Global models applied to agricultural and trade policies ... · Global models applied to agricultural and trade policies: a review and assessment Frank van Tongerena, Hans van Meijla,

Agricultural Economics 26 (2001) 149–172

Global models applied to agricultural and trade policies:a review and assessment

Frank van Tongerena, Hans van Meijla, Yves Surryb,∗a Agricultural Economics Research Institue (LEI), P.O. Box 29703, 2502 LS The Hague, The Netherlands

b Institut National de la Recherche Agronomique (INRA), Unité d’Economie et de Sociologie Rurales de Rennes,Rue Adolphe Bobierre, CS61103, 35011 Rennes Cedex, France

Received 2 February 2000; received in revised form 12 July 2000; accepted 21 August 2000

Abstract

The continuing process of global integration bears implications for farmers and related supplying and processing industriesin all parts of the world, but also for the rest of the world economy. An assessment of agricultural and trade policy impacts isbound to be complex and is often supported by quantitative modeling analysis. This article provides an assessment of the presentstate of applied modelling in the area of trade and agricultural policies. We provide in this paper a comparative assessment ofalternative modelling approaches, considering a total of 16 partial equilibrium and general equilibrium models. The assessmentincludes theoretical modelling foundations, datasets employed and institutional aspects, such as model maintenance anddissemination of results. A typology of models is provided by structuring the assessment along a clear set of evaluationcriteria. © 2001 Elsevier Science B.V. All rights reserved.

Keywords:Global modelling; Agriculture; Trade

1. Introduction

The 1990s have witnessed a growing interdepen-dence of national economies. This global trend evenaccelerated since the signing of the Uruguay RoundTrade agreement in Marrakech in 1994. Signs of thisincreasing globalisation of the world economy takeseveral forms. World trade and foreign direct invest-ment (FDI) have grown at a faster pace in the ninetiesthan in the previous decade (WTO, 1998). Financialmarkets are more and more interrelated on a worldwide scale and any shocks occurring in one finan-

∗ Corresponding author. Tel.:+33-2-2348-5390;fax: +33-2-2348-5380.E-mail addresses:[email protected](F. van Tongeren), [email protected] (Y. Surry).

cial place have repercussions elsewhere in the world.Another sign of this world globalisation is the ongoingcreation of regional trade agreements and a consolida-tion of existing ones. On the policy side, globalisationof the world economy led policy makers, governmentsand international institutions to tackle global policyissues such as global warming and transboundary en-vironmental problems and come up with some realpolicy proposals. This awareness among policy mak-ers also manifests itself in the trade policy area with aglobal consensus to launch a new round of trade policynegotiations despite a bad start at Seattle in November1999.

This overall trend of globalisation of economicactivity and simultaneously a global orientation ofpolicy discussions is continuing to represent an enor-mous challenge for economists. Assessments of such

0169-5150/01/$ – see front matter © 2001 Elsevier Science B.V. All rights reserved.PII: S0169-5150(00)00109-2

150 F. van Tongeren et al. / Agricultural Economics 26 (2001) 149–172

phenomena has required the development of globaleconomic models. Such effort in global modelling isnow well established and has become an integratedpart of the economics field. Furthermore, the devel-opment of such models has been facilitated by therapid improvement in information and communicationtechnologies (ICT). In parallel with the globalisationof economic activity and policy making, economicsresearch is globalising as well. Large scale globalmodels are increasingly supported by internationalteams and international groups of stakeholders.

The purpose of this article is to review the presentstate of applied modelling used to examine the globalimpacts of agricultural and trade policies. Althoughthe scope of this review is limited by a deliberate biastowards agricultural trade issues, we should also notlose sight that, nowadays, many global phenomenago beyond agricultural and trade policy issues, someof which are environmentally related. This led us toadopt an open-minded strategy in conducting this re-view of global models. More specifically, we reviewednot only global models with an exclusive agriculturaland trade policy focus but also looked at models whichmay also be used to assess other global problems suchas global macroeconomic and environmental issues.

The continuing process of global integration bearsimplications, not only, for farmers and related sup-plying and processing industries in all parts of theworld, but also for the rest of the world economy.An assessment of likely policy impacts is bound tobe complex and should be supported by quantitativemodeling analysis that explicit the relations of coun-tries between each other, and that explicit relevantinteractions between the agricultural sectors and therest of the economy.

There is no model which can serve all purposes.The choice of theoretical framework, the extent ofregional and sectoral desegregation and the choiceof datasets and estimation methods determine thedomain of applicability of the model. Potential usersof applied models should be aware of strengths andweaknesses of alternative approaches. This reviewprimarily presents factual information about relevantmodels, and presents this information in a structuredformat so as to highlight common features, differencesand areas of applicability of modelling approaches. Inthis way we hope to assist the model user in makinghis own assessment.

Table 1 lists the sixteen models discussed in thisreview, their initiating bodies (institutions and/orpersons) and their current status. A first distinctionin Table 1 is that between partial equilibrium mod-els focusing on agriculture on the one hand, andeconomy-wide models on the other hand. These latterglobal models are designed in such a way that theycould address agricultural and trade policy issues butalso other global problems such as global macroeco-nomic and environmental ones. A large part of thesemodels are still operational and are currently usedfor policy- and outlook analysis. For these reasons,they are in a continuous state of flux of evolvingthrough successive updates and changes. This situ-ation led us to conduct the model review at a pointof time (early 1999) and provide a snapshot of allthese models as they functioned at the beginning of1999. Changes that could have occurred to the op-erational models since this date are not consideredin this article. Information on the individual modelshas been gathered using published papers and journalarticles, unpublished working documents, electronicwww/documents and personal contacts. An annex tothis article, which is available on request, provides adetailed description of each of the models reviewed.

The requirement that the model should be relativelyrecent and likely to be used in the 1990s has led us toexclude important precursors such as the IIASA BasicLinked System (Parikh et al., 1988), The GOL modeldeveloped by USDA-ERS (Roningen and Liu, 1983),OECD’s MTM model (Huff and Moreddu, 1990)and the Tyers–Anderson model (Tyers and Anderson,1992). We have also excluded single-commodity trademodels and linear (or non-linear) programming mod-els that attempt to describe input–output relationshipsfor a certain production process in great detail.

Although this review of global models has an agri-cultural and trade policy focus, it does not pretend tobe exhaustive. The authors are well aware that someglobal models have been excluded from this review.This is especially true in the area of environmentalmodelling and more especially in the area of globalwarming where a large and continuous research efforthas taken place over the last 10 years.1 Also, existing

1 For a full review and state of the art in models of climaticchange, see the various chapters of the OECD Workshop on Eco-nomic Modelling of Climate Change (OECD, 1998).

F.va

nTo

nge

ren

et

al./A

gricu

ltura

lE

con

om

ics2

6(2

00

1)

14

9–

17

2151


global trade models are increasingly taking environ-mental issues into account. In the same vein, we donot intend to review the large body of work dealingwith the specification of environmental models whichis out of the scope of this review.2

2. Evaluation criteria

This section introduces our set of evaluation criteria,which are subsequently used to describe and comparealternative modelling approaches. The set of criteria isbased on the recognition that applied modelling formsa combination of theory and empirical data, both ofwhich deserve due attention in policy relevant mod-elling.

The general ‘filter’ for inclusion of models in thecurrent review has been that the model should

• be multi-region and global in nature;• have relevance for agriculture, trade and natural re-

source based activities;• be multi-commodity;• have a medium term time frame (around 5 years);• be an equilibrium model (i.e. not models that project

demand-supply gaps using primarily technical rela-tions);

• be of recent vintage, and likely to be used in someform in the 1990s;

• be an applied model, i.e. uses a combination of the-ory and empirical data;

• have a strong policy orientation.

2.1. Conceptual framework: definition and scope

2.1.1. Market equilibrium models versustime series projection models

Although time series projection models have beenexcluded from the list, it is useful to distinguish this ap-proach from equilibrium models.Time series projec-tion modelsattempt to forecast the future on the basisof extrapolation of historical data. These models typ-ically put more emphasis on the statistical behaviourof time series data than on the economic theoreticalunderpinnings of behavioural equations. A projectionmodel may, for example, project commodity supply

2 For a recent review of these environmental models, see Conrad(1999), and Faucheux and Levarlet (1999).

based on agronomic data on acreage and yields with-out taking into account farmers’ responses to changingmarket prices. For a discussion of this method and alarge scale application, see Alexandratos (1995). An-other well known global model that relies heavily onthe projections method is found in Rosegrant et al.(1995).

On the other hand,market equilibrium modelscon-tain the response (behaviour) of economic agents tochanges in prices (costs), and prices adjust so as toclear markets. The objective of these models is the de-termination of equilibrium prices and quantities on (in-terrelated) sets of markets.3 In a fully fledged globalequilibrium model, there will typically be endogenousprices attached to world markets as well as domes-tic markets. This class of models is firmly establishedwithin mainstream economics where the behaviouralresponse of suppliers and buyers is typically derivedfrom optimising assumptions: given a description ofthe production technology, the supplier chooses a com-bination of inputs such that costs are minimised fora given level of output. Given a description of con-sumer preferences, the buyer determines his preferredconsumption bundle such that his/her utility is max-imised for a given level of his/her budget. Standardassumptions include constant returns technology, ho-mothetic preferences, and markets characterised byperfect competition. While these basic theoretical as-sumptions underlie equilibrium modelling, the opti-misation process is usually not modelled explicitly.Rather, a reduced form approach is common, wheredemand and supply are specified as functions of in-come, prices and elasticities.

