Methods to Aggregate Import Tariffs and their Impacts on Modeling Results Janine Pelikan and Martina Brockmeier

Transcription

1 Methods to Aggregate Import Tarffs and ther Impacts on Modelng Results Janne Pelkan and Martna Brockmeer Janne Pelkan, Insttute of Market Analyss and Agrcultural Trade Polcy, Federal Agrcultural Research Centre (FAL), Bundesallee 50, Braunschweg, Germany, phone: , fax: , Emal: Martna Brockmeer, Insttute of Market Analyss and Agrcultural Trade Polcy, Federal Agrcultural Research Centre (FAL), Bundesallee 50, Braunschweg, Germany Abstract When mpacts of WTO market access proposals are analyzed wth economc trade models, t s necessary to aggregate tarff data from the detaled tarff lne level to the model level. In ths artcle mport tarffs and mplemented mport tarff cuts are aggregated from the 6- dgt tarff lne level wth trade weghted average, the Trade Restrctveness Index (TRI) and the Mercantlst Trade Restrctveness Index (MTRI) by consderng bound and appled tarff rates. The resultng tarffs are substtuted for the orgnally appled mport tarffs of the GTAP data base. Multlateral trade lberalzaton scenaros are then mplemented and the welfare effects are compared. Keywords: Import tarff aggregaton, tarff cuttng scenaros, bound and appled tarffs, WTO negotatons, CGE modelng, agrcultural trade polcy JEL classfcaton: F13, F17, Q17 1

2 1 Introducton It s a great challenge for economsts to predct the mpact of multlateral trade lberalzaton. A study by UNCTAD (2003), for example shows that the results of welfare effects of lberalzaton studes can vary wdely. Ths may be attrbuted to the type of model (partal or general, statc or dynamc), the calculated scenaros or the chosen parameters. Furthermore, the data bases used are heterogeneous and lead to dvergences n each of the model results (Hess and Cramon-Taubadel, 2006). In the measurement of trade protecton, possbltes for dstorton exst on two levels. Frst the protecton can appear n many varatons: there are tarffs as well as non tarff trade barrers. Import tarffs can be expressed n value terms (ad valorem), as specfc and compound tarffs, or n the form of tarff rate quotas (TRQs). In order to compare the varous nstruments, they must be transferred nto a common metrc. For ths, however, no perfect soluton exsts (Cpollna and Salvatc, 2006). In most data bases, the polcy nstruments are thus converted nto ad valorem equvalents (AVEs). But dstortons emerge due to ths transformaton as well. Salamon (2006), for example, shows for mlk that the converson of specfc tarffs nto AVEs strongly depends on the bass year used. Secondly, the aggregaton of the data base also leads to dstortons. Trade polcy s carred out on the very detaled level of tarff lnes, for whch several mllon peces of tarff nformaton exst. The consumpton and producton data needed n the models are, however, avalable at a much hgher aggregaton level. Due to the rare avalablty of these data at the tarff lne level, there are only a few practcal possbltes for carryng out a theoretcally based aggregaton. Up untl now, manly smple or trade weghted averages of the tarff data were used n modelng (Manole and Martn, 2005). Anderson and Neary (1994, 2003) developed two theory-based ndces: the Trade Restrctveness Index (TRI) and the Mercantlst Trade Restrctveness Index (MRTI). Whle the frst ndex measures the unform welfare equvalent trade restrcton, the second ndex s an 2

3 mport volume equvalent method. There s a body of growng lterature on these ndexes. In some studes these ndces are calculated n a general equlbrum framework. Others use a partal equlbrum approach. In ths paper we concentrate on the TRI and MTRI and ther applcaton n modelng. For ths purpose, Chapter Two of ths paper descrbes the theory of the mplemented tarff aggregaton methods and gves a short overvew of the current lterature on ths topc. In Chapter Three a smple, partal equlbrum calculaton of the TRI and the MTRI s appled to aggregate tarffs from the 6-dgt tarff lne level to the level of the Global Trade Analyss Project (GTAP) model. Takng mport demand elastctes from the lterature and usng the MAcMap and COMTRADE data base, we calculate aggregated bound and appled mport tarffs. The results are then compared wth the tarffs derved by trade weghtng. How do the dfferent aggregaton methods nfluence smulaton results of the GTAP model? Is t possble to fnd a ratonale for the dfferng of the results or are they dfferent by chance? To answer these questons mport tarffs aggregated wth the help of the TRI and MTRI are addtonally substtuted for the orgnal appled mport tarffs of the standard GTAP data base. A multlateral trade lberalzaton scenaro accordng to the G-20 proposal (October 2005) s then mplemented at the tarff lne level. For each scenaro bound and appled tarffs are taken nto account. Fnally the modelng results are compared by consderng the welfare effects. Thereby, the focus of the dscusson s on the EU-27. The results are dscussed n Chapter Four. The paper ends wth a concluson. 2 Economc Theory of Tarff Aggregaton Methods In the lterature, a great varaton of tarff aggregaton methods are descrbed and used for emprcal analyss. In ths study the trade weghted average tarff, the TRI and the MTRI are appled. Therefore, only these three methods are dscussed n the theoretcal part of ths paper. For a broader dscusson of tarff aggregators see Cpollna and Salvatc (2006). 3

