1 GTAP-AGR : A Framework for Assessing the Implications of Multilateral Changes in Agricultural Policies Roman Keeney and Thomas W. Hertel2 GTAP Technical Paper No.24 August, 2005 1 Pronounced gee-tap-ag, the ‘R’ is silent. 2 The authors thank Hsin Huang, Hans Jensen, Wusheng Yu, and Allen Rae for valuable comments on an earlier draft of this paper. The authors also gratefully acknowledge partial support for this research under Coop Agreement No. 43-3AEK-3-80053 between the U.S. Department of Agriculture, Economic Research Service and Purdue University GTAP-AGR Specific Sets Set Name3 Description Elements4 AGRI_COMM Farm Level Sectors pdr, wht, gro, v_f, osd, c_b, pfb, ocr, ctl, oap, rmk, wol FOOD_COMM Food Processing Sectors cmt, omt, vol, mil, pcr, sgr, ofd, b_t LSTK_COMM Livestock Prod. Sectors ctl, oap, rmk pdr, wht, gro, v_f, osd, c_b, pfb, cmt, omt, vol, mil, pcr, FEED_COMM Feedstuffs Commodities sgr, ofd PROG_COMM Program Commodities pdr, wht, gro, osd, c_b Non-Agricultural NAGR_COMM PROD_COMM less AGRI_COMM elements Commodities 3 For standard GTAP sets and their composition, please see McDougall and Dimaranan (2002). 4 For definitions of these codes for aggregate commodities, please see Table 1 describing the commodity/sector aggregation. For more information regarding the composition of the aggregate commodities/sectors, please see McDougall and Dimaranan (2002). i List of Tables Title Page Number Table 1. Sector Aggregation 33 Table 2. Regional Aggregation 34 Table 3. Factor Supply Parameters 35 Table 4. CES Substitution Parameters in Agriculture 36 Table 5. Estimates of Feedstuff Substitution 37 Table 6. Private Consumption Mapping 38 Table 7. Own-Price Elasticities in Private Consumption 39 Table 8. Income Elasticities in Private Consumption 39 Table 9. Trade Elasticity Comparison 40 Table 10. Farm Household Income Share 40 Table 11. GE Elasticities for US and Canada 41 Table 12. GTAP-AGR GE Elasticity Decomposition: Canada 41 Table 13. GTAP-AGR GE Elasticity Decomposition: USA 42 Table 14. GTAP GE Elasticity Decomposition: USA 42 Table 15. Percent Change in Exports by Commodity 43 Table 16. Percent Change in Exports by Region 44 Table 17. Percent Change in Labor Employment 45 Table 18. Percent Change in Farm Income 46 Table 19. Percent Change in Regional Welfare 47 Table 20. Welfare Decomposition 48 GTAP-AGR Figures Title Page Number Figure 1. Factor Market Segmentation 49 Figure 2. Production Technology for Agricultural Sectors 50 Figure 3. Feedstuffs Substitution for Livestock Sectors 51 Figure 4. Two Stage Budgeting of Seale et al. 52 Figure 5. Eliciting Farm Level Elasticities 53 Figure 6. Incidence Comparison GTAP and GTAP-AGR 54 Figure 7. Model Validation 55 ii GTAP-AGR: A Framework for Assessing the Implications of Multilateral Changes in Agricultural Policies GTAP Technical Paper No. 24 Roman Keeney and Thomas W. Hertel Abstract Global models of agricultural trade have a long and distinguished history. The introduction of the GTAP data base and modeling project represented a significant advance forward as it put modelers and trade policy analysts on common ground. After an initial generation of GTAP based modeling of agricultural trade policy using the standard modeling framework, individual researchers have begun introducing agricultural specificity into the standard modeling framework in order to better capture the particular features of the agricultural economy pertinent to their research questions. This technical paper follows in that same tradition by reviewing important linkages between international trade and the farm and food economy and introducing them into the standard GTAP modeling framework, offering a special purpose version of the model nicknamed GTAP-AGR. We introduce this agricultural specificity by introducing new behavioral relationships into the standard GTAP framework. We focus particular attention on the factor markets, which play a critical role in determining the incidence of producer subsidies. This includes modifying both the factor supply and derived demand equations. We also modify the specification of consumer demand, assuming separability of food from non-food commodities. Finally, we introduce the important substitution possibilities amongst feedstuffs used in the livestock sector. Where possible we support these new behavioral relationships with literature-based estimates of both the mean and standard deviation of behavioral parameters. The express purpose of this is to support systematic sensitivity analysis with respect to policy reform scenarios performed with GTAP- AGR. In addition to documenting these extensions to the standard modeling framework, the paper has an additional goal, and that is to gauge the performance of the GTAP-AGR model and how it differs from the standard GTAP framework. We do this primarily by comparing the farm level supply and demand response in terms of policy incidence for the two frameworks. In addition, we evaluate the ability of both models to reproduce observed price volatility in an effort to validate the performance of these models. Finally, we evaluate the results of the two models in a side-by- side comparison of results from full liberalization of agricultural and non-agricultural support. 