,976 0.64 (0.25) 0.91 5,797South Africa 0.17 (0.01) 0.45 2,731 0.69 (0.24) 0.85 1,464Spain 0.67 (0.03) 1.34 2,037 1.11 (0.37) 1.31 2,086Turkey 0.42 (0.02) 0.73 3,818 0.92 (0.17) 0.98 2,850Ukraine 0.26 (0.01) 0.52 59,537 0.82 (0.32) 0.87 35,685USA 0.97 (0.05) 0.15 228,092 1.23 (0.22) 0.35 95,308Uruguay 0.19 (0.01) 1.09 519 0.73 (0.29) 1.38 408Zimbabwe 0.38 (0.02) 2.10 131 1.18 (0.85) 1.92 144

reported in table 2 along with the standarddeviations.

Once we have our Ti and θ estimates byeither approach 1, 2, or 3, we infer the values ofwi and Li from identities in the model. UsingSi ≡ ln Ti − θ ln wi, we calculate the land rentalrate in country i as:12

(16) wi = exp([ln Ti − Si]/θ

Note that since land is the only factor ofproduction, wi can be thought of as returnsto the entire bundle of resources associatedwith a unit of land. Using this estimate, base-line cropland estimates can be solved from theland market identity that relates total croplanddomestic product and land rental rates:

(17) Li =(

N∑n=1

12 EK (p. 1776) use actual wages to estimate wages for the base-line, while using the model itself to solve for intermediate inputprices. We ignore intermediate inputs and use the model itself tosolve for land rental rates.

Values of wi and Li are reported in table 2 forapproaches 1 and 3. The Li are not necessarilyin recognizable units such as hectares or acresgiven how wi is calculated in equation (16).We can nonetheless get a sense of the valid-ity of Li by evaluating whether it gets theranking of countries by cropland area correct.In this respect, the estimates are quite rea-sonable. A simple linear regression of FAOcropland areas on Li yields an R2 of 0.76 and0.80 for approaches 1 and 3, respectively. SinceLi is a function of wi yet has a good fit, thissuggests that wi must also be quite reason-able. Note that approach 3 generates a smaller,and therefore possibly more realistic, range ofwi than does approach 1. This occurs mainlybecause there is less variation in the estimatesof Ti.

The final parameter to estimate is α. We firstcalculate this for individual countries, then finda unified α by taking a GDP-weighted aver-age. We get α = 0.21, with a standard deviationis 0.05.

Counterfactuals are evaluated according toseveral criteria. One is the change in land rentalrates, w′

n − wn, where w′n denotes the new land

rental rate that solves equation (8) under the

nloaded from

Table 3. Counterfactuals 1 and 2: Large Changes in Trade Costs

Baseline to Autarky: Baseline to Zero Gravity:% Change in Net Welfare % Change in Net Welfare

Approach 1 Approach 3 Approach 1 Approach 3

Argentina −2.6 −4.4 50.6 30.6Australia −0.6 −0.5 37.1 20.0Brazil −0.3 −0.2 27.7 15.8Bulgaria −0.0 −0.0 37.5 17.9China −0.2 −0.2 15.9 8.5Ethiopia −0.2 −0.1 28.4 15.3France −2.5 −2.0 28.4 15.3Greece −4.7 −4.4 66.2 32.1Hungary −0.8 −1.4 57.0 30.2Italy −3.3 −2.5 45.3 22.1Japan −0.2 −0.1 39.1 16.5Mexico −3.9 −2.9 37.9 19.0Morocco −0.4 −0.3 43.0 22.5Peru −1.3 −1.2 69.2 32.9Romania −0.3 −0.8 44.5 23.3Russia −0.1 −0.1 35.3 18.1South Africa −1.6 −1.3 43.5 22.8Spain −5.5 −3.6 43.2 21.5Turkey −1.6 −1.8 42.1 21.7Ukraine −0.0 −0.0 22.4 10.4USA −0.5 −0.6 19.1 11.7Uruguay −1.8 −1.4 59.4 29.6Zimbabwe −0.3 −0.3 76.3 37.6