4 2.1 Trade Weghted Average Tarffs The tarff aggregaton based on the weghtng of the mport values s the most commonly used aggregaton method n modelng (Manole and Martn, 2005). Employng ths method the aggregated tarff t agg s calculated as a weghted average of the ad valorem tarffs t for good (=1,,.n): 1 * (1) t Agg = w t n = 1 The weght * w s thereby based on the mport quantty q and the world market prce p w : (2) w * = n q p = 1 w q p w Ths method consders the relatve mportance of trade flows. The greater the mportance of a product for trade, the greater s the weght gven to the product n the aggregaton. One advantage compared to other aggregaton methods s that the mport values of tarffs are accessble and nternatonally documented up to the 6-dgt level. Thus, the necessary data can, for example, be easly calculated wth the WITS 2 Software (World Integrated Trade Soluton) from the COMTRADE data base 3. The problem wth ths method s that the measured protecton tself endogenously nfluences the aggregaton. If an mport tarff ncreases, and, as a consequence, the mport demand decreases, the weght of the tarff also loses mportance. In contrast, the welfare loss of a tarff ncreases dsproportonately hghly wth the ncrease of the prevalng mport tarff. The tarffs have a greater effect on welfare and traded quanttes n the case of a relatvely elastc mport demand than wth a relatvely For the sake of smplcty we only consder one product group n ths equaton and abstract from regonal aggregaton. cf.: Unted Natons Commodty Trade Statstcs Data Base: 4

5 nelastc mport demand functon. However, the mport weghted tarff shows hgh values especally for tarffs of products wth nelastc demand. Addtonally, prohbtve tarffs are assgned weghts of zero, although the welfare losses are at maxmum. Ths endogenous bas consequently leads to an underestmate of the tarff restrctons. There are dfferent approaches n the lterature wth whch an attempt s made to mnmze the endogenous bas of the mport weghtng. For example, Bouët et al. (2004) propose a weghtng based on reference groups. Here t s assumed that the mport structure of smlar countres s a proxy for the free trade structure of a gven country. The weghted mport value V* s calculated usng the mport values of the reference groups (RefGrp): (3) V * r s = 1 r = 1 = V r Re fgrp n m n = 1 r = 1 m V V r r s Re fgrp V r ReGrp(s) refers to the value of product mported by the reference group of country s from country r. The second term of the equaton presents a normalzaton factor, wth whch the share of the total mports of a country s are gven n proporton of the total mports of the reference group. Thus, that the sze of the reference group does not nfluence the results. The aggregated tarffs are then calculated accordng to the Formula (1) where * r s w s defned as (4) w * r s = n V = 1 * r s V * r s In the lterature, the weghng of tarffs wth non-dstorted values s often presented as a preferable aggregaton method. The OECD (1997), for example, descrbes a weghtng va the value of the mports or the producton wthout trade barrers as deal. Leamer (1974) proposes a model n whch duty free mport values are estmated wth a functon of the 5

6 varables GNP, populaton and endowment of resources. These calculated mport values are used for the weghng of tarffs to make a consstent weghng possble. Anderson and Neary (2005) show however, that weghtng wth the undstorted mport values can reduce the endogenous bas. In contrast, a correct presentaton of welfare and traded quanttes s not gven, snce mport demand elastctes are not consdered. Another dsadvantage of ths method s that the mport values under free trade are hard to measure. To correctly estmate the weghts, much more nformaton would be needed and a complete mport demand model would have to be specfed and estmated. 2.2 Trade Restrctveness Index The Trade Restrctveness Index (TRI) developed by Anderson and Neary (1994) enables welfare based aggregaton of the tarffs. It represents the trade restrctons that are welfare equvalent to the ntal dsaggregated protecton structure on the aggregated level. Accordng to Kee at al. (2005b), t therefore answers the queston: What s the unform tarff that would keep welfare constant? Anderson and Neary (1996, 2005) derve the TRI as a general equlbrum applcaton from the dstance functon developed by Deaton (1979). They defne the TRI as an nverse, unform tarff factor whch compensates the representatve consumers for a current welfare change, whle holdng the balance of trade constant. Most current studes on the TRI use ths general equlbrum applcaton (.e., Anderson and Neary, 2005; Bach and Martn 2001; Salvatc, 2001). The advantage of ths method s the theoretcal consstency. However, t does not enable one to capture a detaled tarff structure, snce the necessary data are not avalable at ths detaled level (Cpollna and Salvatc, 2006). For the general equlbrum applcaton the tarffs are aggregated frst wth the help of other methods (.e., wth trade weghts) and then the welfare equvalent protecton level s calculated. 6

7 It s, however, possble to mplement the TRI partally (Anderson and Neary, 2005). Bureau and Salvatc (2004a and 2004b), for example, calculate the product specfc TRI for the agrcultural sector of the EU-15 and the USA. They aggregate the bound tarffs from the HS-8 level to the GTAP sector level. The necessary elastctes are taken from the GTAP data base and are thus only avalable at the aggregated level. The mport demand s calculated n ths study wth a CES (Constant Elastctes of Substtuton) Functon. Kee et al. (2005b) calculate the TRI for 91 countres under consderaton of tarff and non tarff trade barrers. For ths purpose they aggregate all appled tarffs from the HS-6 level to the country level, so that one tarff per country s dsplayed. The mport demand s modeled wth a lnear functon. Bureau et al. (2000) calculate the percentage changes of the TRI based on bound tarffs at the HS-8 level for the EU and the USA. Here, the TRI based on data from the year 1995 s compared wth the TRI n the year The necessary mport demand elastctes are estmated at the 4-dgt tarff lne level. In contrast to the trade weghted tarff, the TRI s theoretcally based. Dfferent studes show, however, that varous smplfyng assumptons must be accepted to emprcally calculate the TRI. Partcularly a lack of data avalablty at the HS-level complcates the use of ths concept n appled modelng. In the partal applcaton of the TRI no cross prce effects or ncome effects are consdered and changes n trade polcy do not capture ntersectoral effects. Also, the assumpton of a small country s very restrctve and does not permt the consderaton of terms of trade effects. Thus, the tarffs cannot affect the world market prces, and the welfare equvalence s expressed only through a change n the allocaton effcency. 7

