Chapter 5. Determinants of Economic-Based Protection from Technical Barriers to U.S. Agricultural Exports

Transcription

1 Chapter 5. Determnants of Economc-Based Protecton from Techncal Barrers to U.S. Agrcultural Exports 5.1. Introducton Despte the efforts of the nternatonal tradng communty to lmt msuse of techncal barrers through the Uruguay Round Agreements, the 1996 USDA Survey results suggest that these regulatory measures contnue to provde dsgused economc-based protecton for domestc ndustres. Poltcal economy s a paradgm that explans government nterventon n the market, where polcy choce s endogenously determned by ndvdual agents and polcymakers actng as ratonal maxmzers gven ther sets of preferences. The nteracton between economcs and poltcs may result n polces whch are net welfare decreasng, such as the provson of economc-based protecton through techncal barrers. Emprcal models quantfy the economc and poltcal ncentves and abltes of ndvdual agents and polcymakers to affect regulatory outcomes. An emprcal approach smlar to those used n prevous studes of other trade and agrcultural polcy decsons can be appled to analyze the poltcal economy factors underlyng the ncdence and mpact of questonable techncal barrers. In ths chapter, two fundamental emprcal models are exposted to dentfy the determnants of economc-based protecton from questonable techncal barrers as they are appled to U.S. agrcultural exports. The purpose of the emprcal models s not to test the poltcal economy theory, but rather to dentfy the common factors underlyng observed regulatory levels and to quantfy the economc and poltcal relatonshps that gve rse to questonable techncal barrers. The frst model addresses the ncdence of questonable barrers. The dependent varable n ths model measures the presence or absence of questonable techncal barrers appled by country. The second emprcal model addresses the mpact of questonable techncal barrers. The dependent varable measures the estmated trade mpact from questonable techncal barrers as a percentage of 1996 U.S. agrcultural exports to that country. For each of these two models a number of econometrc specfcatons are estmated. Independent varables n the econometrc models represent the nfluence of market and poltcal economy factors, as well as the possble nfluence of survey desgn characterstcs on the estmaton results. The varables reflect nfluental measures dentfed n the theoretcal and emprcal poltcal economy models dscussed n Chapter Three ncludng characterstcs of trade, agrculture, and the aggregate economy. Such a characterzaton can then be reorganzed to 100

2 provde proxy measures for the nterests of ndvdual agents, effectve poltcal nfluence, polcymaker preferences, and nsttutonal structures, as depcted n Fgure 3.3. Secton 5.2 ntroduces the dependent and ndependent varables used n the econometrc estmatons to represent the poltcal economy of techncal barrers. A unvarate PROBIT model that quantfes the ncdence of questonable techncal barrers s presented n Secton 5.3 where the dependent varable s a bnary representaton of the presence or absence of questonable techncal barrers by country. Secton 5.4 presents a TOBIT model to measure the mpact of questonable techncal barrers. The dependent varable measures the estmated percentage revenue loss from questonable techncal barrers, dstrbuted contnuously and censored at zero. A summary and conclusons are presented n Secton Econometrc Models of Questonable Techncal Barrers to U.S. Agrcultural Exports The poltcal economy theory presented n Chapter Three provdes a conceptual framework for emprcal analyss of the economc-based protecton provded by questonable techncal barrers. Ths s the underlyng theoretcal paradgm represented n the followng emprcal models. An approach smlar to those used to model other agrcultural and trade polcy decsons s appled to analyze the reported ncdence and mpact of questonable techncal barrers. Snce ndvdual agent preferences, effectve poltcal nfluence, polcymaker preferences, and nsttutonal structures cannot be measured drectly, a number of varables are specfed as proxy measures for the underlyng poltcal economy relatonshps. Taken together, the set of ndependent varables selected for ncluson n the models gves a broad characterzaton of the lkely poltcal economy nfluences on the regulatory decsons. 5.2.a. Measures of Incdence and Impact The ncdence and mpact of questonable techncal barrers to U.S. agrcultural exports, as presented n Chapter Four, provdes measures of potental economc-based protecton on a cross-sectonal bass at one pont n tme. There are several ways to quantfy protecton usng ths data. The presence or absence of barrers dentfed as questonable reflects the ncdence of protecton, whle the percentage estmated trade mpact reflects the mpact of protecton. 5.2.a.1. Presence of Questonable Techncal Barrers The presence or absence of questonable techncal barrers ncludes all observatons n the 1996 USDA Survey summed by country (n=134). Ths s a bnary varable, YN, where one s used to ndcate the presence of one or more questonable techncal barrers to U.S. agrcultural exports and zero s used to ndcate the absence of such barrers. There are 63 one-observatons 101