3 1. Introduction There is a long and distinguished history of global modeling of agricultural markets and the impact of developed country liberalization on the developing world (van Tongeren et al. 2001). Some of the earliest work in this area was that of Valdes and Zietz (1980) who conducted a highly disaggregated, commodity by commodity analysis of the developing country impacts of trade reform. Following in this tradition, extensive work was undertaken at ERS/USDA in the 1980’s using the multi-commodity, partial equilibrium SWOPSIM framework (Roningen and Dixit, 1989; Krissoff, Sullivan and Wainio, 1989). Meanwhile, the econometrically-based, partial equilibrium analysis of Tyers (1985) and Tyers and Anderson (1986) contributed effectively to raising public awareness of the “disarray in world food markets” due largely to developed country policies. In Europe, and especially at the OECD, the general equilibrium modeling work of Burniaux and his co-authors also had a substantial impact in the 1980’s and early 1990’s (Burniaux and Waelbroeck, 1985; Burniaux et al., 1990; Burniaux and van der Mensbrugghe, 1991). At the International Institute for Applied Systems Analysis (IIASA) there was an important project seeking to build a global model “from the ground up” by linking a series of unique national general equilibrium models sponsored by IIASA, and culminating in the publication of the volume by Parikh, Fischer, Frohberg, and Keyzer (1988). All of these efforts represented multi-year, indeed sometimes decade-long, commitments by the authors to gather data, program models, estimate parameters and conduct policy simulations to assess the impact of agricultural policies on world food markets. Each one of these models had its own unique features, seeking to capture different key aspects of world food markets. For example, Tyers and Anderson emphasized the dynamics of supply and world- domestic price transmission. Burniaux and Waelbroeck focused heavily on factor markets and rural-urban migration. The IIASA model sought to bring in more of the physical constraints on food production. As such, each of these efforts represented many different “views of the world”. Indeed these models were generally seen as extensions of the authors, as they were rarely used by others who were not co-authors (SWOPSIM being an important exception). Since the mid-1990’s, the analytical landscape has changed dramatically with the advent of the Global Trade Analysis Project – nick-named GTAP (Hertel, 1997). Now, almost all of the individuals and agencies conducting analysis of the global implications of agricultural liberalization make use of the GTAP data base – and a global applied general equilibrium model.5 Proponents of the GTAP approach will argue that this has facilitated great strides in our understanding of global trade due to improved data quality, the associated advancement of greatly improved tools for modeling and analysis, and the widespread replication of results (largely non- existent in the global modeling area prior to GTAP – with the exception of the SWOPSIM work). With regard to agricultural trade in particular, the shift towards general equilibrium modeling has had many advantages, including: (a) greater theoretical consistency, (b) improved welfare analysis, (c) exhaustive coverage of the farm and food complex, and (d) integrated treatment of agriculture and non-agriculture liberalization. However, there have also been disadvantages associated with this GTAP-based, general equilibrium approach to the modeling of agricultural trade. One of these has been the tendency to abstract from specific structural features that characterize global food and agricultural markets. Critics argue that the GTAP-based models are overly simplistic and do not capture many 5 A few examples are: (Anderson et al. (2001).; Diao et al. (2003); Francois, van Meijl, and van Tongeren (2003); Frandsen et al. (2002); Harrison et al. (1996)). For a more extensive, but still very partial, listing, visit the GTAP web site: www.gtap.org . A simple search for the keyword “agriculture” turns up more than 100 GTAP applications. 4 important characteristics of the agricultural economy. They also argue that the GTAP parameters need more solid econometric foundations.6 While we count ourselves among the advocates of the GTAP approach, and shudder to think of developing a unique global data base and model for each project/institution, we are also inclined to agree with the criticisms of model structure and parameters. Therefore, the primary purpose of this paper is to outline one approach to redressing some of these concerns, while retaining the advantages of the GTAP-based, global general equilibrium approach. The goal of this paper is two-fold. The first goal is to re-introduce detailed agricultural structure into global general equilibrium trade modeling and underpin this with econometric estimates from the literature. In so doing, we build on recent work by the OECD (2001) which seeks to characterize the degree of factor market segmentation between the farm and non-farm sectors as well as improving the representation of input substitution possibilities in farm