8 2.3 Mercantlst Trade Restrctveness Index The Mercantlst Trade Restrctveness Index (MTRI) s also defned by Anderson and Neary (2003) and based on CORDEN (1966). Wth the help of ths ndex, the mport volume equvalent protecton s measured. The ndex shows how strongly natonal protecton dstorts the mports from the rest of the world. It s defned on the bass of an aggregated tarff, resultng n the same mport volume at world market prces as the ntal vector of nonaggregated tarffs (Anderson and Neary, 2003). Thus, the ndex measures the unform tarff whch keeps mports constant (Corden, 1966). Analogously, to the TRI, most emprcal calculatons of the MTRI are done as a general equlbrum calculaton (Cpollna and Salvatc, 2006). Anderson and Neary (2005) or Antman and Salvatc (2005), for example, use ths form. Whle work n the frst study s done on the HS-4 level, t s carred out n the second study at the GTAP model level. The detaled tarff data are brought up to the approprate level wth trade weghts n both studes. The MTRI can also be calculated wth the help of a partal equlbrum applcaton. Thereby, the MTRI s often expressed as an mport value equvalent tarff orgnally proposed by Corden (1966). Kee et al. (2005b) derve the MTRI for example on the bass of an mport value equvalent tarff and call t Overall Trade Restrctveness Index (OTRI). In the framework of the above mentoned study, they calculate appled tarffs and non tarff barrers at the HS-6 level and aggregate them to the source and product generc country level. Therefore, they show only one supra-sector aggregated tarff per country. Bureau and Salvatc (2004b) calculate the product specfc MTRI for the agrcultural sector of the EU-15 and the USA. The mport and tarff data n ths study are also dentcal wth the data used to calculate the TRIs. Bureau et al. (2000) calculate the changes n the MTRI for the EU and the USA through the mplementaton of the Uruguay Round Agreement. Also for ths study the same base data s used for both TRI and MTRI. 8

9 In the partal equlbrum applcaton t also holds true for the MTRI that no ntersectoral effects are documented and the assumpton of a small country works restrctvely. 3 Methodology Ths chapter descrbes how the emprcal results of ths paper are generated. The underlyng formulas for the TRI and the MTRI are devated and the data set s specfed. Fnally, the extenson of the GTAP model and the mplemented scenaros are descrbed. 3.1 Emprcal Estmaton of the TRI and MTRI The emprcal estmaton of the TRI and MTRI s carred out accordng to a partal equlbrum approach that s appled to the entre tarff data base of the GTAP model. Followng Feenstra (1995) and Anderson and Neary (2005), the TRI s calculated from a lnear mport demand curve gven below: (5) q = a b p Where the constant a stands for the pont at whch the demand curve of product ntersects the ordnate and b descrbes the slope of the demand curve. At the prce p the quantty demanded of product s q. The welfare loss (W ) arsng from an mport tarff, can be formally presented as: p d p w (6) W = qdp qdp qd ( pd pw ) = ( a pd 0.5b pd ) ( a pw 0.5b pw ) ( a b pd( pd pw )) p w world market prce of product p d domestc prce of product, p d = p w + p w t 9

10 Rearrangng the expresson n equaton (6) results n the followng term: (7) W 2 ( t pw ) = 2 b The total welfare loss of a gven country can be expressed as W = W, so that the welfare equvalent tarff t TRI s expressed mplctly wth the followng equaton: n n 2 TRI [( t pw ) b ] = 1 = 1 2 (8) [( t pw ) b ] = The left hand sde thereby shows the hypothetcal welfare loss, whle the rght hand sde expresses the welfare loss whch actually occurs. A rearrangement of equaton (8) results n equaton (9): (9) t TRI n = n = 1 = 1 ε w * [ ε w * t ] The MRTI ( t MTRI ) s calculated based on a concept of Corden (1966) and Anderson and Neary (2005). It s also derved from a lnear mport demand functon. In ths applcaton the mport value s used as the relevant metrc to calculate the MTRI. Such that the ndex s mplctly defned by Equaton (10): n (10) [ pw [ a b (1 + tmtri ) pw ] ] = [ pw [ a b (1 + t ) pw ] ] = 1 n = 1 The left hand sde of the equaton descrbes the total value of mports that s created when the unform tarff s employed. The rght sde of the equaton defnes, n contrast, the total value of all mports under the gven tarffs. Through a rearrangement of Equaton (10) the followng equaton can be derved: 10

11 (11) t MTRI = n = 1 n = 1 ε w [ ε w * * t ] Equatons (9) and (11) serve to calculate the TRI and the MTRI n the emprcal part of ths artcle. The sum over dfferent regons has to be calculated n addton to the sum over dfferent products n many of the emprcal analyss cases. 3.2 Data Set For the emprcal part of ths analyss the data s taken from the lterature. Ths data has been adjusted to address the specfc queston under consderaton, e.g., the calculated trade weghted average tarff only ncludes tarff lnes for whch elastcty data exsts. Otherwse t s not possble to correctly compare the dfferent methods. The tarff data used for the calculatons stem from the MAcMap 4 (Market Access Map) data base, developed through a combnaton of the nformaton from the data bases COM- TRADE, TRAINS, AMAD and the WTO data base. MAcMap provdes nformaton on preferental tarffs, tarff rate quotas (TRQs) and a consstent converson of specfc tarffs nto AVEs. The nformaton on preferental tarffs s taken from the TRAINS data base and extended wth natonal sources. AVE calculatons are based on the medan unt value of world-wde exports orgnatng from a reference group to whch an exporter belongs. These values are computed usng a three year average trade flow based on the perod (Bouёt et al., 2004). The calculaton of the AVEs for the bound tarffs s conducted wth the help of average world mport unt values for the same year (Bchr et al., 2006). If a tarff lne ncludes TRQs, these are converted wth the help of the fll rate of the AMAD data base. If the fll rate s less than 90 %, the tarff s used wthn the quota. If the fll rate