3 and 71 zero-observatons of the YN varable ncluded n the sample (see Appendx B for a lst of countres ncluded n each category). 5.2.a.2. Percentage Trade Impact from Questonable Techncal Barrers The mpact of questonable techncal barrers appled to U.S. agrcultural exports s measured n the 1996 USDA Survey by the estmated loss, or potental loss, n 1996 export revenues attrbutable to these barrers assumng a fxed world prce. To correct for magntude dfferences n trade among countres and commodtes t s useful to normalze the estmated trade mpact values from the survey. Ths s a censored normal varable, PETI, where percentage mpact s measured as the estmated trade mpact (from the 1996 USDA Survey) dvded by the total value of 1996 U.S. agrcultural exports to the country mposng the barrer. Agan, observatons are summed by countres (n = 134). Values range from zero, for those 71 countres where no questonable techncal barrers were dentfed, to 412, where the estmated trade mpact from techncal barrers n 1996 was large relatve to total 1996 agrcultural exports to that country. In the prevous chapter, Table 4.28 shows a dstrbuton of countres n the sample by the percentage estmated trade mpact. 5.2.b. Measures of the Poltcal Economy Determnants of Questonable Techncal Barrers Followng the approach used n the emprcal studes revewed n Chapter Three, a number of ndependent varables are dentfed to serve as proxes for the poltcal economy determnants of questonable techncal barrers due, n part, to the fallblty of partcular sngle measures (Bollen 1980; Beghn and Kherallah 1994). Snce the measures of ncdence and mpact of economc-based protecton represent cross-country observatons, the ndependent varables n the econometrc models more closely follow those measures used n the crosscountry emprcal studes of other agrcultural and trade polces, nstead of cross-commodty analyses or studes that focus on votng decsons for a specfc polcy ssue. As dscussed n Chapter Three, Honma and Hayam (1986) and Anderson and Hayam (1986) categorze the ndependent varables n ther studes as measures of comparatve advantage, agrcultural share n the economy, and terms-of-trade. The productvty rato of labor n agrculture relatve to manufacturng, and the factor rato of agrcultural land to captal endowment per worker serve as measures of comparatve advantage. The share of agrculture n the labor force and n natonal GDP serve as measures of nfluence from the agrcultural sector n the economy. Terms-of-trade are measured as the export value of agrcultural relatve to manufactured products. Fnally, dummy varables are ncluded to dentfy members of the EU, non-allance countres, and Japan. Leamer (1990) groups the ndependent varables n hs emprcal model nto categores reflectng the economc sze of the country, characterstcs of the country, and characterstcs of the commodty. Natonal GNP measures economc sze of the country. Populaton per GNP and arable land per GNP serve as measures of country characterstcs. Captal per man-hour, land per 102

4 man-hour, and a seres of dummy varables represent characterstcs of the commodtes ncluded n the study. Grll (1988) dvdes the determnants of protecton n hs study nto measures of the state of the domestc and world economy, domestc compettveness, and structural changes related to shfts n comparatve advantage or growth. The real exchange rate, measures of unemployment and productvty are used as ndcators of the state of the economy. Import penetraton serves as a measure of comparatve advantage and a tme trend serves as a measure of structural change. DeGorter and Tsur (1991) ncludes a measure of the dfference n GDP between agrcultural and rural sectors as a proxy for ncome endowment dfferentals. Other ndependent varables n the model are per-capta arable land, percentage rural populaton, and dummy varables for net exporters versus mporters, ncome groups, and the commodty rce. Beghn and Kherallah (1994) group the ndependent varables n ther study nto measures of cvl lbertes, poltcal systems, publc fnance, and others. Both cvl lbertes and poltcal systems are captured by a seres of dummy varables ncludng a dummy varable representng OECD membershp. Ratos of export dversfcaton (export value of the crop to total exports), tax revenues (taxes on ncome to total tax revenue), and tax nstruments (export tax revenue from the crop to total ndrect taxes) serve as measures of publc fnance. The rato of export to mport values, agrcultural share n GDP, ncome, a socal equty ndex, and a measure of comparatve advantage are ncluded n the category of other varables. In ths dssertaton a total of 23 ndependent varables are consdered as proxy measures to represent the poltcal economy determnants of the ncdence and mpact of questonable techncal barrers n the emprcal specfcatons. 47 Followng the general approach used n other cross-country studes, the ndependent varables can be categorzed nto three broad groups: characterstcs of trade, characterstcs of the domestc agrcultural sector, and characterstcs of the aggregate economy. There are 11 varables ncluded as proxy measures for characterstcs of trade, fve varables ncluded as proxy measures for characterstcs of the agrcultural sector, sx varables ncluded as proxy measures for characterstcs of the aggregate economy, and one varable ncluded to measure the nfluence of the survey desgn. The names and defntons of the ndependent varables are lsted n Table 5.1 and a longer descrpton of each varable s ncluded n the followng sectons. All of the exogenous varables are contnuous except for the dummy varables used, respectvely, to represent WTO membershp and to reflect survey desgn as nfluenced by the presence of an FAS post n some countres but not others. As the poltcal economy paradgm represents an teratve process as depcted n Chapter Four, t s assumed that the determnants of 1996 questonable techncal barrers exsted n pror years. The ndependent varables represent trade, agrculture, and aggregate economc characterstcs n 1995, the year pror to collecton of the USDA Survey, except for those varables that represent ether projected 47 The potental for proxy varables to ntroduce bas nto the estmaton has been wdely recognzed snce such varables are mperfect representatons of the underlyng latent true varable. However, research has shown that the asymptotc bas n parameter estmates that are ncluded s worse f some mportant proxy s omtted (Maddala 1992; Greene 1997). 103