Note:Approach 1 uses θ = 2.83 and Ti from table 2. Approach 3 uses θ = 4.96 and Ti from table 2. The percentage change in world trade is −100, −100, +1,393,and +1,151 in the four scenarios, respectively.

counterfactual simulation. Higher land rentalrates are positively correlated with welfare,since this reflects increases in income on thesupply side. Another criterion is the change incrop prices (P′

n − Pn), where P′n denotes the

new price that solves equation (7) in the coun-terfactual simulation. This price reflects thecosts of purchasing on the demand side andhas a negative relation to welfare. We can alsodefine “welfare” as the real GDP of the sector,denoted Wn = Yn/Pα

n . The percentage changein welfare is:

100 ×[

W ′n

Wn− 1

= 100 ×[

Y ′n

P′αn

Yn− 1

= 100 ×[

w′nLn + YO

wnLn + YOn

− 1]

Note that noncropland value added (YOn )

remains fixed throughout. Further note that formobile land, only the price effect is operativein equation (18). Instead of looking at changes

in land rental rates, we will look at changes incropland area.

Main Results

Counterfactual 1:We have provided some mea-sures of openness to imports at the countrylevel (table 1). We now calculate overall globalopenness in the crop sector. To begin doingso, we first report the results of raising tradebarriers to their autarkic levels. We let dni goto infinity for n �= i such that countries areforced to equate their production and con-sumption. Results are reported in table 3 andare not as extreme as one might expect. Wediscuss the results for approach 1 first. Whileevery country suffers a small loss in net wel-fare, the maximum fall is only 5.5% (Spain),and the median reduction in welfare is only0.6% (Australia). (In Bulgaria and Ukraine,the losses round to zero.) These numbers sug-gest that in terms of net welfare, existing tradecosts are high enough to approximate a state ofautarky. Net welfare change is plotted againstper capita GDP in figure 3. The largest falls in

nloaded from

7 8 9 10 11

Greece

Figure 3. Move to autarky: Scatterplot ofwelfare change and per capita GDP

net welfare tend to occur on the high end of theincome distribution.13 In particular, there are anumber of rich countries that import a largeshare of their consumption but are not espe-cially competitive. Italy, Spain, and Greece arethe most extreme examples. According to theGTAP data, they import an average 31% oftheir grains and oilseeds consumption, whilethe average for all twenty-three countries is11%. Their crop prices go up by an average56% in Counterfactual 1, while the average forall twenty-three countries is a 10% increase(not reported in the table). This result is hardlyspecific to these three countries,as suggested byfigure 3, but is the most pronounced for them.

We also consider Counterfactual 1 under analternative set of parameter estimates,given byapproach 3 in table 2. One of the changes is thatθ = 4.96, which means that yield distributionsare broader and a country’s relative strengthsand weaknesses across crops within the cropsector are less pronounced.The median changeis about the same as before (−0.8% forapproach 3 versus −0.6% for approach 1),although the maximum change is somewhatsmaller (−4.4% for Greece in approach 3versus −5.5% for Spain in approach 1).

Counterfactual 2:The results for Counterfac-tual 1 suggest that existing trade barriers arequite high. We can get a better sense of this byconsidering the opposite experiment: an elim-ination of all trade costs and barriers. In thiscase,we set dni = 1 in equation (2) for all n and i.

13 A note of caution is in order. Since the model concerns only theagricultural sector, the welfare results might in general be of greaterimportance in poorer countries, all else the same. First, agriculturehas a greater share of the economy of poorer countries. Its share ofGDP has a −0.73 correlation with per capita GDP. Second,food hasa greater share of the consumption bundle in poorer countries. Itsshare of final consumption has a −0.81 correlation with per capitaGDP.These points are not taken into account in our welfare results.

This creates a situation of what might be calledzero gravity. The effect on trade volumes ismuch more extreme:The volume of world tradeincreases 1,393% for approach 1. This sug-gests that the volume of trade would increaseapproximately fifteen times if all forms of tradecosts could be eliminated. Every country has asubstantial gain in net welfare, with a medianof 42.1% (Turkey) and a maximum of 76.3%(Zimbabwe). Clearly, international crop mar-kets are very far removed from the frictionlesstrading system envisioned in many traditionalmodels of international trade.