12 s greater than 100 % n contrast, the out of quota tarff s mplemented. Should the fllng rate be between 90 % and 100 %, a smple average of the n and out quota tarff s created (Bouёt et al., 2004). The trade data used as weghts come from the COMTRADE data base and reflect an average of the years 2000, 2001, and The nternal trade n the EU-15 and the new member states s excluded from the calculaton. Furthermore the tarff data of the EU-15 s transferred to the 12 new EU member states. Thus, the EU enlargement s smulated n concordance wth the model smulatons. The mport demand elastctes used are estmated by Kee et al. (2005a) at the 6-dgt tarff lne level wth a sem-flexble translog functon. The method s based on work by Kohl (1991) and Harrgan (1997), whereby mports are ncluded as nputs n a GDP functon. World market prces, factor endowment and a Hcks neutral productvty are ncluded as exogenous varables n the functon. The calculaton wth a GDP based functon requres that the mported goods, be processed domestcally. Aganst the background of ncreasng vertcal ntegraton and the assumpton that even n the case of mported processed products a value addng occurs through transport and marketng n the mportng country, ths GDP based functon s beng used more frequently to estmate such elastctes (Kee et al., 2005a). 3.3 Computaton of Dfferent Scenaros In the emprcal part of ths paper a multlateral trade lberalzaton scenaro s carred out to show the effects of the dfferent tarff aggregaton methods on modelng results. Therefore, an extended verson of the comparatve statc standard mult regonal general equlbrum GTAP model 5 s appled. 5 The framework of the standard GTAP model s well documented n the GTAP book (Hertel, 1997) and avalable on the Internet ( 12

13 Before the actual smulatons are carred out, some pre-smulatons are mplemented to extend the model structure and to update the protecton rates. The focus of the extenson s on the EU-27. Therefore nstruments of the Common Agrcultural Polcy (CAP) and the common budget of the EU are ncluded nto the GTAP model. Based on the results of the pre-smulaton, a base run s conducted whch projects the exogenous varables populaton, GDP and factor endowment up to the year Addtonally, the Agenda 2000, the EU enlargement and the EBA agreement as well as the Md Term Revew (MTR) are mplemented n 2004, 2010 and 2014, respectvely. The base run only consders poltcal nterventon n the EU-15 and n the 12 canddate countres. Developments n other regons, lke the Farm Bll of the USA or Chna's WTO access, are not yet taken nto account. Parallel to the base run, a scenaro s mplemented as well. It takes account of the same projectons and polcy shocks (Agenda 2000, EU enlargement, EBA agreement and MTR), but n the tme perod from 2010 to 2014, t addtonally mplements the WTO scenaros. Ths fnal step s n the centre of the followng analyss. Concernng market access the agrcultural tarffs are cut accordng to the proposed ters and the cappng of the G-20 proposal of October 2005 (compare Table 1). In addton to the tarff cuts to open agrcultural markets, tarffs of non-agrcultural commodtes are reduced by 50 % and 33 % n the developed and developng countres 6, respectvely. Least Developed Countres (LDCs) are exempted from tarff reductons n all scenaros. 6 Country classfcaton nto developng, developed or ndustralzed countres s appled accordng to the WTO classfcaton. Thus economes n transton are not explctly named. 13

14 Table 1: Agrcultural Tarff Cuts of the G-20 Proposal Used n the Smulatons Developed Countres Developng Countres tarff rate (%) tarff cut (%) tarff rate (%) tarff cut (%) >75 75,00 >130 40,00 > ,00 > ,00 > ,00 > , , ,00 Cap: 100 % Cap: 150 % Source: G-20 (2005). Agrcultural export subsdes are also elmnated n all scenaros. For the domestc support pllar, we follow the assessment of Brnk (2006) and Blandfort (2005) that nether of the currently avalable proposals wll hghly constran domestc support. Therefore, domestc support s kept unchanged for all countres and regons n all scenaros. In total, three scenaros are conducted by employng dfferent aggregaton methods based on the trade weghted average, the TRI and the MTRI. It s necessary to aggregate the tarffs before and after the tarff cut accordant to the measure under consderaton. Therefore, a program s developed applyng the Statstcal Analyss Software (SAS). Fgure 1 shows how the tarffs are calculated for each scenaro. Both the ntal base data and the reduced tarffs are generated from the 6-dgt tarff lne level to the model level. In a frst step, the orgnal base data n the GTAP-model are replaced by the new base data. After ths the GTAP data base contans the new appled tarffs at the base level. In a second step, the data are cut at the 6-dgt tarff lne level n the data base. Thereby, bound and appled tarffs are taken nto account. The reduced tarffs are aggregated wth the help of the same aggregaton measure that s used for the generaton of the base data. Fnally, a shock s mplemented n the GTAP model accordng to the dfference of the tarff rate before and after the cut. 14

15 Fgure 1: Calculaton of the Base Data and the Tarff Cuts by Dfferent Aggregaton Measures SAS HS6-dgt tarff lnes (MAcMaps) SAS HS6-dgt tarff lnes (MAcMaps) Aggregaton usng trade weghts, TRI or MTRI Implementaton of the tarff cuts New base data at model level Aggregaton usng trade weghts, TRI or MTRI GEMPACK Implementaton of the shock New tarff at model level 4 Results In Chapter 2 t s shown that only a few partal equlbrum applcatons of the TRI or MTRI exst. In the studes by Bureau and Salvatc (2004a and 2004b) and Bureau et al. (2000) only the bound tarffs are analyzed, whle Kee et al. (2005b) only concentrate on the appled tarff. Untl now there has been no study n whch bound and appled tarffs for the calculaton of the TRIs or the MTRIs are consdered. Anderson and Martn (2005) descrbe that the dfference between bound and appled tarffs s of great mportance partcularly for developng countres. In contrast, n the hgh ncome economes, there s only a slght dfference between bound and appled tarffs. However, all studes explan ths statement on the bass of average tarffs (e.g., Walkenhorst and Dhel, 2003) or trade weghted tarffs (e.g., Bchr et al and Lard, 2002). In many cases, only the dfference between bound and appled MFN tarffs s studed, as for example n Lard (2002). But the dfference between bound and appled tarffs exsts because of two 15