5 future characterstcs or an average across several years pror to Summary statstcs for the ndependent varables are shown n Table 5.2 and a matrx of correlaton coeffcents s shown n Table 5.3. Table 5.1. Independent varable names and defntons used n the econometrc models Varable Varable Defnton Unts Characterstcs of Trade TRDBAL 1995 agrcultural trade balance (exports mports) $ per capta GTDBAL growth n exports growth n mports percent USTRD 1995 agrcultural trade balance wth the U.S. (exports to $ per capta the U.S. mports from the U.S.) GUSTRD average growth n agrcultural trade balance percent wth the U.S. (change n exports to U.S. mports from U.S.) GDPEXP 1995 percentage of exports n GDP percent PENTR 1995 agrcultural mport penetraton relatve to domestc percent value-added n agrculture TAPPL Projected 1999 appled MFN average tarff rate for percent agrcultural mports SLACK Dfference between 1999 bound tarff rate and projected percent appled tarff rate on agrcultural mports TAPPLF Projected 1999 MFN tarff rate faced by agrcultural percent exports TREDF Change n projected 1999 tarff faced by agrcultural percent exports as a result of Uruguay Round commtments from the country s tradng partners WTO 1996 WTO membershp (1 = full, 0 = other) 0,1 Characterstcs of Agrculture VAAG 1995 value-added n agrculture $ per capta GDPAG 1995 GDP from the agrcultural sector percent LABAG 1995 labor force employed n agrculture percent KL 1995 captal/land rato (tractors/land rato) number per 1000 hectares LL 1995 labor/land rato (agrcultural workers/land rato) number per hectare Characterstcs of the Aggregate Economy GDP 1995 GDP $ per capta PVT 1995 prvate consumpton $ per capta GFPRCE average annual growth n nomnal food prces percent RURAL 1995 rural populaton percent XCHG 1995 rato of offcal to parallel exchange rate rato GOVT 1995 percentage of government consumpton n GDP percent Measurement Issues FAS 1996 FAS post n country (1 = yes, 0 = no) 0,1 104

6 Table 5.2. Summary statstcs for varables used n the econometrc models Standard Varable Mean Devaton Medan Maxmum Mnmum Number of Observatons Dependent Varables YN PETI Characterstcs of Trade TRDBAL GTDBAL USTRD GUSTRD GDPEXP PENTR TAPPL SLACK TAPPLF TREDF WTO Characterstcs of Agrculture VAAG GDPAG LABAG KL LL Characterstcs of the Aggregate Economy GDP PVT FPRCE RURAL XCHG GOVT Measurement Issues FAS

7 Table 5.3. Correlaton coeffcents for all observatons n the data set YN PETI TRDBAL GTDBAL USTRD GUSTRD GDPEXP Dependent Varable YN 1 PETI Characterstcs of Trade TRDBAL GTDBAL USTRD GUSTRD GDPEXP PENTR TAPPL SLACK TAPPLF TREDF WTO Characterstcs of Agrculture VAAG GDPAG LABAG KL LL Characterstcs of the Aggregate Economy GDP PVT GFPRCE RURAL XCHG GOVT Measurement Issues FAS

8 PENTR TAPPL SLACK TAPPLF TREDF WTO VAAG Characterstcs of Trade PENTR 1 TAPPL SLACK TAPPLF TREDF WTO Characterstcs of Agrculture VAAG GDPAG LABAG KL LL Characterstcs of the Aggregate Economy GDP PVT GFPRCE RURAL XCHG GOVT Measurement Issues FAS GDPAG LABAG KL LL GDP PVT GFPRCE Characterstcs of Agrculture GDPAG 1 LABAG KL LL Characterstcs of the Aggregate Economy GDP PVT GFPRCE RURAL XCHG GOVT Measurement Issues FAS RURAL XCHG GOVT FAS Characterstcs of the Aggregate Economy RURAL 1 XCHG GOVT Measurement Issues FAS

9 5.2.b.1. Characterstcs of Trade There are 11 ndependent varables ncluded n the emprcal models to represent the characterstcs of trade among countres. Agrcultural trade balance (TRDBAL) measures the strength of the agrcultural export sector relatve to that of the agrcultural mport sector. As exports decrease relatve to mports, domestc agrculture producers may have more to gan from regulatory nterventon. Data for the 1995 value of agrcultural exports and mports by country, n U.S. dollars, s reported n FAO (1997). Values for agrcultural trade balance range from -$ per capta (Unted Arab Emrates) to $ per capta (Ireland). The mean agrcultural trade balance s slghtly postve at $2.61 per capta (South Afrca, Indonesa). The medan value s -$8.82 (Span). A measure of change n trade balance (GTDBAL) may capture a shft n the relatve strength of the domestc agrcultural sector. The percentage growth n exports mnus growth n mports s calculated from data reported n World Bank (1997). The number of observatons for ths measure s lmted, as data s avalable for only 94 countres. Values range from percent (Argentna) to percent (Lesotho). The mean value s 1.97 percent (Guatemala) and the medan value s 0.45 percent (Austra, Israel). As the 1996 USDA Survey only measures questonable techncal barrers to U.S. agrcultural exports, the per-capta agrcultural trade balance wth the U.S. (USTRD) and the average percent change n ths trade balance (GUSTRD) are proxes for the blateral tradng relatonshp. Agan, as blateral exports decrease relatve to mports domestc producers may have more to gan from regulatory nterventon. The correlaton between the overall agrcultural trade balance and the agrcultural trade balance wth the U.S. s 0.514, ndcatng that whle net exporters are also exporters to the U.S., there s not a strong relatonshp between the multlateral and blateral levels of trade. In addton, snce the dependent varable s specfc to the U.S. the measures of trade balance wth the U.S. may also provde some ndcaton of a retalatory tradng strategy. As agrcultural exports to the U.S. decrease relatve to agrcultural mports from the U.S., polcymakers may be more lkely to mpose questonable techncal barrers on U.S. exports. Data for the 1995 per-capta agrcultural trade balance between the U.S. and other countres (n U.S. dollars) and the average percentage change n such a balance between 1992 and 1995 are calculated from data reported n FAS (1997). The values for per-capta agrcultural trade balance wth the U.S. range from -$ (Hong Kong) to $ (New Zealand). The mean value s -$3.33 (Pakstan) and the medan s -$1.29 (Bangladesh). Values for the change n such a trade balance range from percent (Serba) to percent (Uzbekstan). The mean value s 2.84 percent (Bosna) and the medan value s 0.08 percent (Tawan, Trndad and Tobago, Belgum, Swtzerland). The percentage of exports n natonal GDP (GDPEXP) ndcates the dependence of the economy on nternatonal markets and s one proxy for outward-lookng market orentaton. Ths s a broader measure of export dependence than that provded by the agrcultural trade balance. A polcymaker may be less lkely to restrct trade through questonable techncal barrers when a country s more dependent on nternatonal markets as an outlet for domestc products. A 108