Depending on assumptions made about the flexi-bility of production factors, equilibrium models canbe classified as short term, medium term or long term.Short term (in the Marshallian sense) means that someproduction factors are fixed, and are not allowed toreallocate between alternative uses. The fixed factorswill typically be capital, agricultural land, and per-haps agricultural labour. Medium term models allow

3 This does not deny the existence and relevance of disequilib-rium situations. Temporary shortages and excess supply situations(which may, for example, arise as a consequence of price or quan-tity regulations) can very well be captured in equilibrium models,for example, by allowing for stockpiling and depletion. The keypoint is that these models catch market interactions in a coherentand theoretically sound way.

F. van Tongeren et al. / Agricultural Economics 26 (2001) 149–172 153

for reallocation of all production factors as responseto some exogenous events. Finally, long term mod-els would also model endogenous capital formation.Within the group of market equilibrium models, wecan identify partial and economy-wide models.

2.1.2. Representation of national economies:partial versus economy-wide models

Partial models treat international markets for aselected set of traded goods, e.g. agricultural goods.They consider the agricultural system as a closedsystem without linkages with the rest of the econ-omy. Effects of the rest of the domestic and worldeconomy on the agricultural system may be includedin a top-down fashion by altering parameters andexogenous variables. Partial models are in principleable to provide much product detail. See also Meilkeet al. (1996) who give a summary of global partialequilibrium models adapted to agricultural trade.

Partial models may be single- or multi-product.Multi-product models are able to capture supply anddemand interrelationships among agricultural prod-ucts. Most partial models include linear or log-linearbehavioural equations, which allows the representa-tion of supply and demand relationships (responses)prevailing in the markets under study.4 They also in-corporate into their supply and demand relationshipsexogenous variables such as technical change, worldpopulation and household income.

Partial models of international trade in agriculturegenerally focus on trade in primary commodities. Thatis, they capture agricultural supply, demand and tradefor unprocessed or first-stage processed agriculturalproducts without taking into account trade in pro-cessed food products, despite the fact that the lattercommodities represent an increasing share of worldtrade.

The main area of application of partial equilibriummodels is detailed trade policy analysis to specificproducts, which represent only a small portion of theactivities of the economy in question. This (small sec-tor) condition implies that policy-induced changes on

4 Some (single country) partial equilibrium modellers attempt toovercome the shortcomings that are implied by linear and log-linearbehavioural equations by estimating flexible functional forms. See,e.g. Frohberg et al. (1997). To our knowledge this has not yetbeen applied in global trade models, however.

the rest of the economy (outside the farm sector) areso small that they can safely be ignored.

On the other hand, economy-wide modelsprovidea complete representation of national economies,next to a specification of trade relations betweeneconomies. A first step in moving from partial equi-librium to economy-wide modelling is to introducesupply and demand equations for an aggregate resid-ual commodity. By imposing regularity restrictionson the supply and demand elasticities of the amendedmodel, one obtains a model that includes demand andsupply interactions between agricultural commodi-ties and other commodities in a consistent way. Afuller economy-wide specification is obtained whenthe model is closed with respect to the generation offactor income and expenditures, which requires theexplicit specification of factor markets for land, labourand capital. In other words, the essential generalequilibrium features are captured by including fac-tor movements between sectors, next to allowing fordemand interactions. Economy-wide models captureimplications of international trade for the economyas a whole, covering the circular flow of income andexpenditure and taking care of inter-industry relations.

There are three broad classes of economy-widemodels: macro-econometric models, input–outputmodels and applied general equilibrium (AGE) mod-els. Macro-econometric models do not concern ushere, since they will not zoom in on agriculture, butrather are concerned with macro-economic phenom-ena such as inflation and exchange rates. Input–outputmodels provide a comprehensive description ofinter-industry linkages and a full accounting of pri-mary incomes earned in production activities.

AGE models do also usually contain full input–output detail, but on top of that they contain equationsthat describe the behavioural response of producers,consumers, importers and exporters and possibly otheragents in the economy.5 AGE models are specifi-cally concerned with resource allocation issues, thatis, where the allocation of production factors overalternative uses is affected by certain policies or ex-ogenous developments. International trade is typicallyan area where such induced effects are important

5 Limitations of open Leontief input–output models include fixedprices, exogenous final demand, perfectly elastic factor supplies,and an inability to demonstrate welfare effects.


consequences of policy choices. Needless to say, suchinduced effects are not visible in partial models. In theface of changing international prices, resources willmove between alternative uses within the domesticeconomy, or even between economies if productionfactors are internationally mobile. Only if a completedescription of the multi-sectoral nature of the econ-omy is provided, can such developmental issues beanalysed.

2.1.3. Regional scopeMulti-region models differ with respect to their

regional coverage.6 Global trade models attempta closed accounting of the selected commoditytrade flows for the entire world. If the model iseconomy-wide, the global model also includes aglobally closed income accounting system. At theother end of the scale, a model might focus ontrade between a selected set of trading partners,without attempting a globally closed accounting.Or it might even single out one group of countries,such as the EU-15, and describe its trade on worldmarkets.

A globally closed database does not imply that allregions or countries distinguished are treated with thesame amount of detail. An intermediate position isfrequently adopted, wherein the model’s database isclosed with respect to the world, but only selectedregions are treated with a great amount of detail, andconfining the description of other regions to a smallerrange of variables that are of crucial importance.

2.1.4. Linked individual country models orparametric differences between regions

There are two broad approaches with respect to themodelling of individual economies within the globaleconomic system. One approach starts by giving a de-tailed representation of individual economies, takinginto account much of the institutional and economicdetails of the individual countries, and subsequentlylinking individual country models through trade flows,

6 In accordance with the international trade literature, ‘regional’has a supra-national meaning in this report, and not an intra-national (provinces, etc.) one. A ‘country’ corresponds to thenotion of a nation state. Whenever this article refers to regions, wemean an aggregate of individual countries. Regional aggregationsof countries, therefore, do not necessarily represent a coherentgeographical space, for example, a ‘Rest of the World’ region.

capital flows and possibly factor mobility betweencountries.7

The other route to global modelling starts byassuming the same modelling structure for all in-dividual economies, and representing differencesbetween economies in terms of data and parametersonly. This approach yields a relatively transparentmodel structure, since there is only one economicmodel. This in turn greatly facilitates both the datahandling aspects as well as the interpretation of re-sults. In the linked country models approach, theindividual country models may be based on differenttheoretical assumptions, which may make it difficultto disentangle model results into the effects of ex-ogenous events on the one hand and differences intheories on the other hand. A disadvantage of the‘one model fits all’ approach is clearly its limitedcapability to handle structural differences betweeneconomies.8

2.2. Specification and modelling issues

2.2.1. Dynamic versus comparativestatic specifications

Dynamic models allow the analysis of lagged trans-missions and adjustment processes over time. Alter-natively, the comparative static approach studies thedifferences between equilibria resulting from differentassumptions on exogenous data or policy variables.The time path between equilibria is ignored in com-parative static models.

Dynamic models can be used to trace the accu-mulation of stock variables, whereas static modelsare unable to do this. In comparative static models,policy changes have no effect on the accumulation ofstocks — e.g. capital stock — and the associatedchanges in production possibilities. For short-runagricultural analysis the implications of accumulatingcommodity stocks may be relevant as well.

Dynamic features can be incorporated in equilib-rium models in several ways. The most frequentlyused approach is to specify a recursive sequence of

7 The United Nations Project LINK (Klein and Su, 1979) is awell known example of this approach to global modelling.

8 An intermediate approach has been followed by the GLOBUSmodel (Bremer, 1987), which included three prototypical modelstructures for, respectively, developed market economies, centrallyplanned economies and developing economies.


temporary equilibria. That is, in each time period themodel is solved for an equilibrium, given the exoge-nous conditions prevailing for that particular period.In between periods, stock variables are updated, ei-ther exogenously (e.g. population) or as a result of theequilibrium outcomes of the preceding period (e.g. in-vestment demand leading to a changed capital stock inthe next period). Recursive dynamics do not guaranteetime-consistent behaviour. In contrast, in intertem-poral equilibrium models agents display optimalbehaviour over time as well as within periods.Intertemporal models are usually tantamount tousing rational expectations assumptions.9 Suchforward-looking behaviour leads to equilibrium timepaths that move towards a long-run steady state (ifit exists). A main reason to incorporate such in-tertemporal features into general equilibrium mod-els is the desire to model savings rates endoge-nously, and hence to allow the model to generatealternative (endogenous) growth rates. In such mod-els, a policy change can have a lasting effect onthe economy’s growth rate through changes in theaccumulation of capital stocks. A feature whichis impossible with a fixed savings rate assump-tion.10

Comparative static models are sometimes used togenerate projections of policy impacts at some futurepoint in time. Such projections are not to be confusedwith econometric forecasts, but are achieved by con-structing an artificial future dataset that is consistentwith the model’s assumptions — a so-called baseline— and subsequently conducting a policy experimenton the basis of this projected dataset. The artificial fu-ture dataset is constructed by making assumptions onthe growth of exogenous variables and parameters andsubsequently letting the model solve for an equilib-rium that is consistent with these assumptions. Typi-cal projections with AGE models rely on exogenousforecasts of GDP, factor endowments and factor pro-ductivity.

9 Furthermore, by modelling the intertemporal equilibrium be-haviour of economic agents, as well as equilibria within periods,such models are able to counteract the Lucas (1976) critique oneconomic models.10 If knowledge is included as a production factor that can beaccumulated, growth rates become endogenous, even with a fixedsavings rate (see endogenous growth literature, for example, Gross-man and Helpman (1991)).

2.2.2. Modelling of international trade

2.2.2.1. Assumptions concerning the nature of tradedgoods: homogeneous versus heterogeneous goods.In classical trade models, goods are assumed to besimilar in the eyes of buyers. In such a market, thegoods of one producer perfectly substitute for thoseof another and are calledhomogeneous. If the num-ber of suppliers is sufficiently large, the market willapproach the perfect competitive outcome and pricesacross suppliers will be equalised. Homogeneity andcompetitiveness also imply that each actor in themarket is either a buyer or a seller of the good, butnever both, since each actor is either able to producethe good with non-negative profits at the prevailingmarket price or not. This implies that a country canonly be an exporter or an importer of a certain good,and models that include this assumption describe onlyinter-industry trade.