We also consider Counterfactual 2 under analternative set of parameter estimates, givenby approach 3 in table 2. In this case, worldtrade grows by 1,151% as trade costs areeliminated, which is a smaller amount thanunder the parameters of approach 1. Themedian and maximium welfare gains are alsomore subdued, at 21.5% (Spain) and 37.6%(Zimbabwe), respectively. This is primarilybecause θ is higher in approach 3, which meansthat yield distributions are broader. Since acountry’s relative strengths and weaknessesacross crops within the crop sector are less pro-nounced, comparative advantage exerts less ofa force, and trade does not rise by as much asit did in approach 1.

If we repeat the general idea of Counterfac-tuals 1 and 2 for different changes in bilateraltrade costs (reducing each of the 506 dni param-eters by the same proportion), we are able toestimate the elasticity of trade volumes withrespect to trade costs. We find that a 1% reduc-tion in overall trade costs increases world tradevolumes by 2.0–2.5% (based on simulationswhich we lack space to report).

Counterfactual 3: The previous two counter-factuals suggest that “globalization” has notreally come to the world crop sector, if the met-ric is global volume of trade. We now focuson import trade costs (as represented by mn)that are more likely to be reducible. We reducethe import trade costs of each country to thelevel of the country that is most open in thisregard. This happens to be the United States,for which −θm = 5.88. All other trade costs inequation (12) are retained. We report resultsfor approach 1 only.

Changes in net welfare, crop prices, and landrents are in columns 1–3 of table 4. Everycountry experiences an increase in welfare,with a median of 17.9% (Bulgaria). The netwelfare changes mask much larger changesin crop prices and rental rates for land. Allcountries except Argentina, Brazil, and the

nloaded from

Table 4. Counterfactuals 3 and 4: Liberalized Import Policy

Land is fixed by agricultural sector Land is mobile across agricultural sectors

Net Crop Land rental Net Crop CroplandCountry welfare prices rates welfare prices area

Argentina 13.4 57.7 78.3 1.1 −4.7 306.7Australia 5.0 −24.4 −6.2 7.0 −26.4 −33.3Brazil 3.0 14.2 26.6 4.7 −18.7 16.3Bulgaria 17.9 −56.9 −9.4 25.3 −64.0 −63.0China 4.6 −39.5 −29.3 10.7 −37.0 −68.7Ethiopia 6.4 −27.6 −4.0 10.2 −35.6 −43.9France 7.8 −25.2 5.0 12.0 −40.2 −32.6Greece 77.4 −93.7 −24.7 107.6 −96.3 −80.7Hungary 45.0 −59.8 69.4 26.9 −66.0 396.5Italy 32.0 −75.7 −20.6 52.3 −85.1 −65.1Japan 14.4 −69.9 −57.0 29.0 −68.4 −95.3Mexico 25.3 −74.7 −47.1 51.5 −84.8 −96.8Morocco 13.3 −38.3 8.0 18.4 −53.5 −16.4Peru 38.6 −88.6 −74.0 77.6 −92.6 −99.7Romania 40.0 −60.1 57.3 26.1 −65.0 322.6Russia 22.5 −46.1 30.9 19.6 −55.5 72.7South Africa 8.2 −41.2 −20.1 20.4 −56.9 −76.7Spain 25.5 −64.0 1.0 30.9 −70.5 −9.2Turkey 21.2 −58.6 −1.3 28.1 −67.4 −5.3Ukraine 9.8 −35.4 −1.3 15.4 −47.8 −24.7USA 5.9 70.0 72.9 0.0 0.0 175.1Uruguay 43.2 −87.7 −53.1 79.4 −92.9 −96.5Zimbabwe 38.4 −87.6 −59.1 68.3 −90.5 −98.4

Note:Values are percentage changes. In both counterfactuals, import trade costs for each country are lowered to the level of the country that is most open in thisregard (the United States). Approach 1 parameters are used. World trade increases 775% and 1,102% in the left and right scenarios, respectively. Crop pricesrefer to those faced by buyers.