10 postve correlaton coeffcent of between the percentage of exports n GDP and the level of GDP ndcates that dependence on nternatonal markets tends to grow as countres become relatvely more wealthy. The 1995 percent of exports n GDP by country s found n World Bank (1997). The values range from four percent (Hat) to percent (Sngapore). The mean value s percent (EU) and the medan value s percent (Senegal, New Zealand). The value of 1995 agrcultural mports relatve to domestc value-added n agrculture (PENTR) measures mport penetraton relatve to the sze of the domestc sector. Smlar to the measures of trade balance, as agrcultural mports ncrease relatve to domestc value-added, producers may perceve havng more to gan from regulatory nterventon, but conversely, consumers have more to lose f the dfference between the domestc and world prce ncreases as a result of regulatory nterventon that nsulates the domestc sector from nternatonal markets. A postve statstcal relatonshp between ths measure and regulatory nterventon would mply the former effect domnates the latter, and a negatve relatonshp mples the opposte. As shown n Table 5.3, agrcultural mports tend to be hgher relatve to domestc producton n countres that are more dependent on the nternatonal export markets. The correlaton between agrcultural mport penetraton and the overall percentage of exports n GDP s To construct the measure of mport penetraton, the value of 1995 agrcultural mports, reported n FAO (1997), s utlzed, together wth the 1995 domestc value-added n agrculture, reported n World Bank (1997). Values range from percent n Hong Kong to 2.66 percent n Inda. The mean value s percent (Unted Kngdom, Trndad and Tobago, and the Netherlands). The medan value s percent (Malaysa). As shown by the range of values and the large standard devaton, relatve to the mean value, there s substantal varablty n agrcultural mport penetraton among countres. The post-uruguay Round weghted appled tarff on agrcultural mports, ncludng the calculated tarffcaton of non-tarff barrers, (TAPPL) provdes a measure of economc-based protecton provded through trade polces other than techncal measures. Possble substtuton between polcy nstruments s ndcated f protecton provded by techncal barrers ncreases as other barrers decrease. The dfference between commtments to 1999 bound tarff levels and projected post-uruguay Round appled tarff levels on agrcultural mports (SLACK) provdes a measure of slack, or the ablty of a government to ncrease ts tarff level and stll meet ts oblgatons under the GATT Agreements. Data for both the projected 1999 appled mostfavored-naton tarff rate and the dfference between the bound and projected 1999 rates s found n Fnger, Ingco, and Rencke (1996). 48 Values for the projected 1999 tarff rate range from percent (Brazl) to percent (Japan). The mean value s 8.02 percent (Sngapore) and the medan value s 1.10 percent (a number of countres, prmarly former republcs of the USSR). Values can be negatve as the projected future tarff ncludes the tarffcaton of nontarff barrers. As export subsdes are converted to tarffs they enter the calculaton as negatve values. Therefore, countres wth negatve projected future tarffs are those that subsdze agrcultural exports and the mean values for tarffs are lower than those usually reported n the lterature for agrcultural products. Values for the dfference between the bound 48 The estmated of projected 1999 weghted tarff rates on agrcultural products ncludes the tarrfcaton of nontarff barrers (see Ingco 1995 for calculaton procedures). 109

11 tarff and the projected future appled tarff range from 0.20 percent for Japan to percent for Zmbabwe. The mean value s percent (Chle) and the medan value s percent (Malaw, Zamba). There are several proxy measures ncluded n the models that are used to represent a retalatory tradng regme. The average appled tarff faced by agrcultural exports from each country (TAPPLF) and the reducton n such tarff levels as mandated by the Uruguay Round Agreements (TREDF) are measures of trade restrctveness faced by the agrcultural export products of each country. A polcymaker mght mplement a retalatory strategy of ncreased economc-based protecton when faced wth more restrctve polces on domestc agrcultural exports. Fnger, Ingco, and Rencke (1996) provdes data for both the projected 1999 average tarff rate appled to agrcultural exports from each country and the reducton n such a rate as a result of Uruguay Round commtments for agrcultural products exported by country. 49 The average appled tarff rate faced by agrcultural exports ranges from percent (Bolva) to percent (Barbados). The mean value s percent (Zmbabwe) and the medan s 6.60 percent (Japan). The reducton n the average appled tarff rate faced by agrcultural exports due to Uruguay Round commtments ranges from 1.20 percent (Venezuela) to percent (Turkey). The mean s percent (Sweden) and the medan s percent (Ngera, Ghana). Membershp n the WTO ndcates a country has agreed to abde by the provsons of GATT 1994 ncludng the SPS and TBT Agreements and the Agreement on Agrculture. Ths s represented by a dummy varable (WTO) n the econometrc models. Countres that face the possblty of nternatonal challenge n the WTO may be less lkely to provde economc-based protecton through techncal barrers. Conversely, countres that have agreed to lmt the future use of tarffs and other non-tarff barrers may be more lkely to ncrease the use of questonable techncal barrers n order to mantan protecton for domestc agrculture. A negatve coeffcent would ndcate that the former effect domnates the latter. Seventy-one percent, or 95, of the countres n the Survey were members of the WTO b.2. Characterstcs of Agrculture There are fve ndependent varables ncluded n the emprcal models to represent the characterstcs of agrculture among countres. Per-capta value-added n agrculture (VAAG) s a measure of the absolute contrbuton of agrculture to an economy. As the per-capta value- 49 The estmate of the 1999 average appled tarff rate on agrcultural products ncludes the tarff equvalents of nontarff barrers whch were constraned by the Uruguay Round Agreement on Agrculture based on rates specfed n the country schedules converted to ad valorem equvalents usng World Bank prce projectons (see Ingco 1995 for calculaton procedures). 50 For comparson, smlarly constructed proxy varables for characterstcs of trade n the U.S. equal $ per capta agrcultural trade balance, percent growth n trade balance, 11.0 percent of exports n GDP, percent agrcultural mport penetraton, 10.8 percent projected appled tarff rate on agrcultural products, 0.1 percent dfference between the post-uruguay Round bound and appled tarff rates, 37.6 percent average tarff faced by U.S. agrcultural exports, 46.7 percent reducton n the average appled tarff faced, and Membershp n the WTO. 110