16 components. One s the bndng overhang that descrbes the dfference between bound and MFN tarffs accordng to Francos and Martn (2003). The other s due to the dfference between MFN and preferental tarffs (Anderson and Martn, 2005). What s the dfference between bound and appled tarffs for dfferent regons of the world, f calculatons are based on the TRI or the MTRI? Ths queston s addressed n the followng emprcal part of ths paper. A clarfcaton of ths queston s partcularly nterestng for WTO lberalzaton scenaros, snce only the reducton of appled tarffs can nduce trade effects Tarffs Computed wth SAS Table 2a and 2b represent the aggregated tarffs for selected agrcultural commodtes whch are appled n the GTAP smulatons. Regons and products are chosen accordng to the GTAP-aggregaton that s used n ths study (see Appendx A1.1 and A1.2). For llustraton purposes the three dmensonal matrces of the GTAP data base are aggregated nto two dmensonal arrays. Thus, the tables only show source generc tarffs. The trade weghted average, the MTRI and the TRI are presented for bound and appled tarff rates. The tarff cuts are mplemented accordng to the G-20 proposal as descrbed n Chapter 3.3. Hence, the reduced source generc tarff s named G-20. The last rows of Table 2b summarze the results for the agrcultural sector as a whole. Here, the source and product generc tarffs are presented. Tables 2a and 2b reveal that all aggregaton methods lead to a hgh gap between bound and appled tarff rates. Therefore, some economes wth WTO developng country status do not have to reduce ther appled tarffs at all. For example, the ACP countres are not forced to reduce ther tarffs for cereals. Also, Brazl s allowed to leave tarffs for beef, other meat and sugar unchanged, whle Inda does not have to change ther mport tarff polces n the sugar and beef sector. One reason for the hgh dfference between bound and 7 Ths s only when, the reduced appled tarffs are not prohbtve. 16

17 appled tarffs n ths study s the consderaton of blateral preferental tarff rates n addton to the MFN rates. 17

18 Table 2a: Aggregated Source Generc Tarffs by GTAP Sector and Regon n % Trade weghted MTRI TRI Mlk products BT a AT b G-20 c BT AT G-20 BT AT G-20 EU USA Japan Oceana rwtoic d Brazl Inda ACP LDC rwtodc e ROW f n.a n.a n.a Cereals BT AT G-20 BT AT G-20 BT AT G-20 EU USA Japan Oceana rwtoic Brazl Inda ACP LDC rwtodc ROW n.a n.a n.a Beef BT AT G-20 BT AT G-20 BT AT G-20 EU USA Japan Oceana rwtoic Brazl Inda ACP LDC rwtodc ROW n.a n.a n.a a) BT = ntal Bound Tarff b) AT = ntal Appled Tarff c) G-20 = tarff after the mplementaton of the G-20 proposal d) rwtoic = Rest of all Industralzed WTO member Countres e) rwtodc = Rest of all Developng WTO member Countres f) ROW = Rest of the World (non WTO members) Source: Authors calculatons. 18

19 Table 2b: Aggregated Source Generc Tarffs by GTAP Sector and Regon n % Trade weghted MTRI TRI Other Meat Products BT a AT b G-20 c BT AT G-20 BT AT G-20 EU USA Japan Oceana rwtoic d Brazl Inda ACP LDC rwtodc e ROW f n.a n.a n.a Sugar BT AT G-20 BT AT G-20 BT AT G-20 EU USA Japan Oceana rwtoic Brazl Inda ACP LDC rwtodc ROW n.a n.a n.a All Agrcultural Products BT AT G-20 BT AT G-20 BT AT G-20 EU USA Japan Oceana rwtoic Brazl Inda ACP LDC rwtodc ROW n.a n.a n.a a) BT = ntal Bound Tarff b) AT = ntal Appled Tarff c) G-20 = tarff after the mplementaton of the G-20 proposal d) rwtoic = Rest of all Industralzed WTO member Countres e) rwtodc = Rest of all Developng WTO member Countres f) ROW = Rest of the World (non WTO members) Source: Authors' calculatons. 19

20 The results of the trade weghted average and the two equvalence ndces are close to each other. Expectedly, the correlaton coeffcent accordng to Pearson and Bravas shows a statstcally sgnfcant correlaton between all methods (compare Table 3). Ths result s n accordance wth Bureau and Salvatc (2004b) who also found a close relatonshp between these tarffs when they calculated the trade weghted average tarff and the TRI for the EU and the USA. Table 3: Correlaton coeffcent accordng to Pearson and Bravas for the dfferent aggregaton methods a Correlaton of: Bound tarffs Appled tarffs G-20 Import weghted and MTRI * * * Import weghted and TRI * * * MTRI and TRI * * * a) The table shows the correlaton coeffcents for aggregates of all tarffs whch are used n the GTAP data base (ncludng non agrcultural products). It s not only based on the sample presented n Table 2a and 2b. The * denotes that coeffcents are statstcally sgnfcant at the 0.1 % level. Source: Authors' calculatons. Furthermore, Anderson and Neary (2005) show n a theoretcal and emprcal analyss that the dfference between the TRI and the mport weghted tarff s nfluenced by the varance of the ntal tarff structure. Accordng to ther regresson analyss, the percentage excess of TRI over the trade weghted tarff s postvely and sgnfcantly related to the coeffcent of varaton of the tarffs. Hence, f there s no varance n the ntal structure and all tarff lnes of a sector have dentcal tarffs, respectvely, the same tarff would be calculated by all aggregaton methods. In Inda for example all bound tarff lnes n the beef sector are fxed at 100 % and all appled tarffs at 35 %. Consequently, all aggregaton methods show the same aggregated bound and appled tarff rates for ths sector. 20