United States experience a fall in crop prices. Inthese three countries,crop prices rise by 57.7%,14.2%, and 70.0%, respectively, because theyhave sufficient competitiveness to respond ina major way to new opportunities in for-eign markets (table 1). Greece, Peru, Uruguay,and Zimbabwe experience more than an 80%drop in crop prices because they were initiallyamong the least open countries (table 1). Thisgives Greece the largest increase in net wel-fare (77.7%), also in part because it is one ofthe least competitive countries (S estimated tobe −2.62) (table 1).

Every country faces a drop in land rentalrates except for nine, including Argentina,Brazil, and the United States. Since their cropsectors expand under falling trade barriers,they might be called natural exporters. Brazil,however, has the smallest increase in net wel-fare (3.0%). The benefit of rising land rentalrates (26.6%) is offset by the higher prices thatconsumers pay (14.2%).

France’s crop prices for buyers fall by 25.2%due to the import liberalization, while landrental rates increase by 5.0% (table 4). Notethat unlike the previous examples, this cannot

be explained by the well-known three-paneldiagram of international trade because it wouldassume that there is one good that is consumedand produced. The consequence is that if thedomestic price falls, then producers must beworse off. In our model, France as a nation canconsume a much broader mix of crops than thecrops specialized in by France’s producers. Asa result, the price index for crop buyers is notthe same as the price index for producers. Theconsumer price index can fall while producersnonetheless receive higher prices for the sub-set of crops that they produce, due to demandfrom foreign markets. The latter effect man-ifests itself in the form of higher land rentalrates.

Counterfactual 4: In previous simulations,land area was held fixed by sector. We nowrepeat Counterfactual 3 but allow land to bemobile across agricultural sectors, with landrental rates held fixed (Li is now endogenous,while wi is now exogenous). Although realitylies somewhere in between these two extremes,of course, this provides a sensitivity check onthe results. Results are provided in columns4–6 of table 4. Since land rental rates no longer

nloaded from

change, all welfare changes arise from changesin crop prices. The United States is the onlycountry to experience no change in welfare(crop prices do not change, since U.S. opennessdoes not change, and land is mobile). For everyother country, there is an increase in welfare,with the median being 25.3% (Bulgaria). Thepositive change in welfare is similar to theresult under fixed cropland area (Counterfac-tual 3). However, in 78% of the countries, thewelfare increase is larger (Counterfactual 4).This is largely due to the ease by which landcan be shifted toward crop production in keyproducing countries. Increases in cropland areain Argentina, Brazil, Russia, and the UnitedStates more than make up for the contrac-tion in cropland experienced in most othercountries. This is verified by the fact that cropprices fall for all countries (except, as noted,the United States), which is the source of thewelfare gain. Despite a fundamental differencein the setup, Counterfactuals 3 and 4 havemany similarities—for instance, the correla-tions between welfare and crop prices acrossCounterfactuals 3 and 4 are 0.87 and 0.95,respectively. In spite of these similarities, wefeel that the fixed cropland assumption is abetter approximation of reality.

Counterfactual 5: In the next two counter-factuals, we examine how an event in onecountry is transmitted to others. Counterfac-tual 5 considers how trade can be a vehicleby which the benefits of new technologiesare spread. Consider, for example, a pledgeby the Monsanto Company to develop seedsthat will double the yields of corn, soybeans,and cotton by 2030 (Pollack 2008). There canbe severe limits to the internationalizationof such technologies, however, due to agro-climatic differences across countries, licensingrestrictions, and imperfectly functioning inputmarkets. Many consumers will therefore bene-fit through only importation of the grain itself.Are trade costs, however, low enough for thebenefits of such innovations to be spread?

We consider the effect of a hypothetical 30%increase in the overall yields of the UnitedStates (TUSA). Results are reported in table 5(cropland area is fixed). One effect of thechange is a 4.4% increase in world trade vol-ume. U.S. welfare increases 2.2%, due largelyto a 6.2% drop in crop prices caused by anincrease in supply. In addition, there is a 2.8%increase in land rental rates as the U.S. cropsector expands to supply foreign markets. U.S.exports increase by 12.5%,while imports fall by13.5%. All other countries have a fall in crop

prices, which benefits consumers. This happensbecause they increase their imports (between1.4% and 12.3%). Only Argentina has a wel-fare decrease, and it is slight, at 0.1%. Thishappens because Argentina is a close com-petitor of the United States, with nearly ashigh a competitiveness (table 1) and of com-parable distance to major Asian and Europeanmarkets. The welfare of every other countryincreases.