12 added ncreases the ablty of agrcultural producers to both organze and nfluence polcy may ncrease, ether because the total number of ctzens s smaller or the total ncome of the sector s larger. Total value-added n agrculture, n U.S. dollars, s reported n World Bank (1997). When the per-capta value-added of agrculture s calculated from ths data, values range from $26.31 (Mozambque) to $ (Greece). The mean value s $ (Ukrane, Georga) and the medan value s $ (Mexco, El Salvador). The percentage GDP from agrculture (GDPAG) provdes a measure of relatve contrbuton of agrculture to the economy. In countres where the agrcultural sector s relatvely larger, producers may fnd t harder to organze and effectvely nfluence polcy decsons (Olson 1985). Table 5.3 shows a postve correlaton of between the percentage GDP from agrculture and the percentage of labor employed n agrculture. In addton, pror research results ndcate that the relatve contrbuton of agrculture to an economy tends to decrease as the economy grows. Polcymakers are more lkely to tax agrculture n less wealthy economes, where the relatve contrbuton of agrculture s larger, and provde protecton to agrculture n more wealthy economes, where the relatve contrbuton s smaller (Honma and Hayam 1986; Grll and Sassoon 1990). There s a correlaton n the data set between the percentage contrbuton of agrculture to GDP and the level of GDP. As shown n Table 5.2, the percentage of agrculture n GDP ranges from zero (Sngapore, Kuwat, Hong Kong) to percent (Georga). The mean contrbuton of agrculture s percent (Turkey) and the medan value s percent (Croata, Kazakhstan, Ecuador, Tunsa, Angola). The number of ndvduals employed n the agrcultural (LABAG) sector s one proxy for effectve poltcal nfluence. However there s some dsagreement n the lterature over the expected sgn of ths varable n determnng protecton levels. As the percentage labor force n agrculture ncreases, there s a larger proporton of the populaton who may exert pressure for ncreased economc-based protecton. However, followng an Olsonan argument, as the percentage labor force n agrculture ncreases, the ablty of the group to organze, overcome free-rder problems, and lobby successfully for protecton decreases. The 1995 percentage labor force n agrculture by country s reported n World Bank (1997). The percentage of the labor force employed n agrculture ranges from zero (Sngapore) to percent (Malaw). The mean value s percent (Azerbajan) and the medan value s percent (Domncan Republc, Colomba). As a country becomes relatvely more captal ntensve, ts comparatve advantage n agrcultural producton ncreases (Anderson and Hayam 1986; Gardner 1993). The captal-land factor rato, measured as the number of tractors n use per 1000 hectares of arable and cropland (KL) serves as one proxy for nternatonal compettveness n agrculture. As the compettveness of the domestc sector ncreases, relatve producer ncentves to lobby for economc-based protecton may decrease. As shown n Table 5.3 when the number of tractors per 1000 hectares ncrease, the percentage of the labor force employed n agrculture decreases wth a correlaton coeffcent of 0.403, and the absolute contrbuton of agrculture to the economy, measured as per-capta value-added, ncreases wth a correlaton coeffcent of The number of tractors n use and hectares of arable and cropland by country n 1995 are reported 111