21 The mport weghted tarff s n general lower than the equvalence based methods, because the trade weghed tarff assgns hgh weghts to tarffs for products wth relatvely nelastc demand functons. The nfluence on the welfare and the trade values s, however, hgher, f the mport demand functon s relatvely elastc. Ths s consdered n the TRI or MTRI and leads to a further devatons between the ndces. Accordng to the underlyng theory, the MTRI s always lower than the TRI. Ths can be explaned ntutvely. If an ntal tarff vector of non-aggregated tarffs s replaced by an aggregated tarff ndex, then the lower tarffs ncrease and hgher tarffs are reduced. Snce the mport volumes or values must reman constant n the MTRI calculaton, the welfare losses of a tarff ncrease are less than the welfare gans of a tarff reducton. Consequently, the welfare ncreases when the MTRI s mplemented. A hgher tarff s needed to mantan the welfare at ts orgnal level. For ths reason the TRI s always hgher than the MTRI (Anderson and Neary, 2005) Welfare Effects Computed wth GTAP How s the outcome of the smulatons nfluenced by dfferent tarff aggregaton measures? Table 4 reveals the welfare effects of trade lberalzaton whch follow the mplementaton of the G-20 proposal. The welfare effects are dscussed on the bass of the equvalent varaton n ncome expressed n mllon US $. The man focus of the analyss s on the EU-27. The columns of Table 4 represent the welfare change as subtotals of market access lberalzaton and of the abolshment of export subsdes n dfferent regons of the world. Addtonally, t s dfferentated between the mpact of agrcultural tarff reducton and nonagrcultural tarff reducton. The decomposton of the results show how much of the welfare effect stems from lberalzaton of the EU market and how much s due to the lberal- 8 For formal proof of ths statement see also ANDERSON and NEARY, 2005, p. 66ff. 21

22 zaton of thrd country markets. The fnal row of each scenaro ndcates the sum over all welfare effects of each pllar. The mplementaton of the trade weghted average tarff generates an ncrease n welfare of $ 72 bllon for the world. Developng countres and LDCs receve 60 % of these global welfare gans, whle the ndustralzed countres only receve 37 %. The remanng 3 % are dstrbuted to non WTO member countres (ROW). They are partcpatng n the gans from non agrcultural trade lberalzaton because of the terms of trade effect. The reducton of agrcultural tarffs contrbutes wth more than two thrds or $ 49.6 bllon to the overall welfare gan. Developng countres retan about $ 28.6 bllon of these gans whch come manly from market access lberalzaton between thrd countres ($ 25.6 bllon). A more detaled examnaton shows that most of the gans for developng countres are generated by lberalzaton of ther own agrcultural markets. The overall welfare effect of the export subsdy abolshment s small compared to the gans nduced by the reducton of mport tarffs. But the decomposton shows that there s a wde margn of results whch compensate each other and lead to a small overall effect. In most economes, especally n developng countres, the elmnaton of export subsdes leads to a welfare loss. The abolshment reduces the supply of agrcultural products on the world market. Consequently, the world market prces rse and net food mportng countres suffer from hgher prces. The EU-27 s payng most of these subsdes and expectedly shows welfare gans of around $ 10 bllon due to the terms of trade and the allocaton effect followng the abolshment of ths trade dstortng nstrument. 22

23 Table 4: Welfare Effects of Trade Lberalzaton Accordng to the G-20 Proposal (2001 US$ mllons) Tarff of agrcultural products Tarff of non Export subsdes Scenaro from TC a) from EU From TC agrcultural from EU From TC World to EU to TC To TC products to TC to all regons Weghted Mo. $ EU Other IC DC LDC ROW total MTRI Mo. $ EU Other IC DC LDC ROW total TRI Mo. $ EU Other IC DC LDC ROW total a) TC = Thrd Countres Source: Authors' calculatons. If the MTRI s used as tarff aggregaton measure, the overall welfare gans ncrease to $ 76.6 bllon. The dstrbuton of welfare gans between ndustralzed economes (41 %) and the group of developng countres ncludng LDCs (57 %) s smlar to the prevous scenaro. 23

24 The reducton of agrcultural and non agrcultural tarff dstortons generates $ 52 bllon and $ 24.7 bllon, respectvely, for the world as a whole. Half of the welfare gan from tarff reducton s collected by developng countres. Compared to the frst scenaro, the benefts of trade lberalzaton partcularly ncrease n ndustralzed economes. Thus, the mplementaton of the G-20 proposal on mport equvalent tarffs nduces hgher overall welfare results n the EU and other ndustralzed countres. Developng countres show a hgher ncrease n welfare for non agrcultural products than n the frst scenaro. However, the effect of agrcultural tarff reducton dffers only slghtly between both scenaros for these economes. The TRI scenaro predcts an overall welfare gan of $ bllon whch s more than twce as hgh as n the two other scenaros. Industralzed countres are recevng 28 % of these gans, whle developng economes and the LDCs are obtanng 71 %. Developng countres receve $ bllon of the welfare gans from non agrcultural market lberalzaton. In contrast, the terms of trade for other ndustralzed countres worsens, so that the welfare effect of non agrcultural market lberalzaton becomes negatve. However, other ndustralzed countres obtan $ 45.2 bllon through the reducton of agrcultural mport tarffs. Accordngly, ther overall welfare gan adds up to $ 41.2 bllon. The welfare effect followng the abolshment of export subsdes s negatve n ths scenaro. The terms of trade effect s more evdent due to the hgher mport tarffs n the ntal stuaton, and welfare decreases n all countres except the EU-27. The ntal tarffs are much hgher compared to the other two scenaros, when the TRI s mplemented as tarff aggregaton measure. A reducton of these hgh ntal tarffs accordng to the G-20 proposal n the agrcultural sector, and the proportonal tarff cut n the non agrcultural sector results n hgher cuts and therefore also n hgher welfare effects. Most of the addtonal gans can, however, be attrbuted to the reducton of the non agrcultural tarffs. Therefore, t s not surprsng that the reducton of welfare equvalent tarffs leads to 24