Looking more closely,the benefits are spreadunevenly across countries. For example, cropprices fall by 4.0% in Mexico but fall by only0.3% in Russia.This cannot be due to our proxyfor general openness, since Mexico and Russiahave a very similar −θm coefficient (−0.70 and−0.75, respectively). It may be due to the factthat Mexico and the United States share a bor-der and are in NAFTA. On the other hand,the difference may be explained by distance.Based on the relative size of these coefficientsin table 1, Russia’s lengthy distance appears toexplain why it benefits less. In short, geographymatters.

Counterfactual 6: Another plausible meansby which countries can gain from trade—consumers in particular—is when one countryexpands its area in agricultural production.Brazil, for example, currently farms about 175million acres and is said to have room to doubleits available cropland to equal the scale of theUnited States,without clearing any more of theAmazon rainforest (Barrionuevo 2007). As anillustration of how this might affect the welfareof other nations, we consider a hypothetical30% expansion of acreage in Brazil.

The right three columns of table 5 reportthe effects on welfare, crop prices, and landrental rates (cropland area is fixed).The largestimpact is in Brazil, with a net welfare gainof 5.9%. Crop prices drop by 17.4% andland rental rates drop 17.8%. Overall worldtrade increases by 1.5%, with Brazil’s exportsincreasing by 70.9%. The effects are fairlyminor for other countries. The United Statesexperiences a small, 0.8%, drop in exports anda larger, 12.9%, rise in imports. However, dueto the large size of the U.S. crop sector, U.S.crop prices, land rental rates, and welfare areessentially unaffected.

For certain aspects of the results, an inter-esting pattern emerges. Changes in crop priceson distance are plotted in figure 4. The rela-tionship shows that countries located closeto Brazil (Peru, Argentina, Uruguay) bene-fit the most from its expansion. This suggeststhat while tariffs and nontariff barriers are

nloaded from

Table 5. Counterfactuals 5 and 6: Diffusion of an Event Across Countries

Yield increase: TUSA increases by 30% Cropland expansion: LBRA increases 30%

Net Crop Land rental Net Crop Land rentalCountry welfare prices rates welfare prices rates

Argentina −0.1 −0.9 −0.9 −0.1 −1.3 −1.2Australia 0.1 −1.5 −1.2 0.0 −0.4 −0.3Brazil 0.0 −0.5 −0.4 5.9 −17.4 −17.8Bulgaria 0.0 −0.2 −0.2 0.0 −0.1 −0.1China 0.0 −0.4 −0.3 0.0 −0.2 −0.2Ethiopia 0.0 −0.4 −0.4 0.0 −0.2 −0.2France 0.1 −1.8 −1.4 0.1 −1.0 −0.7Greece 0.2 −1.5 −1.2 0.1 −0.8 −0.6Hungary 0.0 −0.8 −0.7 0.0 −0.4 −0.3Italy 0.1 −1.4 −1.1 0.1 −0.7 −0.5Japan 0.0 −0.5 −0.4 0.0 −0.2 −0.2Mexico 0.4 −4.0 −3.1 0.0 −0.3 −0.2Morocco 0.0 −0.9 −0.7 0.0 −0.4 −0.4Peru 0.0 −0.6 −0.5 0.1 −1.5 −1.2Romania 0.0 −0.5 −0.5 0.0 −0.2 −0.2Russia 0.0 −0.3 −0.2 0.0 −0.1 −0.1South Africa 0.2 −2.4 −1.8 0.1 −0.7 −0.6Spain 0.2 −1.6 −1.3 0.1 −0.9 −0.7Turkey 0.1 −1.5 −1.1 0.0 −0.6 −0.5Ukraine 0.0 −0.1 −0.1 0.0 0.0 0.0USA 2.2 −6.2 2.8 0.0 −0.4 −0.3Uruguay 0.0 −0.5 −0.4 0.1 −0.8 −0.7Zimbabwe 0.0 −0.6 −0.5 0.0 −0.2 −0.1

Note: Values are percentage changes. Approach 1 parameters are used. World trade increases by 4.4% and 1.5% in the left and right scenarios, respectively.Crop prices refer to those faced by buyers.