Homogeneity therefore simplifies the task of trademodelling considerably. Since prices are equalised andthere is no other distinguishing characteristic of thegoods, it makes no difference from which suppliera particular purchase is made. The homogeneity as-sumption is therefore associated with a ‘pooled’ mar-ket approach to trade modelling, where we see onlywhat each actor brings to the market (supply) and whatthat actor takes from the market (demand). For obvi-ous reasons, the pooled market approach is also knownas ‘non-spatial’ modelling.

However, these simplifications in modelling havesevere limitations for applied trade research, as thesemodels explain only inter-industry trade and notintra-industry trade. The latter turns out to be animportant phenomenon in trade, since even at highlevels of disaggregation, countries report both exportsand imports in any sector. If intra-industry trade isnetted out, one ignores an important phenomenon ofthe real world and underplays the importance of tradeto each region. Furthermore, these trade models canbe hypersensitive to changes in transportation costsand trade policy wedges, and run the risk of extremespecialisation when sector-specific factors of produc-tion are not present in the model, see also Francoisand Reinert (1997).

One way to introduce intra-industry trade in amodel is to assume that goods are distinguished byother factors than price alone, and hence are viewed


as imperfect substitutes from the perspective of thebuyer. When product differentiation is possible, goodsare calledheterogeneousand the task of trade mod-elling is considerably more complex. First, there is noneed for prices to equalise across suppliers. If goodsare heterogeneous, then different buyers are willing topay different prices to obtain the same quantity of thegood. Hence, independent price movements amongsuppliers are possible. Second, each actor in the mar-ket may be both a buyer and a seller at the same timeif goods are differentiated. This implies that the trademodel has to trace twice as many transactions thanunder the homogeneity assumption. Clearly, the bilat-eral specification provides a richer and more detailedpicture of the market, but requires much more data,parameters, bookkeeping and computational effort.Whether the additional effort is justified by additionalrelevant information depends on the kind of problemsa model has to tackle. Many policy disputes and pol-icy instruments are bilateral in nature, and can only betreated within a bilateral specification of trade flows.

There are two ways to incorporate product differ-entiation into applied trade models. On the one hand,product differentiation can be introduced exogenouslyby assuming that products are differentiated by coun-try of origin. This method introduced by Arming-ton (1969) simply assumes that imports and domesticgoods are imperfect substitutes in demand. The oftenused Armington formulation in applied trade modelsinvokes the assumption that products are differenti-ated by country of origin. In combination with a pref-erence function that is separable in domestic productsand combined foreign products, this yields empiri-cally manageable import functions. Most often a CESfunctional form for preferences is assumed.11 Thisassumption has received much criticism because thesource of product differentiation is exogenously intro-duced on the demand side. Another disadvantage ofthis assumption is that terms of trade effects turn out tobe quite large empirically. The Armington assumptionimplies that each importer, however small the regionmay be, has some degree of market power, and is there-

11 In general, the assumption of (weak) separability of the util-ity function implies that ‘bundles’ of goods have to be identified,where substitution is relatively easy within ‘bundles’ but substi-tution is less easy across ‘bundles’. This is evidently also an em-pirical issue.

fore able to influences world prices. See Brown (1987)for an analytical and numerical assessment of termsof trade effects under the Armington assumption.

An alternative approach is to introduce product dif-ferentiation endogenously at the firm level on the sup-ply side.12 This approach assumes that consumersprefer differentiated goods either to obtain a bettermatch between their preferred variety and those extantin the market place (Lancaster, 1980) or to obtain in-creased variety in consumption (Spence, 1976; Dixitand Stiglitz, 1977). Krugman (1979, 1980) and Ethier(1979, 1982) introduced the concept of monopolisticcompetition into international trade theory. In this ap-proach fixed costs such as R&D or marketing costsare necessary to produce differentiated goods. The in-clusion of fixed costs has some implications for tradepolicies. Next to the traditional gains from trade weget ‘noncomparative advantage’ gains from trade inthe presence of scale economies and imperfect com-petition. First, shocks that increase output at firm levelresult in positive scale effects. Second, there are gainsfrom trade in the form of increased variety (therebyincurring fixed costs and reducing the sales of existingfirms). Thirdly, scale economies imply that the mar-ket can only support a limited number of firms, whichare consequently imperfectly competitive. Trade cre-ates a larger market that can support a larger numberof firms and hence a greater level of competition. Thereduction in market power is called the procompeti-tive effect. The advantage of this approach is that itlocates product differentiation on the supply side andit minimises terms of trade effects. A disadvantage isthat the absence of firm level data makes econometricestimation of elasticities problematic (Winters, 1990).

2.2.3. Representation of policiesA careful representation of policies is an essen-

tial component of global models applied to practicalagricultural and trade issues. The policies consideredcover not only trade policies but also domestic agri-cultural policies which add further distortions betweeninternational and domestic prices. Modelling policy

12 Berkum van and van Meijl (2000) give an elaborate survey ofdeterminants of international trade, such as factor endowments,product differentiation, economies of scale and innovation, andtheir implications for trade patterns and trade policies. Furthermore,they survey empirical evidence on the relevance of various theoriesfor explaining trade in agricultural and food products.


instruments in global models can take two forms. Thefirst one consists of developing a direct structural rep-resentation of the policy instruments through the in-corporation of its mechanisms. The second approach ismore indirect and measures the policy-induced distor-tions through a price-transmission (policy-response)relationship linking international and domesticprices.13 Depending on the values taken by this rela-tionship, it is flexible enough to capture a wide arrayof trade and domestic agricultural policy regimesranging from a perfect transmission of world pricesto perfect insulation. Measures of distortions cap-tured by price wedges and/or tariff equivalents areincorporated into this policy response function.

In the global models under review, these two formsof representing policies are being used extensively.Concerning the price transmission specification, themost common form of modelling policy instrumentsis the perfect transmission case with price wedges andor tariff equivalents. This form is relatively easy to im-plement. In what follows, we illustrate the modellingof policy instruments through two trade policy instru-ments: tariffs and quantitative restrictions.

Tariffs and quantitative restrictions such as quo-tas and voluntary export restraints (VERs) are twotypes of trade policy instruments that are examinedin applied trade models. Tariffs can be introduced ina straightforward manner and are most of the timeexpressed as the percentage by which the domesticprice exceeds the world price; i.e. an ad valorem tar-iff rate. Quotas are more difficult to implement. First,one has to investigate whether a quota is binding ornot. Second, it is difficult to assess what would be thelevel of imports without the quota. Third, one has tomodel who appropriates the rents that accrue from thequota: domestic importers or foreign exporters. Withregard to the second element researchers focus on the

13 For more details on the specification and implementation ofprice transmission equations, see Tyers and Anderson (1992,Chapter 2). Price transmission equations can also be viewed asreduced-form relationships obtained from a structural “politicaleconomy” model explaining the behaviour of policy decision mak-ers who optimise a policy preference (social utility) functionssubject to economic and political constraints. The empirical im-plementation of price transmission equations is often pragmaticand not necessarily derived in a consistent fashion from the un-derlying political economy model. This could explain why such aspecification of policies is not used in global computable generalequilibrium models.

price distortions caused by the quota. There are severalmethods to quantify quotas and other non-tariff mea-sures (Laird, 1997), which basically amount to two al-ternative ways to implement them in applied models:the first is a tariff equivalent representation, while thesecond method specifies quantity restrictions directly.

2.2.3.1. Tariff equivalents or price wedges.The tar-iff equivalent or price wedge of a non-tariff measure isthe difference between the free world price of a goodand the price on the domestic market. These measurescan be relatively easily be observed when goods arehomogenous and free world prices can be obtainedfrom transaction values. For manufactured goods, theformer is a problem and for many (agricultural) com-modities the latter is a problem. A popular methodto arrive at estimates of tariff equivalents is use ofproducer support estimates (PSE).14 A disadvantageof these subsidy equivalents is that they vary consid-erably from year to year, not only through changesin policies but especially through changes in worldprices, exchange rates and the value of domestic pro-duction.

The representation of quantity restrictions as pricewedges is not always an adequate approach. If a quotais not binding in the benchmark, its tariff equivalentwill be equal to zero. However, the quota may becomebinding as the result of a policy simulation. This effectwill not be captured when the quota is approximatedby a tariff equivalent because the tariff equivalent re-mains zero. Additional complications arise in the caseof multi-tier protective schemes like tariff-rate quotas(TRQs). These schemes specify different tariff ratesfor different import quota levels, and it has to be de-termined whether each quota is really binding and/orthe tariff is prohibitive. Also, the implementation ofa quota implies the generation of quota rents, which

14 PSEs were formerly known as ‘producer subsidy equivalents’.These include the transfer from price distortions (i.e. price wedges),and the transfers from government to producers. The transfer fromprice distortions or market price support is again the transfer fromconsumers to producers in the form of price gaps between do-mestic and world prices. The transfer from government expendi-tures includes both direct government payments and indirect trans-fers (e.g. input subsidies, tax concessions). The consumer subsidyequivalent (CSE) measures the implicit tax or subsidy imposed onconsumers. See Cahill and Legg (1990) for a comprehensive de-scription. Both PSEs and CSEs are regularly published by OECD.


accrues as income to the holder of the right to importor export under the quota. The endogenous determina-tion of quota rents and their distribution over holdersof quota rights can only properly be captured by anexplicit representation of these policy measures. See,for example, DeMelo and Tarr (1992) for a discussionof trade quota instruments.