25 a hgh welfare gan. In addton to the aforementoned reason the trade value of the manufacturng sector s much hgher than that of the agrcultural sector. In sum, the reducton based on the TRI expectedly results n hgher welfare gans for sngle country groups and the whole world than the reducton based on trade weghted tarffs. How does ths relate to the already avalable studes on trade lberalzaton? Most of the more recent studes on trade lberalzaton show lower welfare results than the earler ones. The reasons are, for example, the mprovement of data bases through the ncluson of preferental tarffs whch leads to a lower ntal protecton level. Addtonally, t has to be taken nto account that n the meantme some lberalzaton has already occurred. However, ths study shows that the welfare effects hghly depend on the chosen tarff aggregaton method. All recent studes only use trade weghts as aggregaton methods and mght thus stll underestmate the welfare changes of trade lberalzaton. The analyss presented here shows how much and n whch drecton the welfare effects of a trade model can be nfluenced by dfferent aggregaton methods. For future research t would be nterestng to nclude some senstvty analyss on the employed mport demand elastctes. Here, only trade data are consdered for whch elastctes exst. How to deal wth mssng data s another research queston whch would be nterestng to address n future research. 5 Concluson It s well known from the lterature that the tarff data base of a model can nfluence the results of lberalzaton studes sgnfcantly. An essental prerequste for the appled polcy analyss s thus a consstent and transparent tarff data base. The MAcMap data base (Bouёt et al., 2004) therefore presents a very mportant development n ths context. It connects the data bases AMAD, TRAINS, WTO and COMTRADE as well as nformaton from natonal sources. It also ncludes bound and appled tarffs wth consstent converson of the specfc tarffs n AVEs up to the 6-dgt tarff lne level. 25

26 The results of a lberalzaton study could, however, also be nfluenced through the aggregaton of the tarff data to a model compatble level. Dfferent methods are avalable for the aggregaton of tarff data. Usually the smple and the trade weghted average tarff are mplemented. The advantage of these two methods s the comparably low data requrements. However, these methods do not base on a economc theory. Anderson and Neary (1994, 2003), n contrast, developed two theoretcally sound aggregaton methods. These are the TRI (Trade Restrctveness Index), allowng for a welfare equvalent aggregaton of the tarffs and the MTRI (Mercantlst Trade Restrctveness Index), wth the help of whch the tarffs can be aggregated mport volume equvalent. These ndces place hgh requrements on the data used for ths purpose, whch can n many cases not be fulflled. If smplfyng assumptons are made, t s however possble to calculate these ndces wth the already exstng data for the whole mport tarff data base of a trade model. In ths paper we aggregate tarff data wth the help of mport weghts and based on the TRI and MTRI from the 6-dgt tarff lne level to the GTAP model. The results show a hgh dfference between bound and appled tarff rates for all aggregaton methods. The gap between both tarffs arses for developed and to a greater extend for developng countres. Therefore, f the G-20 proposal of the WTO negotatons s appled, some countres do not have to reduce ther appled tarffs of some sectors at all. Analyss of the aggregated bound and appled tarff rates reveals a sgnfcant correlaton between all aggregaton methods. The level of the tarff aggregates can stll be very dfferent though. Whle the trade weghted average tarff and the MTRI are very smlar, the TRI s much hgher. Furthermore, the emprcal analyss shows n accordance wth the underlnng theory that there s a hgh correlaton between the varance of the ntal tarff structure and the dfference between the trade weghted average and the MTRI wth the TRI. The welfare results of the smulatons wth the GTAP model mrror these fndngs. It can be shown that the trade weghted average tarff s a good approxmaton of the theoretcally 26

27 sound MTRI. The welfare effect of the TRI Scenaro, n contrast, exceeds these results by more than three tmes. Also the share of welfare effects due to agrcultural or manufacturng trade lberalzaton changes. Approxmately two thrds of the welfare gans arse from agrcultural trade lberalzaton, f the trade weghted of the MTRI s appled. In contrast, f the TRI s appled manufacturng trade lberalzaton becomes most mportant. However, ths analyss shows that welfare results are hghly dependent on the chosen aggregaton method. If the TRI s assumed to be the most exact aggregaton measure, welfare effects are most lkely underestmated wth trade weghted tarffs. Gven these results, t mght no longer be true that the measurable costs of protecton are not large n percentage terms of the GDP as Krugmann (1995) descrbes n what he called a drty lttle secret. A consderaton mght be especally worthwhle when NTB s are addtonally taken nto account. Acknowledgement We wsh to thank Raner Klepper for valuable assstance wth the Statstcal Analyss Software, as well as Heko Hansen and the conference partcpants of the 10 th Annual Conference on Global Economc Analyss, June, 2007, for helpful comments. Any remanng errors are ours. 27