Figure 4. Brazil expansion: Scatterplot ofchange in crop prices on distance

important in world agricultural trade, the roleof distance should not be overlooked. Evenif countries could eliminate all policy-basedbarriers to trade, the penalty imposed by geo-graphical trade costs limits the benefit from anincrease in supply elsewhere. Geography hasan important impact on the gains from trade.

Conclusions

In this study we propose a new framework forthe analysis of international agricultural trade.We carry out six counterfactual scenarios fortwenty-three countries to illustrate basic pointsabout the model and the world trading systemfor grains and oilseeds. The existence of largeinternational differences in crop yields, cropprices, and land rental rates suggests that thereshould be large gains from international tradein this sector.We find,however, that trade costsare close to those associated with autarkic equi-libria, and very little of the potential gains arebeing reaped.The volume of world trade wouldincrease fifteen times if all trade costs could beeliminated, including those that are difficult toreduce, such as freight costs, and those that aremore easily reduced,such as tariff and nontariffpolicy barriers.

Even if we restrict our focus to the elimina-tion of just those trade costs that are in princi-ple reducible, the welfare gains are quite highfrom freer trade. For example, if all countrieswere to reduce trade costs to the level of thecountry most open to imports, the median fall

nloaded from

in consumer prices is 56.9%. Such reductionscan have large effects on the volume of trade,as a 1% reduction in overall trade costs wouldincrease world trade volumes by 2.0–2.5%. Forthis reason, even small rises in trade costs canbe very harmful. The introduction of exportrestrictions by some crop exporters in 2007–2008 was an example of this, and highlightedthe thinness of many crop markets.

The results differ systematically across coun-tries. Distance inhibits the extent to which anevent in one country is transmitted to othercountries. The level of economic development,as proxied for by per capita income, is alsorelated to many results. Countries with highaverage productivity are more likely to befound on the high end of the global incomedistribution. Countries with high import tradecosts, by contrast, are more likely to be foundon the low end of the global income distri-bution. The insulation associated with highimport trade costs may shield countries some-what from adverse changes in other countries.However,it also means they may gain less whenother countries are able to increase their sup-ply. Regardless of these tendencies, it appearsthat all countries are very far from reaping thepotential gains from trade in international cropmarkets. The framework offers a means forfuture researchers to determine the gains fromtrade from new policy directions.

Funding

The authors acknowledge financial supportfrom the Oregon Agricultural ExperimentStation and the USDA Cooperative StateResearch, Education and Extension Service.

References

Abbott, P. C., P. L. Paarlberg, and P. M.Patterson. 1988. Supplier Substitutabilityby Importers: Implications for Assessingthe 1980 U.S. Grain Embargo. SouthernJournal of Agricultural Economics 20: 1–14.

Alvarez, F., and R. E. Lucas, Jr. 2005. Gen-eral Equilibrium Analysis of the Eaton-Kortum Model of International Trade.National Bureau of Economic ResearchPaper No. 11764.

Anderson, J., and P. Neary. 1996. A NewApproach to Evaluating Trade Theory.Review of Economic Studies 63(1):107–125.

Anderson, J., and E. van Wincoop. 2004. TradeCosts. Journal of Economic LiteratureXLII: 691–751.

Armington, P. S. 1969. A Theory of Demandfor Products Distinguished by Placeof Production. IMF Staff Papers 16:159–178.

Baldwin, R., and A. Venables. 1995. RegionalEconomic Integration. In Handbook ofInternational Economics, ed. G. Grossmanand K. Rogoff, 1597–1644. Amsterdam:Elsevier.

Barrionuevo, A. 2007. To Fortify China, Soy-bean Harvest Grows in Brazil. New YorkTimes, April 6.