13 n FAO (1997). As shown n Table 5.2, the number of tractors per 1000 hectares ranges from approxmately 0.24 (Senegal) to (Japan). 51 The mean s tractors per 1000 hectares (Tajkstan and Barbados) and the medan value s (Albana). The agrcultural labor-land rato (LL) serves as one proxy for the potental economc gans to agrcultural workers. As the rato ncreases, the country s more labor ntensve n agrculture and the economc stakes for each ndvdual worker may decrease. Conversely, as the rato ncreases, agrcultural workers are more concentrated geographcally and may be better able to organze. A negatve sgn would ndcate that the former relatonshp domnates the latter. The correlaton between the labor-land rato and the percentage of the labor force employed n agrculture s As expected the correlaton between the labor-land rato and the captalland rato s negatve, but the relatonshp s not strong n ths data set, as evdenced by a correlaton coeffcent of only The number of agrcultural workers and hectares of arable and cropland by country are reported n FAO (1997). Values range from 0.01 (Australa) to 5.39 (Chna). The mean s 0.81 (Japan) and the medan s 0.38 (Armena) b.3. Characterstcs of the Aggregate Economy There are sx ndependent varables ncluded n the emprcal models to represent the characterstcs of the aggregate economy. Per-capta GDP (GDP) s a measure of ncome n the economy. As ndvdual ncome ncreases consumers demand more safety, producers spend more money to lobby polcymakers, consumers are less lkely to protest prce ncreases, and protecton may ncrease. Data for per-capta GDP, n U.S. dollars, s reported n World Bank (1997). Values range from $91.81 (Mozambque) to $42, (Swtzerland). The mean value s $ (Argentna) and the medan value s $ (Belarus). The level of prvate consumpton (PVT) s an addtonal proxy for the potental economc gans to consumers from lmtng regulatory nterventon. As per-capta prvate consumpton ncreases relatve losses to ndvdual consumers from a margnal prce ncrease due to regulatory nterventon are smaller, but the absolute losses may be larger. Therefore the expected sgn on ths varable s ambguous. A hgh correlaton coeffcent of between the level of prvate consumpton and GDP ndcates that prvate consumpton s also servng as a measure of ncome. Agan, as ndvdual ncome ncreases consumers demand more safety, producers spend more money to lobby polcymakers, consumers are less lkely to protest prce ncreases, and protecton may ncrease. Values for 1995 per-capta prvate consumpton, n U.S. dollars, are reported n World Bank (1997). Per-capta prvate consumpton ranges from $69.06 n Mozambque to 51 Japan has the hghest captal-land factor rato at tractors per 1000 hectares. Ths s lkely to be more reflectve of hgh levels of protecton for the agrcultural sector rather than hgh levels of nternatonal compettveness, ndcatng that there s some msspecfcaton n the use of KL as a proxy for comparatve advantage. 52 For comparson, smlarly constructed proxy varables for characterstcs of agrculture n the U.S. equal $ per-capta value-added n agrculture, 2.00 percent of GDP from agrculture, 3.00 percent of the labor force employed n agrculture, tractors per 1000 hectares, and 0.02 agrcultural workers per hectare. 112

14 $25, n Swtzerland. The mean value s $ per capta (Uruguay). The medan value s $ per capta (Tunsa). The average annual growth n nomnal food prces (GFPRCE) s one proxy measure for overall condtons n the economy. When food prces are rsng rapdly, consumers may not recognze addtonal ncreases resultng from regulatory nterventon n the nternatonal markets. The real prce ncrease effects of regulatory nterventon may be masked by hgh nflaton. Conversely, as the ncrease n food prces due to other factors s greater, consumers may be less wllng to wthstand further ncreases due to regulatory nterventon. The growth n food prces s used nstead of an alternatve measure of overall nflaton snce the focus of the study s regulatory effects on U.S. agrcultural exports. The growth n food prces wll nclude both real and nomnal prce changes n agrcultural markets. Values for the percentage average annual growth n food prce range from 0.50 percent n Fnland to percent n Brazl. The mean value s percent (Turkey) and the medan value s percent (Czech Republc). Only 97 observatons are ncluded for growth n food prces due to lmted data avalablty. The percentage of domestc rural populaton (RURAL) serves as a broader proxy of support for agrcultural nterests n the economy than the percentage of agrculture n the labor force. Agan, the sgn of ths varable s ambguous. As the percentage of rural populaton ncreases, polcymakers are more lkely to support rural nterests through economc-based protecton for domestc agrculture. However, the ablty of rural ctzens to organze decreases as ther number ncreases. The percentage of rural populaton s hghly correlated wth the percentage of the labor force employed n agrculture, wth a correlaton coeffcent of 0.830, as shown n Table 5.3. Percentage rural populaton by country n 1995 s reported n FAS (1997). Values range from zero (Sngapore) to percent (Malaw and Oman). The mean s percent (Slovaka) and the medan s percent (Georga, Ecuador). The rato of the offcal to parallel exchange rate (XCHG) measures over- or undervaluaton of the domestc currency. When the domestc currency s overvalued, the volume of exports decreases and the volume of mports ncrease. Countres wth an offcal exchange rate lower than the parallel rate (overvalued currency wth a rato less than one) may be more lkely to mpose questonable techncal barrers n order to offset the negatve effects of the exchange rate polcy on the net trade balance. Data on 1995 exchange rate ratos are reported n World Bank (1997) but the number of observatons s lmted to 89. The exchange rate rato ranges from 0.20 (Syra) to 1.10 (Jordan). There are many countres that have offcal rates equal to the parallel rate so that the rato s equal to one, resultng n a mean value equal to 0.93 and a medan value equal to one. There s relatvely lttle varaton n the reported exchange rate ratos, a shown by a standard devaton of only The percentage of government consumpton n GDP s used as one measure of nwardlookng market orentaton n the economy. As the proporton of government consumpton ncreases and a country becomes less market-orented n the domestc economy, polcymakers may be more lkely to provde economc-based protecton for the domestc agrcultural sector 113