2.2.3.2. Other policy instruments.Next to borderprotection instrumentsstricto sensu, other relevantpolicies frequently need to be represented in models.For example, in relation to the GATT/WTO commit-ments, ceilings on the volume of subsidised exportsas well as bounds on the value of export subsidiesmay be relevant. In relation to the common agri-cultural policy (CAP) of the European Union, landset-aside and headage premiums are clearly examplesof agricultural polices that do not directly affect bor-der protection, but nevertheless do have an impact ontrade flows.

2.2.4. Theoretical consistencyJudging the theoretical consistency of models has

many facets, and the discussion here is far fromexhaustive.15 At its most basic level, a model’s nu-merical results should be qualitatively in accordancewith the theoretical foundations on which the modelhas been erected. At the level of numerical implemen-tation of the model, theoretical consistency placesrequirements on the parameters used in functionalforms, especially parameters used in demand systemsand supply equations.16

15 It is not a straightforward task to develop a sound set of criteriato judge the theoretical consistency of models. This theme is alsoclosely related to the issue of model validation, which we havenot taken up in this article. In addition the evaluation of theoret-ical validity would require much more information on the indi-vidual models than is available. Since numerical model outcomesare contingent on the particular set of assumptions employed, inparticular policy assumptions and assumptions on exogenous vari-ables, a judgement of the numerical validity would require a studywhere model results are compared against the background of thesame set of assumptions.16 The four essential properties of demand functions are: (1)adding up: at the given level of prices and income, demand equalstotal expenditure, (2) homogeneity: compensated (Hicksian) de-mand is homogeneous of degree zero in prices and uncompen-sated (Marshallian) demand is homogeneous of degree zero inincome and prices together, (3) symmetry: cross-price effects are

Imposing all regularity conditions on the parame-ters used in demand system and the supply equationsof global partial equilibrium models is almost an in-surmountable task. To overcome this problem, globalpartial equilibrium modellers have adopted the follow-ing two strategies: (i) ensure that all parameters (elas-ticities) satisfy the essential regularity conditions (i.e.that own price effects have the right signs and domi-nate the cross-price effects) and/or (ii) impose the fullset of regularity conditions to small components of theglobal model.

The economic structure of general equilibriummodels forces the model builder to exercise a strictdiscipline with regard to the restrictions on param-eters. In particular, a necessary condition for theexistence and uniqueness of an equilibrium solutionis that all excess demand functions are homogeneousof degree zero in prices. This condition is met whenthe regularity conditions on demand and supply equa-tions are satisfied. In a properly designed generalequilibrium model, equality between incomes and ex-penditures will always be satisfied for the economy asa whole. This feature does not always hold for partialequilibrium models because they lack the restrictionsimposed by an economy-wide national accountingframework.

2.2.5. Model closures‘Closing’ the model is the process of classifying the

variables as either endogenous, i.e. values are deter-mined (solved for) by the model, or exogenous, i.e.predetermined outside the model. Model experimentsare conducted by introducing alternative assumptionson exogenous variables.

Alternative model closures can also be employed toconstruct different models from the same basic mod-elling framework. Multi-region models with a globalcoverage can sometimes be transformed into singleregion models by singling out one specific regionand declaring ‘the rest of world’ as exogenous. Simi-larly, economy-wide models can be transformed into

symmetric, (4) negativity: the matrix of own- and cross-pricederivatives of compensated demand functions is negativesemi-definite. In particular this implies that (a) compensated de-mand function slope downward, and (b) own price effects dom-inate cross-price effects. See, for example, LaFrance (1986) andLaFrance and Hanemann (1989). Similar observations hold forequation systems used to model the supply side.


partial models of selected sectors, by specifying aclosure which holds ‘the rest of the economy’ as ex-ogenous. The latter possibility is especially useful ifone wants to compare partial equilibrium outcomesagainst general equilibrium outcomes, thereby check-ing for the presence of significant economy-wideinduced effects following a certain policy changetargeted at the agricultural sector.

There are certain general closure rules that haveto be fulfilled by both partial and general equilib-rium models. Both in a partial and a general equi-librium setting, a valid closure has to assure thatthe number of endogenous variables is equal to thenumber of equations. In addition to this necessarytechnical condition, the closure must specify a valideconomic environment. For example, if the equilib-rium model demands that all buyers exhaust theirbudget, the closure must be such that all buyersare on their budget constraint, and that there areno ‘leakages’ with respect to incomes and expendi-tures.

Since partial models describe only a subsystem ofthe economy, they do not have to be concerned aboutthe so-called ‘macroeconomic closure’, i.e. the treat-ment of the link between investment and savings. Inmodels without intertemporal decision making withrespect to investments, the identity between macroe-conomic investment and savings is guaranteed byfixing either one at some pre-specified level, andrequiring the other variable to accommodate. For ex-ample, a so-called ‘Keynesian closure’ specifies anexogenous investment level and lets savings adjustendogenously. On the other hand, in ‘neo-classicalclosures’, investment is adjusting to savings levels.Since the source of savings may be both domesticand foreign, the closure rule also has implications forthe treatment of the current account balance. If thetrade balance is fixed exogenously, one essentiallyalso fixes the difference between domestic savingsand domestic investments. In addition, in multi-regioneconomy-wide models, the approach of fixing tradebalances at the regional level is a simplifying way toavoid modelling the allocation of global savings toindividual regions. A disadvantage of this approachis the inability to model endogenous changes in thevolume of regional trade balances. It is obvious thatthe specification of the macroeconomic closure canhave profound impacts on model outcomes.

2.3. Data and parameters

2.3.1. DataData requirements are very demanding for

multi-regional models of international trade. Theamount of data is determined by the level of desegre-gation (countries/regions, activities/commodities) andthe theoretical structure (homogeneous/heterogeneousgoods, bilateral/pooled markets).

Not only is the amount of data usually quite large,but the data needed to be mutually consistent. Es-pecially if trade is related to domestic inter-industrystructures, substantial adjustments to the publisheddata are necessary, because the procedures for collect-ing and classifying trade statistics differ from thoseemployed for domestic input–output tables. Whiletrade data with broad coverage are now widely avail-able on a comparable basis, this is certainly not truefor input–output data and for trade protection infor-mation.17 A social accounting matrix (SAM) usuallyunderlies economy-wide models. Although the SAMis sometimes only implicitly present in the databaseof AGE models, it forms the basis for a coherentand consistent description of national economies. SeeLaird (1997) for a description of widely used datasources for international trade analysis.18

It is obvious that regular updating of datasets willimprove the timeliness and relevance of results. Thechoice of base year for a modelling dataset has addi-tional consequences, both for comparative static anddynamic models. The economic conditions that pre-vail at the point of reference determine the conclusionsthat can be drawn from alternative simulations.

2.3.2. ParametersThe parameters used in behavioural equations deter-

mine the response to policy changes, and are thereforea very crucial element in each modelling exercise. Keyparameters usually are: price- and income elasticities

17 A recent joint initiative by USDA/ERS, Agriculture and Agri-food, Canada, the European Commission, UNCTAD, FAO andOECD develops a new Agricultural Market Access Database(AMAD). At the time of writing this contains tariff-line leveldata on market access commitment and implementation of about50 WTO members. AMAD is publicly available since September2000. See Waino et al. (2000) and www.amad.org.18 The annex to van Tongeren and van Meijl (1999) lists thedatasources of individual models in some detail.


and budget shares in demand systems; substitutionelasticities and input cost shares in supply systems;Armington (substitution) elasticities in import de-mand; if economies of scale are included, parametersthat capture the degree of exhaustion of returns to scale(cost-disadvantage ratio). As already mentioned abovein Section 2.2.4 the values of these parameters mustbe determined in consistency with data and theory.

Two approaches to estimating model parameterscan be distinguished: econometric estimation andcalibration. Econometric estimation of parametersshould ideally be done by simultaneous equation es-timation methods that take into account the overallmodel structure. However, given the size of appliedtrade models, identification problems, lack of data,etc., this is not feasible, and one has to resort tosingle-equation estimation methods, using either timeseries or cross-section data. See Jorgenson (1984) foreconometric estimation of AGE models.

Most applied trade modellers resort to calibrationmethods — also called the ‘synthetic approach’ —to generate a set of parameters that is consistent withboth the benchmark data and the model’s theory. Thecalibration approach takes initial estimates of elastic-ities, etc., from outside sources and adjusts certainother parameters in the given functional forms to theinitial equilibrium dataset. Calibration, therefore, ex-ploits theoretical restrictions, equilibrium assumptionsand assumptions on functional forms to arrive at apoint estimate.

3. Model overview

3.1. Partial models

In this section we describe the features of the se-lected partial equilibrium models. We first describe thedesign choices of a prototypical standard multi-regionpartial equilibrium model. This standard model willserve as a point of reference for the individual par-tial equilibrium models. Individual models are sum-marised in alphabetical order in Table 2, which alsoidentifies their non-standard features.

A standard partial equilibrium model has the fol-lowing features:

• regional scope: global coverage;

• regional unit of analysis: parametric differencesbetween countries;

• dynamics: comparative static;• modelling of trade: homogeneous goods;• characterisation of global markets: pooled markets;• representation of policies: ad valorem price wedges

(trade: tariff equivalents);• theoretical consistency: not implied by theoretical

structure;19

• model closures: factor markets and non-agriculturalsectors are exogenous.

