ASYMMETRIC TRADE FLOWS AND THEIR IMPLICATIONS FOR COMPETITIVENESS, EFFICIENCY AND TRADE

Transcription

1 ASYMMETRIC TRADE FLOWS AND THEIR IMPLICATIONS FOR COMPETITIVENESS, EFFICIENCY AND TRADE YOUNGJAE LEE Department of Agrcultural Economcs and Agrbusness Lousana State Unversty AgCenter 242A Martn D. Woodn Hall Baton Rouge, LA Phone: Fax: E-Mal: P.LYNN KENNEDY Department of Agrcultural Economcs and Agrbusness Lousana State Unversty AgCenter 181 Martn D. Woodn Hall Baton Rouge, LA Phone: Fax: E-Mal: Selected Paper prepared for presentaton at the 2016 Agrcultural & Appled Economcs Assocaton Annual Meetng, Boston, Massachusetts, July 31-August 2 Copyrght 2016 by Youngjae Lee and P.Lynn Kennedy. All rghts reserved. Readers may make verbatm copes of ths document for non-commercal purposes by any means, provded that ths copyrght notce appears on all such copes.

2 Asymmetrc Trade Flows and Ther Implcaton for Compettveness, Effcency and Trade Abstract: The asymmetrc trade flow of agrcultural goods can serve as a clue to help understand unobservable agrcultural compettveness, nfrastructural effcency for exports, and net openness to mports. In order to dentfy these three factors from agrcultural trade data, we adapt a trade model developed by Eaton and Kortum. Unlke Eaton and Kortum, we nterpret specfc country dummy varables as proxes representng these three factors. Ths study makes four mportant fndngs. Frst, agrcultural trade flow s strongly related to net openness to mports but less to agrcultural compettveness. Second, agrcultural compettveness s more related to land endowments than economc development. Thrd, economc development mproves nfrastructural effcency for agrcultural exports. Fnally, exstng agrcultural mport restrctons are shown to be puntve. Key words: agrcultural compettveness, gross and net openness to mports, gross compettveness, nfrastructural effcency. JEL codes: F11, F13, F14, F18, O19, Q17, Q Introducton Agrcultural trade data reveals asymmetrc trade flows. Such asymmetry n the flow of goods can be explaned through resource endowments, economc development, and market openness. For example, consder two countres; country L s a land-locked, land abundant country whle country S s a small sland country. Naturally, country L s more effcent n producng agrcultural products than s country S whle country S s more effcent n producng fshery products than s country L. Gven ths, the relatve and absolute prce of agrcultural products s lower n country L than n country S and the opposte holds for fshery products. Despte the fact that each of these countres has both absolute and comparatve advantage n an ndustry, there s no trade between these two countres because trade costs, gven the poor market nfrastructure, are greater than the dfference n prces between the two countres. As economes develop they mprove ther market nfrastructure, thus reducng the cost of movng ther products from domestc producer to a foregn fnal consumer. Under ths crcumstance, each country can specalze n producng the good for whch ts opportunty cost s lowest, creatng opportuntes for gans from trade between the two countres. Wthout trade barrers, the two countres mantan ther trade balance at the equlbrum pont. Later, country S mposes a tarff on mports from country L, whle country 1

3 L mantans no trade barrer. Ths unbalanced market openness shocks the trade balance between the two countres, generatng asymmetrc trade flow,.e., more from country S and less from country L. Based on ths smple llustraton, ths study emphaszes three possbltes as follows: 1) resource endowments determne agrcultural compettveness, 2) economc development promotes market effcency, and 3) market openness determnes agrcultural trade flow. Knowng that resource endowments, economc development, and market openness are more complex n the real world, we attempt to descrbe how these three factors can explan asymmetrc agrcultural trade flows. In the tradtonal gravty trade model, trade cost s treated as symmetrc, mplyng that costs n movng agrcultural products from country L to country S s equal to costs n movng from country S to country L. Recently, however, trade economsts have examned components of trade costs nducng asymmetrc trade flow. Bergstrand, Egger, and Larch (2013) state that, n realty, trade costs (and trade flows) are not blaterally symmetrc. They confrmed asymmetrc effects of natonal borders n blateral trade between the U.S. and Canada by usng an alternatve trade model allowng for asymmetrc blateral trade costs. Trade costs are defned as the dfference between the margnal cost of producton by the domestc producer and the prce pad by the end user n a foregn country (Khan and Kalrajan, 2011 and Anderson and van Wncoop, 2004). Based on where trade costs occur, we categorze trade costs as () behnd the border costs ncurred n the exportng country j, T j,x, () between the border costs ncurred between exportng and mportng country j and, T b,x, and () beyond the border costs ncurred n the mportng country, T,x. Then, the end user n the mportng country pays 1 T j,xt b,xt,x $ for an agrcultural good x whose margnal cost of producton s $1 per unt n exportng country j. 1 One mportant pont s that T b,x depends upon the dstance, d, between j and, mplyng symmetry because the dstance and nsurance premum on dentcal sea routes s equal regardless of the drecton of trade. On the contrary, T j,x and,x T are asymmetrc because they are affected by polcy-nduced factors 2