15 through regulatory nterventon. The correlaton coeffcent s only between the percentage of government consumpton n GDP and the percentage of exports n GDP, ndcatng that countres that are less market-orented n the domestc economy are not necessarly less marketorented n the nternatonal economy. The 1995 percent of government consumpton n GDP s reported n World Bank (1997). Values range from four percent (Domncan Republc) to percent (Angola). The mean value s percent (Russa, Span, Italy, Algera, Tunsa) and the medan value s percent (Albana, Morocco, New Zealand, Kenya, Kazakhastan, Panama, Bulgara, Belgum, Ireland) b.4. Measurement Issues Snce prmary data s beng used to measure the ncdence and mpact of economc-based protecton provded through techncal barrers, t s necessary to account for possble measurement bas n the resultng data set assocated wth the survey desgn characterstcs. As descrbed n Chapter Four, the 1996 USDA survey process occurred n four stages n order to obtan consstency and accuracy of the results. One remanng source of possble bas s the relance on FAS feld offces as the prmary data collecton agents. There are 50 feld offces whch cover 132 countres and two regonal tradng blocks representng 98 percent of the 1996 U.S. export market for agrculture, forestry and fsh products. The physcal locaton of a FAS post n a country may ncrease the awareness of feld personnel to ssues wthn that country relatve to other countres that are covered by the post. Therefore, the amount of protecton observed and reported n the survey may be too hgh n countres where feld posts are located or too low n countres where feld posts are not located. A dummy varable s ncluded n the econometrc models to account for the presence or absence of a FAS post n each country. A postve sgn on the FAS varable would ndcate that the presence of an FAS post has an mpact on the results, but would not ndcate whether the protecton s over-represented n those countres wth an FAS post or under-represented n countres wthout an FAS post. Countres where an FAS post are located are ndcated wth an astrk n Appendx B. 5.2.b.5. Relatonshp to the Poltcal Economy Paradgm Taken together, the set of ndependent varables selected for ncluson n the econometrc models gves a broad characterzaton of the lkely poltcal economy nfluences on regulatory decsons to enact questonable techncal barrers to U.S. agrcultural exports. As the varables selected are all proxy measures for the poltcal economy relatonshps descrbed n Chapter Three, there s more than one possble mappng to the categores of ndvdual agents, effectve poltcal nfluence, polcymaker preferences, and nsttutonal structures, as depcted earler n 53 For comparson, smlarly constructed proxy varables for characterstcs of the aggregate economy n the U.S. equal $26, per-capta GDP, $17, per-capta prvate consumpton, 3.5 percent average annual growth n food prces, percent of rural populaton, an exchange rate rato of one, and 16 percent of government consumpton n GDP. 114

16 Fgure 3.3. One such reorganzaton of the ndependent varables s depcted n Table 5.4 and descrbed brefly below. Table 5.4. One possble reorganzaton of the ndependent varables nto categores of the poltcal economy paradgm Varable Varable Defnton Indvdual Agents KL 1995 captal /land rato TRDBAL 1995 agrcultural trade balance GTDBAL growth n trade balance USTRD 1995 agrcultural trade balance wth the U.S. GUSTRD average growth n agrcultural trade balance wth the U.S. LL 1995 labor/land rato PENTR 1995 agrcultural mport penetraton relatve to domestc producton PVT 1995 prvate consumpton GFPRCE average annual growth n nomnal food prces Effectve Poltcal Power LABAG 1995 labor force employed n agrculture RURAL 1995 rural populaton GDPAG 1995 GDP from the agrcultural sector VAAG 1995 value-added n agrculture GDP 1995 GDP Polcymaker Preferences GDPEXP 1995 percentage of exports n GDP GOVT 1995 percentage of government consumpton n GDP Insttutonal Structures XCHG 1995 rato of offcal to parallel exchange rate TAPPL Projected 1999 appled MFN average tarff rate for agrcultural mports SLACK Dfference between 1999 bound tarff rate and projected appled tarff rate on agrcultural mports TAPPLF Projected 1999 MFN tarff rate faced by agrcultural exports TREDF Change n projected 1999 tarff faced by agrcultural exports as a result of Uruguay Round commtments from the country s tradng partners WTO 1996 WTO membershp Measurement Issues FAS 1996 FAS post n country 115

17 There are nne varables that may be consdered as proxes for the stakes of ndvdual agents. As the comparatve advantage of the domestc agrcultural sector decreases, the stakes for producers ncrease, and the relatve ncentves for producers to lobby for economc-based protecton may ncrease. As a country becomes relatvely more captal ntensve, ts comparatve advantage n agrcultural producton ncreases (Anderson and Hayam 1986; Gardner 1993). Therefore the captal to land ratos are one measure of producer stakes. Bhagwat (1982b) argued that pressure from mport competton could be vewed as an addtonal sgn of comparatve advantage. The four measures of agrcultural exports relatve to mports (total agrcultural trade balance, the change n agrcultural trade balance, blateral trade balance wth the U.S., and the change n the blateral trade balance) can thus be vewed as further measures of producer stakes. The labor-land rato serves as an addtonal proxy for the potental economc gans to agrcultural workers, as descrbed above. The value of agrcultural mports relatve to the sze of the domestc producton sector serves as another measure of Bhagwat s pressure from mport competton and represents relatve stakes to both producers and consumers. The percentage of prvate consumpton n GDP and the average annual growth n nomnal food prces are addtonal proxes for the potental economc gans of consumers from lmtng regulatory nterventon. There are fve varables ncluded n the econometrc models that mght be consdered proxes for effectve poltcal nfluence. The percentage of the labor force employed n agrculture reflects the ablty of producers to organze. The percent of the populaton that s rural s a broader measure, whch may account for the secondary mpact of changng producer surplus n rural communtes as well as the drect mpact of regulatory decsons on agrcultural producers. The percentage of GDP derved from the agrcultural sector s used as a measure of the relatve contrbuton of agrculture to the economy. Per-capta value-added n agrculture s used to measure the absolute contrbuton of agrculture n the economy. As the absolute and relatve contrbuton of agrculture to the economy ncreases, producers may have a greater ablty to nfluence regulatory decsons. Conversely, as the absolute and relatve contrbuton of agrculture ncreases, producers may fnd t harder to organze and ther effectve nfluence may decrease. Per-capta GDP s one measure of ncome that may reflect agents fnancal ablty to nfluence polcy. Smlar to prevous cross-country studes there are no readly avalable proxes to measure polcymaker preferences. Two varables ncluded n the econometrc specfcatons mght be consdered ndcators of such preferences. The percentage of exports n GDP s one measure of outward-lookng market orentaton and the percentage of government spendng n GDP s one measures of nward-lookng market orentaton. These varables may reflect polcymaker preferences for freer trade and open markets nternally. There are sx varables that could represent nsttutonal structures. The rato of the offcal to parallel exchange rate measures the mpacts of domestc currency manpulaton on trade. Two varables are ndcators of a substtuton tradng strategy between polcy nstruments: 116