The standard multi-region applied partial equilibriummodel framework consists of an economic structurethat includes for each region constant elasticity sup-ply and demand equations which determine domesticprices. The standard model is multi-product by na-ture to capture supply and demand interrelationshipsamong agricultural products. Therefore, supply anddemand equations are functions of own and crossproduct prices. The interactions between the agri-cultural product groups are taken into account whileinfluences of factor markets and the rest of the econ-omy are treated as exogenous. Supply and demandrelationships incorporate therefore exogenous vari-ables such as population, household income and tech-nical change. Each sector produces one homogeneousgood that is perfectly substitutable both domesticallyand internationally. A region’s international trade isviewed as the difference between the regions’ supplyand demand and brought to the world market (pooledapproach, no bilateral trade). For each product, worldmarket clearing price balances global trade. Worldprices for each product feed back into domestic pricesthrough a set of equations which specify wedges be-tween world price and domestic price. All policiesare inserted as ad valorem price wedges. Finally, the

19 Generally speaking, the specification of supply and demandrelationships in all selected partial equilibrium models has beenconducted in such a way that the own price effects always dominatecross-price effects. In some cases, the homogeneity condition mighthave been imposed. Some partial equilibrium models may satisfyfor some of their components regularity conditions implied byeconomic theory. Given the large size of most of these partialequilibrium models, it has been impossible to undertake a fullscreening of all their supply and demand relationships and tocheck whether they satisfy all the regularity conditions. However,when documentation on these models permits, we identified thecomponents which satisfy regularity conditions.

F.va

nTo

nge

ren

et

al./A

gricu

ltura

lE

con

om

ics2

6(2

00

1)

14

9–

17

2161


standard model is comparative static in nature. In thispaper the standard SWOPSIM model is an exampleof a typical standard partial equilibrium model.

The standard partial equilibrium model can be mod-ified to capture the following elements:

• possibility that policy instruments can be repre-sented explicitly and in a detailed fashion;

• inclusion of autonomous shifters into behavioural(supply and demand) relationships to generate pro-jections.

In general, all the selected models are pretty closeto the standard model. They differ from the standardmodel because they are recursive dynamic (AGLINK,FAO World Model, FAPRI, GAPsi), endogenise landallocation (AGLINK, FAO World Model, WATSIM),model explicitly quantitative policies (AGLINK,ESIM, GAPsi, MISS20 and WATSIM) or includebilateral trade by using the Armington assumption(SWOPSIM, one application). Besides the designchoices the models differ in their product and countrycoverage, which leads to a rather large differences infocus.

3.2. Economy-wide models

As we did in Section 3.1 for partial equilibriummodels, we first define a prototypical model as a pointof reference against which the features of individ-ual economy-wide models can be compared. Individ-ual models are summarised in alphabetical order inTable 3, which also presents their non-standard fea-tures. We choose as our standard a multi-region AGEmodel, which has the following characteristics in termsof the criteria introduced in Section 2.

A standard economy-wide model has the followingfeatures:

• regional scope: global coverage;• regional unit of analysis: parametric differences be-

tween countries/regions;• dynamics: comparative static;• modelling of trade: Armington;

20 The MISS and AGLINK models are the only two partial equi-librium models which are explicitly concerned about regularityconditions in some of their components. For the former model,supply equations satisfy the regularity conditions while for latter,feed demand relationships are theoretically consistent.

• characterisation of global markets: bilateral traderelations;

• representation of policies: ad valorem price wedges(trade: tariff equivalents);

• theoretical consistency: implied by model structure;• model closure: endogenous volumes and prices on

all markets, including factor markets. Exogenous:factor endowments, policy instruments. Macro clo-sure: ‘neo-classical’, savings driven investment atglobal level (endogenous trade balance).

The main features of the standard multi-regionAGE model correspond closely to those attributedby Baldwin and Venables (1995) to ‘first generation’models: comparative static, constant returns to scalein production, perfect competition on all markets,Armington assumptions for imports. In addition, ourstandard model has a database with global coverage,i.e. in principle global economic activity is cov-ered. ‘Standard models’ included in this review areRUNS, GREEN, GTAP and MEGABARE. Withineach regional economy of a standard multi-regionAGE model, inter-industry linkages are captured byan input–output structure. Demand for factors of pro-duction is derived from cost minimisation, given asectoral production function (nested CES) that allowsfor substitution between inputs. Typically, substitu-tion is allowed only between primary factors — land,labour, capital — while intermediate inputs are usedin fixed proportion with output (Leontief technology).Each sector produces one homogeneous good thatis perfectly substitutable domestically but substitutesimperfectly with foreign goods (Armington assump-tion). Next to the binary distinction ‘domestic versusforeign’, the multi-region nature of the model enablesa distinction of traded commodities according theirregion of origin. That is, bilateral trade flows are cap-tured.21 Factor markets for land, labour and capitalare included, endowments for these primary factorsare given and the factors are fully employed.

Labour and capital are assumed to be fully mobileacross domestic sectors, while land is imperfectly

21 One strand of AGE models uses, in addition to Armington styleimports, a CET transformation function that models the split ofdomestically produced goods into exported commodities and thosedestined for the domestic market. An advantage of this method isthat it dampens the size of terms of trade effects that emerge inArmington models, see DeMelo and Robinson (1989).

F.va

nTo

nge

ren

et

al./A

gricu

ltura

lE

con

om

ics2

6(2

00

1)

14

9–

17

2163


mobile and tied to agricultural production. Con-sumer demand is derived from utility maximisationunder a budget constraint, and consumers allocatetheir expenditures over domestic and foreign goods.A government actor levies various types of indirecttaxes and subsidies including import tariffs and exportsubsidies. All policy instruments are specified as advalorem price wedges. All factor markets and com-modity markets are assumed to clear, which yieldsequilibrium solutions to factor- and commodity pricesas well as the corresponding equilibrium quantities.

All regional economies are linked through bilateralcommodity trade and through interregional investmentflows. As discussed in Section 2.2.5, there are differ-ent approaches to deal with this latter aspect. If one iswilling to assume a constant current account balancein all regions, then the difference between regionalsavings and investments is essentially predetermined,and as a consequence the aggregate level of the sav-ings — investment balance is also predetermined. Ifone wants to allow for endogenous determination ofthe current account balance, the standard model mustinclude a mechanism to redistribute aggregate savingsover regions.

We also classify under the heading ‘standard model’those which include a recursive sequence of tempo-rary equilibria.22 Recursive models do generate timepaths for endogenous variables, but there is in fact nobehavioural linkage between periods. As a result, theequilibrium solution in each period can essentially becalculated without reference to earlier or later periods.

‘Second generation’ models add increasing returnsand imperfect competition in some of the sectors, al-lowing for estimates of scale and variety effects, asdiscussed in Section 2.2.2. These models are com-parative static in nature. Examples included here arethe Michigan BDS and WTO models. In contrast,‘third generation’ models include time consistent for-ward looking behaviour and endogenous savings rates,hence allowing for the modelling of short run dynam-ics. The G-cubed model is an example of this brand.

22 The Baldwin and Venables (1995) ‘first generation models’ arecomparative static. However, single region recursive AGE modelshave a long tradition, starting from the work of Adelman andRobinson (1978) on Korea. While the standard recursive approachallows for accumulation of capital stocks, investment behaviour isnot forward looking in these models.

Most of selected economy-wide models — whetherdynamic or static — share in common the fact thattheoretical consistency is met by adopting restrictivestructures for consumers’ preferences and firms’ tech-nologies. Hence, the former are often represented byexplicitly additive structures while firms’ technologiesare also specified using (nested) CES functions. Withthe exception of GTAP, all the selected applied generalequilibrium models are specified with (nested) CESfunctions, and linear expenditure systems. Less restric-tive flexible functional forms have been proposed toallow for different degrees of substitutability betweenfactors of production or between consumption goods(see Patridge and Rickman (1998) for a discussion).When incorporating flexible functional forms into anAGE model, care has to be exercised that the param-eterisation meets certain curvature properties to en-sure existence and uniqueness of equilibria. The GTAPmodel is the only one which attempts to adopt a lessrestrictive structure to represent consumer expendi-tures. In fact, consumers’ preferences in this modelare specified through a non-homothetic constant dif-ference of elasticities (CDE) demand system (Hanoch,1975) which allows budget shares to vary with income.

The standard, ‘first generation’ multi-regionalAGE model is a firmly established workhorse ininternational trade analysis. While retaining mostof the standard assumptions, certain special fea-tures are introduced into some models to capturespecific issues, such as developing country agricul-ture (RUNS) or aspects of the EU’s CAP (someGTAP applications). Recursive dynamic variationsof the standard model are now commonplace inresearch in the field of global climate change(GREEN, MEGABARE). Imperfect competitionversions have gained ground in trade liberalisationof manufactures, and are likely to be used in theassessment trade liberalisation in services (WTO,BDS, GTAP). The most recent development is theintertemporal modelling of macroeconomic inter-actions between financial markets and real sectors(G-cubed).

The size of the data collection effort for globalmodels has in the past forced modellers to be rathereconomical as regards the regional and sectoral dis-aggregation. Two collaborative efforts to reduce thisentry barrier exist to date: INFORUM and GTAP. TheGTAP database is specifically tailored to the needs of


general equilibrium modellers, and this has certainlycontributed to its wider usage, also by non-GTAPmodelling teams.

4. Assessment

We started this survey with the claim that no modelcan serve all purposes. Following the criteria set outin Section 2, Table 4 gives an overview of the designchoices made in the surveyed models, and serves as anaid to get an overview of the current state of the field.