production. We also explicitly identify farm households as entities which: (a) earn income from both farm and non-farm activities, (b) pay taxes, and (c) consume food and non-food products based on an explicit utility function. Our consumer demand system is based on recent work by Seale, Regmi and Bernstein (2003) which provides international cross-section estimates of price and income elasticities of demand for food products in more than 100 countries. Finally, we incorporate newly available estimates of trade elasticities7 (Hertel, Hummels, Ivanic and Keeney, 2003). The second goal of this work is to assess the difference in model outcomes that occur due to the altered specification of structure and parameters. We do this in terms of an agricultural liberalization experiment focusing on the three pillars of support, oft discussed in the literature assessing WTO implementations. Specifically, we fully liberalize export subsidies, tariffs, and agricultural domestic support, and focus on the welfare, trade, and price impacts that arise from this experiment with an eye toward identifying the sources of difference in simulation results between the two models. 2. Model Design The GTAP-AGR model represents a special purpose version of the GTAP model, designed to capture certain structural features of world agricultural markets that are not well-reflected in the standard GTAP model – or indeed in most other global trade models currently in use. These features are required in order to provide a more realistic representation of the farm and food system. They are also necessary in order to capitalize on recent econometric work on key elasticities in the global agricultural economy. The discussion of model design is broken into subsections, each dealing with a different aspect of the model. At each stage we discuss both the revised economic theory as well as the parameters used to specify that part of the model. Due to the specificity of this model structure and the associated parameters, GTAP-AGR is no longer as readily flexible with respect to commodity and region aggregation. (The standard GTAP model can be run, without modification, for any commodity aggregation, ranging from 1 to 57, and for any number of regions up to 85 using the version 6 data base.) In particular, the user must either leave the farm and food sector fully disaggregated or she must undertake her own calibration of certain key parameters. The 20 farm and food products as well as the aggregate manufacturing and services sectors are identified 6 See Hertel (1999) for an assessment of GTAP-based analysis of global trade policy in light of John Whalley’s “hidden challenges for AGE analysis” (Whalley, 1986). 7 These trade elasticity estimates are from Hertel et al (2003) and they have recently been incorporated into the public release of GTAP Version 6. 5 in GTAP-AGR are listed in Table 1. Aggregation of non-food activities is less problematic, as the structure here follows that in the standard GTAP model. Non-food activities have been grouped into 6 broad sectors for purposes of the present study. Regional disaggregation requires the user to specify additional parameters – the easiest approach is to let the disaggregated regions inherit the parameters from the parent region. However, it is hoped that the user will supplement the parameter file with estimates from the newly disaggregated focus countries. In the present study, we work with the 23 regions identified in Table 2. 2.1 Standard GTAP: The Point of Departure Our initial point of departure is the GTAP model of global trade (version 6.2). GTAP is a relatively standard, multi-region model which includes explicit treatment of international trade and transport margins, a “global” bank designed to mediate between world savings and investment, and a relatively sophisticated consumer demand system designed to capture differential price and income responsiveness across countries. As documented in Hertel (1997) and on the GTAP web site8, the model includes: demand for goods for final consumption, intermediate use and government consumption, demands for factor inputs, supplies of factors and goods, and international trade in goods and services. The model employs the simplistic but robust assumptions of perfect competition and constant returns to scale in production activities. Bilateral international trade flows are handled using the Armington assumption by which products are exogenously differentiated by origin. This technical paper was under taken with the GTAP 6 database, pre-release 3, available in October of 2004 (see Dimaranan and McDougall, 2005, for documentation of the version 6 data base). We are particularly interested in the specification of domestic support, as the impacts of these subsidies on agricultural production is a contentious issue, and depends importantly on the specification of the agricultural component of the model (OECD, 2001). In the GTAP database, all the different components of OECD PSE data except for market price support are distributed into four classifications of domestic support namely: output subsidies intermediate input subsidies, land-based payments and capital based payments (Jensen, 2002). 