28 6 References Anderson, J. E., Neary, J. P. (2005) Measurng the Restrctveness of Trade Polcy, The World Banks Economc Revew, 8, (2003) The Mercantlst Index of Trade Polcy, Internatonal Economc Revew, 44 (2), (1996) A New Approach to Evaluatng Trade Polcy, Revew of Economc Studes, 21 (6), (1994) Measurng the Restrctveness of Trade Polcy, The World Banks Economc Revew, 8 (2), Anderson, K., Martn, W. (2005) Scenaros for Global Trade Reform, n Poverty and the WTO Impacts of the Doha Development Agenda, Hertel, T. W. and Wnters, A. (Ed.), Chapter 2 Antman, A., Salvatc, L. (2005) EU Trade Polces, Benchmarkng Protecton n a General Equlbrum Framework, LLRR Workng Paper No, 31 Bach, C. F., Martn, W. (2001) Would the Rght Tarff Aggregator for Polcy Analyss Please Stand up? Journal of Polcy Modelng, 23, Bchr, M. H., Jean, S., Laborde, D. (2006) Bndng Overhang and Tarff-Cuttng Formulas, Revew of World Economcs, 142 (2), Blandfort, D. (2005) Dscplne on Domestc Support n the Doha Round, Internatonal Agrcultural Trade Research Consortum (IATRC) Trade Polcy Issue Paper 1 Bouёt, A., Decreux, Y., Fontanage, L., Jean, S., Laborde, D. (2004) A Consstent, ad Valorem Equvalent Measure of Appled Protecton Across the World, The MacMap- HS6 data base, Workng Paper No. 22 Brnk, L. (2006) WTO Constrans on the US and the EU Domestc Support n Agrculture, The October 2005 Proposals, The Estey Centre Journal of Internatonal Law and Trade Polcy, 7 (1), Bureau, J. C., Salvatc, L. (2004 a) WTO Negotatons on Market Access n Agrculture, a Comparson of Alternatve Tarff Cut Proposals for the EU and the US, Topcs n Economc Analyss & Polcy, 4 (1), Artcle 8 (2004 b) WTO Negotatons on Market Access, What we Know, What we Don t and What we Should, n Agrcultural Polcy Reform and the WTO: Where are we headng? Ananna, G., Bohman, M. E., Carter, C. A. and McCalla (Ed.) Bureau, J. C., Fulpon, L., Salvatc, L. (2000) Comparng EU and US Trade Lberalsaton under the Uruguay Round Agreement on Agrculture, European Revew of Agrcultural Economcs, 27 (3), Deaton, A. (1979) The Dstance Functon n Consumer Behavour wth Applcatons to Index Numbers and Optmal Taxaton, Revew of Economc Studes, 46, Cpollna, M., Salvatc, L. (2006) Measurng Protecton: Msson Impossble? Trade AG Workng Paper No. 06/07 Corden, W. M. (1966) The Effectve Protectve Rate, the Unform Tarff Equvalent and the Average Tarff, Economc Record, 42,

29 Feenstra, R. C. (1995) Estmatng the Effects of Trade Polcy, n Handbook of Internatonal Economcs (Ed.) Grossman, G. and Rogoff, K., 3 (30), Francos, J. F., Martn, W. (2003) Formulas for Success? Formula Approaches to Market Access Negotatons, World Economy, 26 (1), 1 28 Harrgan, J. (1997) Technology, Factor Supples, and Internatonal Specalzaton, Estmatng the Neoclasscal Model, The Amercan Economc Revew, 87 (4), Hertel, T. W. (1997) Global Trade Analyss Modelng and Applcatons, Cambrdge Unversty Press, Cambrdge Hess, S., von Cramon-Taubadel, S. (2006) Meta-Analyss of General and Partal Equlbrum Smulatons of Doha Round Outcomes, Invted Paper, IAAE, August 12 18, Gold Coast, Australa Kee, H. L., Ncta, A., Olarreaga, M. (2005a) Import Demand Elastctes and Trade Dstortons, Centre for Economc Polcy Research, Workng Paper No. 4669, London, UK (2005b) Estmatng Trade Restrctveness Indces, The World Bank Group Kohl, U. (1991) Technology, Dualty and Foregn Trade: the GNP Functon Approach to Modelng Imports and Exports, The Unversty of Mchgan Press, Ann Arbor Krugman, P. (1995) Dutch Tulps and Emergng Markets, Councl on Foregn Relatons, Inc. Foregn Affars, July/August Lard, S. (2002) Market Access Issues and the WTO, An Overwew, n Development Trade and the WTO A Handbook, The World Bank, Hoekmann, B., Mattoo, A. and Englsh, P., (Ed.) Washngton DC, Chapter 11 Leamer, E. E. (1974) Nomnal Tarff Averages Wth Estmated Weghts, Southern Economc Journal, 41, Manole, V., Martn, W. (2005) Keepng the Devl n the Detals, A Feasble Approach to Aggregatng Trade Dstortons, ETSG (European Trade Studes Group) conference proceedng, Dubln, September OECD (1997) Indcators of Tarff and Non-Tarff Trade Barrers, Report 1997 Salamon, P. (2006) Wohn bewegt sch der Mlchpres n Deutschland und n der EU? Landbauforschung Völkenrode FAL Agrcultural Research, Sonderheft, 299, Salvatc (2001) Trade Dstorton Indexes and Multregonal AGE Models, The Case of the Common Agrcultural Polcy, Unversta Degl Stud D Roma, La Sapenza Dpartmento D Economa Pubblca, Workng Paper No. 45. UNCTAD (2003) Back to Bascs, Market Access Issues n the Doha Agenda,. Doc. UNC- TAD/DITC/TAB/Msc. 9, Genf Walkenhorst, P., Dhel, N. (2003) Tarff Bndngs, Unused Protecton and Agrcultural Trade Lberalzaton, OECD Economc Studes, 36 29