Eight out of the 16 surveyed models are partialmodels, according to Table 4. Results obtained from

Table 4Basic modelling design choicesa

Partial Economy TotalModels wide

models

Scope of representationNational economies

Partial 8 0 8General 0 8 8

Regional scopeGlobal coverage 8 7 15Non-global coverage 0 1 1

Regional unit of analysisLinked country models 0 1 1Parametric differences 8 7 15

DynamicsStatic 4 3 7Recursive dynamic 4 4 8Forward looking 0 1 1

Modelling of tradeHomogeneous 8 0 8Armington 0 5 5Monopolistic competition 0 2 2Other 0 1 1

Treatment of quantitative policiesTariff/price equivalents 3 5 8Explicit treatment 5 3 8

Data: public data availability?Yes 3 5 8No 5 3 8

ParametersEstimated 2 0 2Calibrated 6 8 14

a The table refers only to standard versions of models.

a general equilibrium analysis will only differ signif-icantly from partial equilibrium results if agriculturaltrade policies lead to noticeable price shifts in othersectors. However, in industrial countries agricultureaccounts for only a small share of GNP. Therefore, thestrength of the linkages of agriculture with other sec-tors is typically not very strong at the level of aggre-gation that AGE models tend to employ. An exceptionmay be those linkages than run through markets fornatural resources, especially land. In contrast, devel-oping countries and countries in transition witness arelatively high share of agriculture in economic activ-ity. There are, therefore, significant second-round ef-fects to be expected from polices that pave the groundtowards regional integration and multilateral trade lib-eralisation, and AGE models provide the only coherentway to analyse these. More generally, policy changessuch as CAP reform and WTO agreements are asso-ciated with impacts that reach beyond the agriculturalsector and involve effects on factor markets for landand labour, which can most fruitfully be studied in ageneral equilibrium framework.

In industrialised countries, there do exist stronglinkages, however, with sectors that are closely relatedto agriculture, either because they deliver key inputssuch as fertilisers, herbicides, agricultural machinery,or because they process primary agricultural products,such as beef processing and dairy industries. High-lighting such interdependencies within the agriculturalcomplex is one area where partial equilibrium modelscan potentially be very successfully used, and someof the recent partial models have taken up this chal-lenge (WATSIM, ESIM). This aspect is also gainingimportance in the presence of dramatically increasingtrade shares of processed food products. Most of thepartial equilibrium models surveyed in this article donot fully exploit this potential advantage because theyhave a focus on trade in primary agricultural com-modities. As a result, there has been a tendency touse AGE models to highlight the forward and back-ward linkages within food supply chains, as well asto incorporate trade in differentiated food products.

The majority of the models has a global coverage,only one of them treats a regional subset of economies(INFORUM). Within the group of models that closestheir accounting with respect to world trade, there aredifferences in regional emphasis. FAPRI focuses onthe US, ESIM on Eastern Europe, MISS focuses on


US–EU interactions, GAPsi emphasises the EU. Aclear regional bias is less obvious in the economy-widemodels with a global coverage. All of them include atleast the major trading regions (US, EU, Asia Pacific).

The commodity coverage of partial models putsmore emphasis and detail on agricultural commodi-ties. Most AGE models include only 1–3 agriculturalsectors. RUNS and GTAP are exceptions in this re-gard. The recent version of the GTAP database hasan amount of agricultural detail that is comparable topartial agricultural models.23

Only one of the models, INFORUM, features linkedindividual country models, while all others favour rep-resentation of differences between economies via dif-ferences in parameters. While in principle, individ-ual country models can capture more regional eco-nomic and institutional details, there are clear dif-ficulties with this approach in terms of consistencyand maintenance, see Section 2.1.4. Indeed, the linkedcountry models approach seems to be less sustain-able, and their contribution to global trade analysis hasbeen rather limited. (The IIASA Basic Linked System,Parikh et al., 1988; The project LINK, Klein and Su,1979).

Comparative static modelling has certainly not goneout of fashion, although eight models favour a recur-sive dynamic approach which permits them to gen-erate time paths of variables and lagged adjustmentpatterns. Forward looking time consistent behaviouris only introduced into one model, G-cubed, whichdoes not have a specific agricultural focus, but concen-trates more on macroeconomic phenomena. Explicitintroduction of time is certainly appealing to policyusers of models, since this relates the model outcomesto concrete time periods. Comparative static modelshave reacted to this demand by generating projectionswithout explicit modelling of the dynamics, see Sec-tion 2.2.1. While this procedure has some appeal, itis also not free of criticism, and some caution shouldbe exercised. Partial models have to make assump-tions on the development of a large number of exoge-nous variables to produce a projected future dataset.In fact, the largest part of the projected future doesnot derive from the model, but from outside assump-

23 At the time of writing, version 4 is the current version of theGTAP database (McDougall et al., 1999). Version 5 is expectedto become publicly available during 2000.

tions. Since the partial model itself does not providea consistency check, it is questionable whether theseassumptions are always consistent among each other.Projections with static general equilibrium models doprovide a consistency check, but these models rely onan extremely small number of assumptions for theirprojections. This implies that a large part of the stepbetween two time periods is ‘explained’ by residualfactors such as total factor productivity (TFP) growthrates which accumulate much of deviations not in-cluded in the original model. Finally, the features ofthe ‘baseline’ in all dynamic models as well as inprojections are critical for the interpretation of policyresults which are obtained relative to the constructedbaseline scenario.

It is striking to note that all partial equilibrium mod-els treat international trade in homogeneous products,while AGE models deal with trade in differentiatedproducts by default. As already mentioned above, thevolume of trade in processed food products is increas-ing relative to trade volumes in primary commodities.Since processed food can be considered to be of amore differentiated nature than primary products, it ishighly relevant to come to grips with trade in differen-tiated products. By excluding intra-industry trade, andlimiting the analysis to net trade, partial models cap-ture the degree to which countries are interwoven onlyimperfectly. If net trade in a certain commodity turnsout to be zero, two economies are unduly qualified asunlinked if in fact there exist intra-industry trade re-lations. These models also run the risk of predictingthe empirically contestable phenomenon of extremespecialisation. Net trade in homogeneous goods alsomakes it impossible to incorporate bilateral trade poli-cies. While the standard treatment of trade in differ-entiated products follows the Armington specification,two AGE models (BDS, WTO) incorporate firm-levelproduct differentiation and economies of scale by de-fault, and the standard GTAP model has been amendedin that direction. These models focus on manufactur-ing and services, where these phenomena are perhapsmore relevant than in agriculture. However, in foodprocessing industries economies of scale and imper-fect competition aspects are certainly relevant as well.A related issue is FDI by internationally operating pro-cessing and retailing firms. This is as yet untreatedin the applied models surveyed, but does require therecognition of economies of scale at the plant level


as well as at the firm level (Markusen and Venables,1998). Scale and variety effects tend to yield ‘largenumbers’ in trade liberalisation studies. It must berecognised, though, that hitherto the empirical basisfor these industrial organisation issues is rather weak.Cross country econometric evidence on key parame-ters that measure scale economies are not yet avail-able.

Eight models attempt to capture explicitly quanti-tative trade restrictions and CAP-type policies, andeight of the models resort to a tariff-equivalent rep-resentation. Policies are typically formulated at thecommodity level or tariff-line level. It is at this levelthat policy makers need information, and partial mod-els are in principle able to get down to the requiredlevel of detail, including specific institutional arrange-ments. Partial models, with their focus on selectedsectors, are in principle able to give a more preciserepresentation of policies, such as quantitative restric-tions. However, our survey of partial model revealsthat some partial models under-utilise that poten-tial and resort to a tariff-equivalent representation ofpolicies.

The inventory of models shows that some datasetsare used by different models. Usually, modellers ad-just the raw data to suit their specific needs, and con-sequently some duplication of efforts occurs. Eightmodelling teams choose to make there dataset publiclyavailable, either free of charge or at cost. This prac-tice, which is increasingly observed within the mod-elling community, is considered a very useful step asit allows others to build on existing (and time consum-ing) work and it increases the transparency of mod-elling results. Sharing of databases has in the pastbeen hampered by well known public good problems,which provide insufficient incentives for individualteams to contribute to database development. The IN-FORUM network provides an early example of an in-stitutional set-up that facilitates sharing of data. IN-FORUM contributors submit (input–output) data in aform that matches their particular country model, anddoes therefore not require major adjustments to a com-mon standard. In contrast, the GTAP framework en-forces uniform standards on regional data and tradedata. In addition, GTAP is supported by a strong groupof institutional stakeholders which puts high require-ments on the quality, timeliness and documentation ofthe data.

It turns out that 14 of the models surveyed here relyon calibration methods, and take their initial parameterestimates from the same published sources that some-times date back a considerable time. Current modelsare dominated by ‘theory’ over ‘observations’. Econo-metric estimation of key behavioural parameters inapplied models is certainly an underdeveloped area,although there are some initiatives to estimate par-tial models in consistence with micro-economic the-ory (ESIM, FAPRI). Recent developments in entropyestimation methods may help to alleviate some of thetechnical problems that one encounters in estimatinglarge scale AGE models with limited data (see Golanet al., 1996).

Although not apparent from our earlier discussions,documentation of models is generally weak and scat-tered, with some notable exceptions (BDS, G-cubed,GTAP). Especially agency based models do not standout by clarity of documentation. Modellers that arerooted in academia face stronger incentives to submittheir work to peer reviews, which increases trans-parency. An important related aspect is the accessi-bility of models and data to outside users, who do notbelong to the organisations or bodies which have (ini-tially) financed or sponsored the development of thesemodels. While eight models offer the possibility toobtain their datasets, the models themselves are oftenproprietary. However, some of the models which arepresented in this review can be considered as ‘publicgoods’ (conditional on certain costs and guarantees)which can be used by or made available to interestedorganisations or persons. Thus, the SWOPSIM modeldeveloped by the economic research service (ERS)of USDA has been made available to numerous aca-demics who worked on the impact of agriculturaltrade liberalisation. The OECD AGLINK model ispresently used by government services of OECDmember countries. A part of the INFORUM modelsand modelling tools are in the public domain. At thepresent time, GTAP represents the most far reachingattempt to public availability, and has now severalhundred users in the academic community as well asin research agencies all over the world.