Bernard, A., J. Eaton, J. Bradford Jenson, andS. Kortum. 2003. Plants and Productivity inInternational Trade. American EconomicReview 93(4): 1268–1290.

Billingsley, P. 1986. Probability and Measure.New York: John Wiley and Sons.

Dimaranan,B.,and R. McDougall (Eds.). 2007.Global Trade, Assistance, and Protection:The GTAP 6 Data Base. West Lafayette,IN: Purdue University Press.

Disdier, A. C., L. Fontagné, and M. Mimouni.2008. The Impact of Regulations on Agri-cultural Trade: Evidence from the SPSandTBTAgreements.American Journal ofAgricultural Economics 90(2): 336–350.

Donaldson, D. 2009. Railroads of the Raj:Estimating the Impact of TransportationInfrastructure. Unpublished manuscript,London School of Economics.

Dornbusch, R., S. Fischer, and P. Samuelson.1977. Comparative Advantage, Trade, andPayments in a Ricardian Model with aContinuum of Goods.American EconomicReview 67(5): 823–839.

Eaton, J., and S. Kortum. 2002. Technol-ogy, Geography, and Trade. Econometrica70(5): 1741–1779.

Effland, A., M. Normile, D. Roberts, and J.Wainio. 2008. World Trade Organizationand Globalization Help Facilitate Growthin Agricultural Trade. Amber Waves 6(3):22–27.

Grant, J., and D. Lambert. 2008. Do RegionalTrade Agreements Increase Members’Agricultural Trade? American Journal ofAgricultural Economics 90(3): 765–782.

Hertel, T. (Ed.). 1997. Global Trade Analysis:Modeling and Applications. Cambridge,MA: Cambridge University Press.

Irarrazabal, A., A. Moxnes, and L. Opro-molla. 2010. The Tip of the Iceberg: Mod-eling Trade Costs and Implications for

nloaded from

Intra-Industry Reallocation. Center forEconomic Policy and Research DiscussionPaper 7685.

Johnson,N.,and S. Kotz. 1970. Continuous Uni-variate Distributions. Boston: HoughtonMifflin.

Lloyd, P., and D. MacLaren. 2002. Measuresof Trade Openness using CGE Analysis.Journal of Policy Modeling 24: 67–81.

McCallum, J. 1995. National Borders Mat-ter: Canada-U.S. Regional Trade Patterns.American Economic Review 85(3): 615–623.

McClanahan, P. 2008. WTO Farm Talks Sput-ter into Action. BRIDGES Weekly TradeNews Digest, October 23, 1–3.

Niroomand,F.,and E. Nissan. 1997.An Inquiryinto Openness in International Trade.Journal of Economic Studies 24(3): 167–178.

Pollack, A. 2008. Monsanto Seeks BigIncrease in Crop Yields. New York Times,June 5.

Reimer, J., and S. Kang. 2010. Estimationof Trade and Domestic Distortions: AnApplication to World Agriculture. Journalof Productivity Analysis, in press.

Trostle, R. 2008. Global Agricultural Sup-ply and Demand: Factors Contributing tothe Recent Increase in Food CommodityPrices. USDA Economic Research Servicepublication WRS-0801.

nloaded from

trade costs and the gains from trade in crop agriculture

Documents

capital gains taxation and shareholder wealth in takeovers

dynamic gains from trade: the role of intermediate inputs...

market reforms and efficiency gains in chile

the welfare gains of trade integration in the european...

optimization-based trade-off analysis of biodiesel crop...

preserving capital gains in real estate transactions

the economic gains of trading stocks around holidays

trade theory and trade facts

reversing gains - aws

itqs in denmark and resource rent gains

local-food movement gains momentum - autogrill

returns to regionalism : an evaluation of nontraditional...

gains from trade without lump-sum compensation

assessing the welfare gains of trade integration in the...

efficiency gains of bancassurance – insurance companies...

who gains from product rents as the coffee

cmc in english language learning: gains and losses

stiffened panel stability behaviour and performance gains...

on productivity performance gains of indonesian firms

accelerated health gains from improved water supply