2.2 Primary Factor Supply Since the path-breaking work of T.W. Schultz (1945), agricultural economists have recognized the importance of off-farm factor mobility – particularly for labor – in determining farm incomes. In his review of US agriculture, Bruce Gardner (1992) highlights this fact and notes that, in the US, farm and non-farm wages have moved together (with the former being lower throughout) since the second World War – largely as a result of steady off-farm migration of workers. Despite this long-term co-movement of wages, there is significant evidence that wage differentials persist in developed economies and that the policy implications of these can be important (Kilkenny, 1993). The limitations of agricultural labor markets have been prominently featured in the development economics literature, as an explanation for the very low level of agricultural supply response in developing countries (de Janvry et al., 1991). If labor and capital were perfectly mobile between agriculture and non-agriculture, as is commonly assumed in applied general equilibrium models, then we would expect to see wages equalized at each point in time for farm and non-farm workers with comparable skills. However, this is not the case. And in some 8 www.gtap.org 6 countries (China is an extreme example), rural-urban wage differentials are quite large (Zhao, 1999). Ideally, we would like to explain the factor market segmentation in terms of underlying barriers to factor mobility – for example the system of Hukou registration in China. However, successfully explaining this limited farm/non-farm, rural/urban mobility across the full range of countries in our model would be a lifetime project. Instead, we specify a constant elasticity of transformation (CET) function which “transforms” farm-labor into non-farm labor (see Figure 1). There are several important characteristics of this function. Firstly, it is constrained by the total labor endowment in the economy. Increased supplies of labor to manufacturing and services must be drawn from agriculture. This is important, as it will force the economy to respect the aggregate resource constraints. Secondly, with a finite elasticity of transformation, it permits wages to diverge between the farm and non-farm sectors. And thirdly, the elasticity of transformation can be calibrated to replicate any desired (non-negative) elasticity of labor supply to agriculture. This third point is particularly handy in light of the econometric evidence available on this subject which typically comes in the form of such factor supply elasticities. Within agriculture, labor is assumed to be perfectly mobile, and similarly for non-agriculture. In addition to segmentation of labor markets, evidence suggests that the segmentation of capital markets may also be appropriate. Therefore, we also introduce a CET function governing capital movements between agriculture and non-agriculture, with full capital mobility (a unique rental rate on capital) within these respective sectors. Land is specific to agriculture in the GTAP database, and only one type of land is distinguished so the modeling of land supply to alternative agriculture activities is treated with the same CET function as the standard GTAP model where land in a given use is imperfectly mobile amongst agricultural uses9. Equations (1) and (2) below represent the CET agricultural and non-agricultural supply of factors in GTAP-AGR, where the index i represents the mobile endowment commodities (labor and capital) and r is a regional index. qoagr(i,r) = qo(i,r) + ETRAEAGNAG(i,r) * [pm(i,r) - pmagr(i,r)] (1) qonagr(i,r) = qo(i,r) + ETRAEAGNAG(i,r) * [pm(i,r) - pmnagr(i,r)] (2) Variable Description (all variables in percent change) qoagr Supply of endowment to agricultural sectors qonagr Supply of endowment to non-agricultural sectors qo Total Supply of endowment pm Market price for endowment pmagr Market price for endowment in agriculture pmnagr Market price for endowment in non-agriculture Parameter Description ETRAEAGNAG Elast. of Transformation between Ag. and Non-Ag. use 9 Huang et al. introduce multiple land types into their GTAP-PEM model to restrict the mobility of land among agricultural uses. 7 In order to parameterize the agricultural CET supply functions in GTAP-AGR, we draw on the excellent report prepared by the OECD (2001). Among the valuable contributions of this report, the annexes provide extensive econometric literature reviews for the EU (Salhofer, 2001) and for North America (Abler, 2001). These authors provide central parameter values for factor supply elasticities for land, labor, and capital supplied to agriculture (see tables A1.3 and A1.4; OECD, 2001), which we use to calibrate the GTAP-AGR CET supply functions. These elasticities10 are reproduced in Table 3 along with the associated standard deviation. The latter are derived based on parameter ranges supplied in the OECD report, coupled with the assumption that these values follow a symmetric, triangular distribution11. Note that the estimated factor supply elasticities are less than one, which is a sharp contrast to the assumption of perfect factor mobility for labor and capital used in most AGE analyses. This means that commodity