Building an applied trade model is a costly ex-ercise, which tends to require several man-yearsof dedicated work on database construction, theoryformulation, parameter estimation and computer im-plementation. In addition, the size of the investment


implies that the basic design choices are to a largeextent irreversible. Once a particular route has beenchosen, the switching cost may become prohibitive.Some developments point towards a further reductionin entry costs to this type of work: (a) convergencetowards standards in model building, where new mod-els can build on established blueprints. (b) A major,and seldom fully appreciated, part of model buildingis devoted to database construction. GTAP has pio-neered institutional innovations that lower the costsassociated with database construction and databasemaintenance considerably. (c) The availability ofpowerful general purpose software packages rendersit obsolete to develop own software to solve largescale models numerically. Additional advantages ofusing packages like GAMS, GEMPACK or GAUSSis the transferability, reproducibility (and thereforecross-checking) of models and ease of maintenance.Early partial equilibrium models have been imple-mented in spreadsheets, which was top technology atthe time. Except for small scale models, and modelsfor pedagogic purposes, spreadsheet models do nothave much to commend them. They are inherentlydifficult to maintain and are very error-prone.

4.1. Classification of the reviewed global models

The assessment of global models has been con-ducted in a qualitative fashion using the set of criteriadefined in Table 4. From this analysis, is it possibleto give a classification of these 16 models which dis-tinguishes groups of models having the same profiles?An attempt has been made here by identifying and or-dering the set of criteria which discriminate best be-tween the possible groups of models. In addition to theclear distinction between partial and economy-widemodels, the other discriminating criteria are whetherthe models are synthetic or econometrically-estimated,dynamic or static, currently used or shelved and inthe case of economy-wide models whether they havea global or non-global coverage and how they specifyinternational trade.24 The results are presented in a

24 The classification process has been supported by a multivariatefactor analysis called ‘multiple correspondence analysis’ (MCA),which is especially designed to process datasets consisting ofqualitative variables with several modalities (Escoffier and Pages,1990). The coding and listing of variables (not presented in this

hierarchical tree-like diagram which provides an intu-itive visual presentation of the model groupings.

Fig. 1 presents the classification of the 16 models.Each branch in the figure is identified with the variablewhich discriminates between groups. The hierarchi-cal structure of models clearly distinguishes in a firststage partial equilibrium from economy-wide models.Within these two broad groups, several sub-groupsare easily identified. Looking first at the partial equi-librium model group, we can identify three pairs ofmodels. The first pair is made up of the FAO andFAPRI models which are econometrically estimateddynamic recursive models. Within the models with cal-ibrated parameters, the second pair of models consistsof static models, such as SWOPSIM, WATSIM, ESIMand MISS, and the third pair is made up of recursivedynamic models, such as GAPSI and AGLINK.

A similar analysis of grouping can be conducted foreconomy-wide models, where we observe that the IN-FORUM model stands on its own. As already observedearlier, this model has a profile which is quite dif-ferent from all other economy-wide models. Not sur-prisingly, the seven, remaining economy-wide mod-els are gathered in groups which can be defined alongthe classification of first-, second- and third-generationmodels. Hence, the model G-cubed is the only thirdgeneration model (forward looking behaviour) whilethe pair formed by the models BDS and WTO are sec-ond generation models (imperfect competition). Thefour remaining economy-wide models can be viewedas first-generation models. Within the latter group thestandard GTAP model is static whereas the GREEN,MEGABARE and RUNS models are recursive dy-namic.

5. A modeller’s research agenda

The distinction between partial- and general equi-librium models runs as a red thread through this article.We have emphasised that both approaches have theirrelative merits and are suited for certain policy ques-tions. We believe that further improvements of both

paper but available upon request from the authors) used in thismultivariate factor analysis has been conducted using the criteriadeveloped in Table 4.


Fig. 1. Classification of reviewed global models.

types of models are desirable, and further researchefforts might be put into the following directions.

As far as partial equilibrium models are concerned,the trade-off between pragmatism and theoretical el-egance is currently biased towards practical usability.Improvements of the theoretical structure of mostpartial models are desirable, not only from a purelyacademic point of view but also in order to avoidmisleading policy conclusions. For example, a modelthat does not guarantee the equality between incomesand expenditures because of an ad hoc treatment ofdemand parameters is likely to lead to wrong conclu-sions. This is especially the case if the model is usedfor longer term projections where income levels andbudget shares are likely to change substantially.

As far as the contribution to the current policydebate is concerned, the partial models will needto take up policy issues related to non-market in-teractions, such as the policy discussion aroundmulti-functionality of agriculture. This requires seri-ous investigations into the modelling of positive andnegative externalities associated with agricultural ac-tivities. A related policy issue concerns the currentdebate on decoupling of income support measures. Intheir current form, none of the models reviewed here

is capable to shed light on the degree in which alter-native income support schemes affect output, pricesor trade. As mentioned before, the current generationof partial trade models focuses on trade in primaryor first stage processed agricultural commodities. Inview of dramatically increasing trade shares of pro-cessed, and differentiated, products, serious effortstowards a more comprehensive modelling of the sup-ply chain will enhance the usability of partial models.Incorporation of bilateral trade flows opens the routeto modelling bilateral trade instruments, which areof great importance in the upcoming round of multi-lateral trade negotiations. There is no way to modelTRQs without a bilateral trade specification.

In as far as partial models display a bias towardspragmatisms, general equilibrium models have atendency to favour theoretical elegance over rele-vance to detailed policy formulation. The usability ofglobal general equilibrium models will be enhancedif country specific institutional and policy details areincorporated. Introducing more ‘pragmatism’ should,however, not sacrifice their theoretical soundness,which is a major advantage of this class of models.

Another potential area for improvement is the dy-namic specification. For many questions, and for many


users of model results, it is important to sketch thetime path towards a new equilibrium.

Within the general equilibrium modelling commu-nity large efforts are currently on-going in modellingglobal environmental issues, such as climate change.There seems to be scope for integrating economicmodels with ecological models, and some experienceis already gained in this respect (Darwin et al., 1996;Nordhaus and Zhang, 1996).

6. Concluding remarks

Ten years ago, the OECD and the World Bank con-vened a symposium that assessed the ‘state-of the-art’in agricultural trade modelling at that time, see Goldinand Knudsen (1990). The field has changed over thepast decade, but to some extent the comments madeat this symposium can be echoed today. Probably themost important innovations have not been theoretical,nor have they been technological. The most signifi-cant changes have been of an institutional nature, al-beit supported by recent computer and communica-tions technologies. Ten years ago, models, data andsoftware were almost exclusively proprietary. Today,it has become more common to exchange computercode and to share databases. This tendency can be ex-pected to be continued in the future. The ‘open source’concept that spurred rapid innovations in some partsof the software industry may very well be the directiontowards which the global trade modelling communityis heading.

Acknowledgements

The authors would like to thank all persons whocontributed to this survey. This includes first of all themembers of the GTAP-EU concerted action project:Søren Frandsen, Chantal Pohl Nielsen, Michael Stæhr(SJFI, Denmark); Martina Brockmeier, Dirk Mane-gold (FAL, Germany); Joseph Francois, MachielRambout (Erasmus University Rotterdam/TinbergenInstiute, The Netherlands); Risto Vaittinen, LeenaKerkela (Higher School of Economics and BusinessAdministration, Finland); Thomas Ratinger (VUZE,Czech Republic); Kenneth Thomson (University ofAberdeen, UK); Bruno Henry de Frahan, (Univer-

sité Catholique de Louvain, Belgium), Akka Ait ElMekki (Ecole Nationale d’Agriculture de Meknès,Maroc); Luca Salvatici (Universita degli studi diRoma La Sapienza, Italy), Paul Veenendaal (LEI, TheNetherlands). We also thank Rob Peters (EuropeanCommission, DG-AGRI) for comments and fruitfuldiscussions. In addition we sincerely thank all indi-vidual researchers and modelling teams who sharedinformation on their models with us. This research hasbeen financially supported by the European Commis-sion under the Fair-6 and Inco programmes (FAIR6CT 98-4141). The content of this article is the soleresponsibility of the authors and does not in any wayrepresent the views of the European Commission orits services.

References

ABARE, 1996. The MEGABARE model: interim documentation.Australian Bureau of Agriculture and Resource Economics,Canberra.

Adelman, I., Robinson, S., 1978. Income distribution policies indeveloping countries: a case study of Korea. Stanford UniversityPress, Stanford.

Alexandratos, N. (Ed.), 1995. World Agriculture towards 2005. AnFAO Study, Food and Agricultural Organization of the UnitedNations, Rome.

Almon, C., 1991. The INFORUM approach to inter-industrymodelling. Econ. Syst. Res. 3 (1), 1–7.

Armington, P.A., 1969. A theory od demand for productsdistinguished by place of production. IMF Staff Papers 16,159–178.

Baldwin, R.E., Venables, A.J., 1995. Regional economicintegration. In: Grossman, Rogoff (Eds.), Handbook ofInternational Economics, Vol. III. Elsevier, Amsterdam (Chapter31).

Berkum, S. van, van Meijl, H., 2000. The application of trade andgrowth theories to agriculture: a survey. Aust. J. Agric. Resour.Econ. 44 (4), 505–542.

Bremer, S.A. (Ed.), 1987. The GLOBUS model. In:Computer Simulation of Worldwide Polictical and EconomicDevelopments. Campus/Westview Press, Frankfurt and Boulder.

Brown, D.K., 1987. Tariffs, the terms of trade and national productdifferentiation. J. Policy Model. 9, 178–203.

Brown, D.K., Deardorff, A.V., Stern, R.M., 1992. A NorthAmerican free trade agreement: analytical issues and acomputational assessment. The World Econ. 15, 15–29.