4 (such as tarff and non-tarff barrers) and market-nduced factors (such as marketng costs and retaler/wholesaler margns). 2 If between the border costs comprse the majorty of trade costs, trade costs wll be approxmately symmetrc. However, trade costs are asymmetrc f behnd and/or beyond the border costs comprse the majorty of trade costs. Tradtonally, the standard gravty approach to modelng trade costs assumes that trade costs are symmetrc,.e., trade costs are a functon of dstances between borders. Recent studes, however, show that advances n technology reduce transportaton costs, but overall trade costs are stll hgh (Khan and Kalrajan, 2011; Alvarez and Lucas, Jr., 2006; Anderson and Wncoop, 2004; Bordo, 2002; and Bruce, 1997). Ths mples that behnd or beyond the border costs may be a relatvely larger component than between the border costs. In ths case, asymmetrc trade costs wll control blateral trade flows. Accordng to agrcultural trade data, trade flows between hgh ncome countres are much greater than those between low ncome countres. Ths mples that the standard gravty model s not the optmal approach when hgh and low ncome countres are ncluded n the same analyss. For example, symmetry restrcts flexblty n explanng trade flows because the prce pad by the end user n an mportng country s determned not only by margnal costs but also by trade costs. Also, trade costs are affected dfferently by the mporter and exporter, resultng n a volaton of the symmetrc condton. In ths context, t s more mportant to llustrate an ndvdual country s margnal cost and asymmetrc components of trade costs affectng trade flows. However, an ndvdual country s margnal cost and asymmetrc components of trade costs must be estmated because they cannot be drectly observed. In ths challenge, based on trade model developed by the prevous studes (Eaton and Kortum, 2002; Waugh, 2010; and Remer and L, 2010) we try to derve blateral trade share equaton as a functon of dstance, margnal cost, behnd the border cost, and beyond the border cost. Then, the three nongeographcal varables n the blateral trade share equaton are defned by parameters determnng ndvdual country s agrcultural competveness, nfrastructural competveness, and market openness. In order to acheve the objectves, ths study s conducted as follows. The followng secton presents the theoretcal framework, dscussng agrcultural prces, trade share, and rents. In the thrd secton, three observatons 3

5 from agrcultural trade data and ther mplcaton n asymmetrc trade flows are dscussed. In secton four, we examne about where asymmetrc trade flows come from and the effects on agrcultural trade flows of asymmetrc components of trade costs. In secton fve, we conduct emprcal analyss n whch we estmate ndvdual country s margnal cost, behnd the border cost, and beyond the border cost. To dentfy the mplcatons of asymmetrc components of trade costs for nternatonal agrcultural trade, two dfferent counterfactuals are smulated n secton sx. The fnal secton concludes the paper wth recommendatons for further research. 2. Theoretcal Framework Consder a world economy composed of N countres. An ndvdual country j produces a set of agrcultural goods, q j x, whch are tradable. There s a contnuum of agrcultural goods ndexed by x 0,1. It s assumed that the prmary (non-produced) factor of agrcultural producton s land and an ndvdual country j s endowed wth an endowment of agrcultural land, L j, and total factor productvty, z j x, vares across agrcultural goods wth a common densty,. To produce agrcultural goods n an ndvdual country j, land and productvty are combned by the followng producton technology functon: q j x z x f L. j In order to produce a gven agrcultural good x, the land requrement depends upon the level of productvty wth constant proportonalty, z x j. Remer and L (2010) noted that there s a relatvely permanent dfference n weather, sol qualty, or technology across countres and that one country wll be relatvely more productve than another (and vce versa) at producng varous pars of agrcultural goods, whch can thus create an envronment conducve to agrcultural trade n the nternatonal market. Wth land beng the prmary factor of agrcultural producton, agrcultural productvty can be defned as output per acre.e., yeld. Yeld s a weather-nduced random varable whch s dstrbuted ndependently and exponentally wth the yeld parameter,. Each country s mean yeld s proportonal to j, wth the constant of proportonalty ndependent of the j 4