18 the post-uruguay Round weghted average most-favored-naton tarff rate appled to agrcultural products and the dfference between the bound tarff and the post-uruguay Round appled tarff on agrcultural products. Two addtonal varables are ndcators of a retalatory tradng strategy: the average appled tarff faced by agrcultural exports and the change n such a tarff level mandated by the Uruguay Round Agreements. 54 WTO membershp ndcates f the country s bound by the new Uruguay Round restrctons Determnants of the Presence or Absence of Questonable Techncal Barrers A smple way to measure the ncdence of questonable techncal barrers s by ther presence or absence. The 1996 USDA Survey ncludes observatons for 134 countres: 63 appled one or more questonable techncal barrers to U.S. agrcultural exports n 1996 and 71 dd not apply such barrers. 5.3.a. Statstcal Model Suppose a bnary dependent varable, YN, s used to measure the presence or absence of questonable techncal barrers where the sample results from N Bernoull trals wth two possble outcomes for each tral; a country ether enacts questonable techncal barrers to U.S. agrcultural exports or does not. The two outcomes are proxes for a contnuous response varable y * = b k xk + u (Maddala 1983). In ths model, y * s the level of economc-based protecton provded by questonable techncal barrers wthn country ( = 1,2,.n), x k s a vector of k descrptve varables for country (k = 1,2,.m), b k s a vector of parameters, and u s a scalar error term. Snce the latent varable y * s unobservable n ths model, ts scale s arbtrary and has only ordnal meanng. The observed varable, YN, s the presence of techncal barrers that are potentally subject to challenge and takes the form y = 1 f y * > 0 y = 0 otherwse. Snce the dependent varable, y, s both qualtatve and dscrete, the regresson technques used for contnuous data are not approprate n ths case. We can construct probabltes from the equatons above so that P( y = 1) = P( y * > 0) = P( u = F ( b x k k > b x ) k k ) = P( u < m k = 1 b k x k ) 54 Note that f the blateral agrcultural trade balance wth the U.S. s nterpreted as representng a retalatory trade strategy, the varables USTRD and GUSTRD mght be more reflectve of nsttutonal structures rather than the stakes of ndvdual producers. 117

19 where F s the cumulatve dstrbuton functon (cdf) for u, the error term. When F s the cdf of a normal dstrbuton the PROBIT model s used. 55 If φ s the standard normal densty functon, the PROBIT model takes the form F( b k x k 1 bk xk / σ ) = 2π e 2 u 2 du where u ~IN (0, and P( y = 1 x ) = φ( b x ) k k k 2 σ ) The model s solved usng the method of maxmum lkelhood where the estmate of the parameters, b k, s the value for whch the probablty of the observed data takes ts hghest value. Snce the model actually estmates standard devaton of u = b k / σ and not b k, the PROBIT model normalzes the Amemya (1981) showed that for a unvarate dchotomous model (whch ncludes the PROBIT specfcaton) the log of the lkelhood functon s globally concave. Therefore a unque soluton to the PROBIT model wll yeld a global maxmum. The coeffcent estmates b k are asymptotcally unbased and effcent pont estmates of the b k that measure the relatonshp between the exogenous varables x k and the underlyng dependent varable y * (Aldrch and Nelson 1984). T-statstcs are calculated for each b k to test the null hypothess that the varable x k has no effect on the underlyng dependent varable y*. Unlke the standard regresson models, typcal scalar measures of sgnfcance are not approprate evaluatve crtera for qualtatve response models (Amemya 1981). Although no sngle method has been wdely accepted, lke an R 2 for lnear models, several alternatve measures have been proposed and used n emprcal studes. The frst measure of sgnfcance s the percentage of correct predctons (Amemya 1981). Predctons are calculated for each observaton where = F F ( b k xk ) where b k s the estmated b k. Then let 55 When F s the cdf of a logstc dstrbuton the LOGIT model s used. The logstc dstrbuton functon has slghtly heaver tals, but usually can t be dstngushed statstcally from the normal dstrbuton unless there s an extremely large number of observatons (Maddala 1983). Aldrch and Nelson (1984) found the two curves to be so smlar as to yeld essentally dentcal results. 56 * 2 The LOGIT model normalzes the standard devaton of y = π / 3 so the coeffcents of the alternatve model can be approxmated wth a mathematcal transformaton (Amemya 1981; Aldrch and Nelson 1984; Agrest 1996). 118