Burniaux, J.-M., van der Mensbrugghe, D., 1990. Arural/urban-north/south general equilibrium model foragricultural policy analysis. OECD Economics DepartmentTechnical Papers no. 33, Paris.

Cahill, C., Legg, W., 1990. Evaluation de l’aide à l’agricultureau moyen des équivalents subvention à la production et à


la consommation théorie et pratique, Revue Economique del’OCDE, Vol 13. Numéro spécial sur la modélisation desconséquences des politiques agricoles, pp. 16–46.

Conrad, K., 1999. Computable general equilibrium modelsfor environmental economics and policy analysis. In: vanden Berg, C.J.M. (Ed.), Handbook of Environmental andResource Economics. Edward Elgar, Cheltenham, Chapter 69,pp. 1060–1089.

Darwin, R., Tsigas, M., Lewandrowski, J., Raneses, A., 1996. Landuse and cover in ecological economics. Ecol. Econ. 17 (3),157–181.

DeMelo, J., Robinson, S., 1989. Product differentiation and thetreatment of foreign trade in computable general equilibriummodels of small economies. J. Int. Econ. 27, 47–67.

DeMelo, J., Tarr, D., 1992. A general equilibrium analysis of USForeign Trade Policy, MIT Press, Cambridge.

Dixit, A.K., Stiglitz, J., 1977. Monopolistic competition andoptimum product diversity. Am. Econ. Rev. 67, 297–308.

Escoffier, B., Pagès, J., 1990. Analyses factorielles simples etmultiples. Objectifs, méthodes et interprétation. Dunod, Paris.

Ethier, W., 1979. Internationally decreasing costs and world trade.J. Int. Econ. 9, 1–24.

Ethier, W., 1982. Decreasing costs in international trade and FrankGraham’s argument for protection. Econometrica 50, 1243–1268.

FAO, 1993. The world food model — Model Specification. FAOMimeograph, ESC/M/93/1, Rome.

Faucheux, S., Levarlet, F., 1999. Energy–economy–environmentmodels. In: van den Berg, C.J.M. (Ed.), Handbook ofEnvironmental and Resource Economics. Edward Elgar,Cheltenham, Chapter 73, 1060–1089.

Francois, J.F., Reinert, K.A., 1997. Applied Methods for TradePolicy Analysis: A Handbook, Cambridge University Press,Cambridge.

Francois, J.F., McDonald, B., Nordstrom, H., 1995. Assessingthe Uruguay Round. In: Martin, W., Winters, L.A. (Eds.),The Uruguay Round and Developing Economies. CambridgeUniversity Press, Cambridge.

Frenz, K., Manegold, D., 1988. Lineares Modell zur Bestim-mung von Preis- und Mengenwirkungen einer restriktivenAgrarpreispolitik in der EG, Arbeitsbericht des Instituts fürlandwirtschaftliche Marktforschung 88/1. Braunschweig.

Frohberg, K., Hartmann, M., Weingarten, P., Fock, A., Wahl,O., 1997. The Central European agricultural simulation model(CEASIM): an overview. Research Report of the Instituteof Agricultural Development in Central and Eastern Europe(IAMO), Halle/Saale, unpublished document.

Golan, A., Judge, G., Miller, D., 1996. Maximum EntropyEconometrics: Robust Estimation with Limited Data, Wiley,Chichester.

Goldin, I., Knudsen, O. (Eds.), 1990. Agricultural tradeliberalisation: implications for developing countries. Organis-tion for Economic Co-operation and Development, Paris.

Grossman, G.M., Helpman, E., 1991. Innovation and Growth inthe Global Economy. MIT Press, Cambridge.

Hanoch, G., 1975. Production and demand models with direct orindirect implicit additivity. Econometrica 43, 395–419.

Henrichsmeyer, W., von Lampe, M., Möllmann, C., 1998.Weiterentwicklung und Andwendung des Welt-Agrarhandels-modells WATSIM für Langfristsimulationen der Welta-grarmärkte sowie der Auswirkungen für die Landwirtschaft derEU und der Bundesrepublik Deutschland unter Einbeziehungdes Modellsystems RAUMIS. Interim Report for the FederalMinistry for Food, Agriculture and Forestry, Bonn.

Hertel, T.W. (Ed.), 1997. Global Trade Analysis: Modelling andApplications. Cambridge University Press, Cambridge.

Huff, H.B., Moreddu, C., 1990. The ministerial mandate trademodel (MTM): OECD economic review, Vol. 13. Special issueon Modeling the effects of Agricultural policies.

Jorgenson, D.W., 1984. Econometric methods for applied generalequilibrium modelling. In: Scarf, Shoven (Eds.), AppliedGeneral Equilibrium Analysis. Cambridge University Press,Cambridge (Chapter 4).

Klein, L.W., Su, V., 1979. Protectionism: an analysis from projectLINK. J. Policy Model. 1 (1), 5–35.

Krugman, P.R., 1979. Increasing returns, monopolistic competition,and international trade. J. Int. Econ. 9, 469–479.

Krugman, P.R., 1980. Scale economies, product differentiation andthe pattern of trade. Am. Econ. Rev. 70, 950–959.

LaFrance, J., 1986. The structure of constant elasticity demandmodels. Am. J. Agric. Econ., 543–552.

LaFrance, J., Hanemann, W.M., 1989. The dual structure ofincomplete demand systems. Am. J. Agric. Econ., 262–274.

Laird, S., 1997. Quantifying commercial policies. In: Francois,J.F., Reinert, K.A. (Eds.), Applied Methods for TradePolicy Analysis: A Handbook. Cambridge University Press,Cambridge.

Lancaster, K., 1980. Intra-industry trade under perfect monopolisticcompetition. J. Int. Econ. 10, 151–175.

Lee, H., Oliviera-Martins, J., van der Mensbrugghe, D., 1994. TheOECD Green model: an updated overview. OECD EconomicsDepartment Technical Papers no. 97, Paris.

Lucas, R.E., 1976. Econometric policy evaluation: a critique. J.Monetary Econ. 1 (2), 19–46.

Mahé, L., Tavéra, C., 1989. Bilateral harmonization of EC andUS agricultural policies. Eur. Rev. Agric. Econ. 15, 325–348.

Markusen, J.R., Venables, A.J., 1998. Multinational firms and thenew trade theory. J. Int. Econ.

McDougall, R.A., Elbehri, A., Truong, T.P. (Eds.), 1999. Globaltrade, assistance and protection: the GTAP 4 database. Centerfor Global Trade Analysis, Purdue University, West Lafayette.

McKibbin, W., Wilcoxen, P., 1999. The theoretical and empiricalstructure of the G-cubed model. Econ. Model. 16, 123–148.

Meilke, K., McClatchy, D., De Gorter, H., 1996. Challenges inquantitative economic analysis in support of multilateral tradenegotiations. Agric. Econ. 14, 185–200.

Nordhaus, W.D., Zhang, Z., 1996. A regional dynamicgeneral-equilibrium model of alternative climate changestrategies. The Am. Econ. Rev. 86 (4), 741–765.

OECD, 1998. OECD workshop report on economic modelling ofclimate change. OECD Headquarters, Paris, 17–18 September,1998.

Parikh, K.S., Fischer, G., Frohberg, K., Guldbrandsen, O., 1988.Towards free trade in agriculture, Matinus Nijhoff, Dordrecht.


Patridge, M.D., Rickman, D.S., 1998. Regional computable generalequilibrium modeling: a survey and critical appraisal. Int.Regional Sci. Rev. 21 (3), 205–248.

Roningen, V.O., 1986. A static world policy simulation(SWOPSIM) modeling framework. Staff Report, EconomicResearch Service, US Department of Agriculture, Washington,DC.

Roningen, V.O., Liu, K., 1983. The world grain, oilseedsand livestock (GOL) model — background and standardcomponents. ERS Staff Report no. AGES830317, Econ. Res.Serv., US Department of Agriculture.

Rosegrant, M.W., Agcaoili-Sombilla, M., Perez, N.D., 1995.Global food projections to 2020: implications for investment.2020 Vision for Food, Agriculture, and the EnvironmentDiscussion Paper 5, International Food Policy ResearchInstitute, Washington, DC.

Spence, M., 1976. Product differentiation and welfare. Am. Econ.Rev. 66, 407–415.

Tangermann, S., Josling, T.E., 1994. Pre-accession agriculturalpolicies for Central Europe and the European Union Studycommissioned by DG I of the European Commission.Göttingen.

Tongeren, F.W. van, van Meijl, H. (Eds.), 1999. Review of appliedmodels of international trade in agricultural and related resourceand environmental modeling. Agricultural Economics ResearchInstitute, Report 5.99.11 (EU-FAIR VI-CT 98-4148, Interimreport 1).

Tyers, R., Anderson, K., 1992. Disarray in World Food Markets:A Quantitative Assessment. Cambridge University Press,Cambridge.

Waino, J., Gibson, P., Whitley, D.B., 2000. Implementation ofUruguay Round tariff reductions. ERS/USDA, AgriculturalOutlook, November 1999.

Winters, A.L., 1990. The role of partial equilibrium agriculturalmodels. In: Goldin, I., Knudsen, O. (Eds.), AgriculturalTrade Liberalisation: Implications for Developing Countries,pp. 453–454.

World Trade Organisation (WTO), 1998. Annual Report, Geneva.Young II, R.E., Westhoff, P.W., Brown, D.S., Adams, G., Womack,

A.W., 1999. Modeling the world agrifood sector: the Food andAgricultural Policy Research Institute approach. Paper Preparedfor The Agricultural Economics Society, Annual Conference1999. Stranmillis University College, The Queen’s University,Belfast, 28 March 1999.

Documents

Global models applied to agricultural and trade policies ... · Global models applied to agricultural and trade policies: a review and assessment Frank van Tongerena, Hans van Meijla,