supply will also be less price-responsive, and more of the benefits of farm subsidies (or losses from their elimination) will accrue to farm households, as opposed to consumers of the farm products. The OECD report also attempts to come up with supply elasticities for purchased inputs. However, there is little econometric evidence to draw on here. One advantage of the general equilibrium framework offered by GTAP-AGR is that these commodity supply responses are endogenously determined – as a function of the factor market assumptions as well as the cost structure of the industry. Therefore, we dispense with the OECD estimates of input supply for fertilizer and other purchased inputs. The supply prices for intermediate inputs are endogenous in the model and determined by the interaction of supply and demand in each of these markets. 2.3 Derived Demands for Agricultural Inputs On the factor demand side, we employ a nested-CES production function which can be calibrated to the three key elasticities of substitution available from the OECD report (Table 4). Specifically, we postulate that output is produced using a constant elasticity of substitution (CES) production function combining two inputs, which are themselves composite inputs (see Figure 2). The first of these is a purchased input aggregate. This is what distinguishes the GTAP-AGR production function from that in the standard GTAP model. The second composite input is a farm-owned (value-added) aggregate. The individual inputs in each of these groups are assumed to be separable from one another. The equations describing the CES input demands for aggregate value added and purchased inputs by agricultural sectors (i.e. the index j refers only to elements of the set AGRI_COMM) follow as (3) and (4). Demands for individual inputs (endowments and intermediates) are given in (5) and (6). In all equations below j refers to an element of AGRI_COMM, r refers to an element of REG, and i refers to an element of TRAD_COMM. The purchased input and farm-owned aggregates are themselves each a CES function of individual farm inputs, the latter corresponding to the value-added aggregation function in the standard GTAP model. This gives us a total of three CES substitution parameters which need to be calibrated. They are calibrated to the OECD central values for the Allen partial elasticities of substitution between: (i) land and other farm-owned inputs (ESUBVA), (ii) land and purchased 10 As a reviewer points out we assume the same transformation elasticity for skilled and unskilled labor even though the expectation is that unskilled labor is more mobile. Our assumption is driven by the lack of econometric evidence to support differential transformation frontiers for the two types of labor employed in agriculture in the GTAP database. 11 The conversion of a mean and lower bound to a standard deviation for the symmetric triangular distribution follows the formula σ=(mean−lowerbound)/ 6 8 inputs (ESUBT), and (iii) among purchased inputs (ESUBPURCH). The values we use for regions in the GTAP-AGR model are based on the OECD (2001) report and are presented in Table 4 along with the implied standard deviation assuming a symmetric triangular distribution12. The user should note that in the parameter file we make use directly of the Allen elasticities and include calibration equations within the model code to determine the actual ESUBVA and ESUBPURCH parameters for sectors in the set AGRI_COMM13. qva(j,r) = qo(j,r) - ESUBT(j,r) * [pva(j,r) - ps(j,r)] (3) qpurch(j,r) = qo(j,r) - ESUBT(j,r) * [ppurch(j,r) - ps(j,r)] (4) qf(i,j,r) = qpurch(j,r) - ESUBPURCH(j,r) * [pf(i,j,r) - ppurch(j,r)] (5) qfe(i,j,r) = qva(j,r) - ESUBVA(j,r) * [pfe(i,j,r) - pva(j,r)] (6) Variable Description (all variables in percent change) qva Demand for farm-owned aggregate input qpurch Demand for purchased inputs aggregate qf Demand for individual intermediate input qfe Demand for individual endowment input qo Sector output pva Price index for farm-owned aggregate ppurch Price index for purchased input aggregate pf Price of individual intermediate input pfe Price of individual endowment input ps Supply price of output Parameters Description ESUBT Elast. of Sub. between farm-owned and purchased inputs ESUBPURCH Elast. of Sub. among purchased inputs ESUBVA Elast. of Sub. among farm-owned inputs Adapting the production parameter ranges from the OECD (2001) report, which covers just six countries, to the 23 regions in the GTAP-AGR model requires a mapping from 6 to 23 regions. We specify this mapping based on similarities in agricultural economies and regions. There is limited evidence available on what supply and substitution elasticity values might be in developing countries. In GTAP-AGR, we simply set the parameter values in all non-OECD countries equal to those for Mexico. It is hoped that future authors will remedy this gap by providing additional 12 Ibid. 13 This facilitates sensitivity analysis since we key all agricultural technology model parameters off of the six regional parameters given by OECD (2001). This allows a complete sensitivity analysis to be conducted with respect to these six Allen elasticities for which we have econometric evidence. 9