6 country. Then, the margnal cost of producng agrcultural goods s j r j / j mc, where r s the rental rate of agrcultural land. 3 So, a country wth a relatvely smaller r / s, on average, more effcent at producng a bundle of agrcultural goods. controls the dsperson of effcency levels,.e., a larger (smaller) results n more (less) varaton n effcency levels relatve to the mean. As ncreases (decreases), t ncreases (decreases) the lkelhood that the margnal costs of the two countres producng the same agrcultural good wll be dfferent, thus leadng to the creaton of more (fewer) ncentves to trade. In ths sense, controls the degree of comparatve advantage. The end user n a foregn country has symmetrc preferences wth respect to bundles of agrcultural goods, wth utlty gven by U 1 0 q x 1 dx 1. The consumer s problem s that of maxmzng utlty for a gven budget by purchasng a bundle of agrcultural goods q(x) from all n number of countres. Then, agrcultural trade occurs n the followng manner. The fnal users purchase each agrcultural good from the lowest prce across all countres. Four factors nfluence whch country s agrcultural good s the lowest prce: () margnal cost, () behnd the border cost, () between the border cost, and (v) beyond the border cost. Therefore, agrcultural trade depends not only on agrcultural productvty but also trade costs. Furthermore, trade costs depend not only on dstance but also mporter and exporter costs. In the followng sectons, we descrbe how agrcultural prces, trade shares, and rents are determned. 2.1 Prce Index for Aggregate Agrcultural Products Due to the ncdence of trade costs, there exsts a gap between the prce pad by the end consumer n mportng country and the margnal cost of producton n exportng country j. Therefore, agrcultural prce n country wll be affected by margnal cost parameters (λ j, θ, and r j ) and blateral trade costs (τ ). The prce ndex for country can be derved usng the moment-generatng functon for extreme value dstrbuton (Eaton and Kortum, 2002) as follows: 5

7 (1) 1 / 1 N 1 p r j j1 1 j, where Γ s the Gamma functon used to express certan types of defnte ntegrals. 4 In equaton 1, s the trade cost from country j to country. The defnton of prce shows proportonal relatonshps between prce and margnal cost, and prce and trade costs. In supplyng agrcultural goods to the fnal user n the mportng country at $1 per unt whch s the lowest prce, the margnal cost of the good n an exportng country s 1 1 / per unt. If trade costs ncrease to, then the country cannot export ther product because the country cannot reduce the margnal cost below 1 /. 2.2 Agrcultural Trade Share s country s expendture share on agrcultural goods from country j. It s also the fracton of all agrcultural goods that country mports from country j. Snce there s a contnuum of agrcultural goods, computng ths fracton s reduced to fndng the probablty that country j s the low-cost suppler to country gven the jont dstrbuton of effcency level and trade costs for any agrcultural good x. Keepng ths n mnd, the expresson for country j s agrcultural trade share wth country s expressed as: (2) 1 rj j N 1 r l 1 l l. l Note that the sum across j for fxed must add up to one,.e., 1. The defnton of trade share shows an nverse relatonshp between trade share and margnal cost of and trade cost of exportng country j and a proportonal relatonshp between trade share and margnal cost of and trade cost of all the other rval countres. j 6

8 2.3 Rent The total value of agrcultural goods produced n country k s equal to r L k k. Also, the total expendture on agrcultural goods n country k can be defned as r L. Here, the parameter 1 represents the noton k k k that agrcultural consumpton s greater than agrcultural producton n country k, mplyng country k s a net mporter of agrcultural goods. The parameter k 1 represents the dea that agrcultural producton s greater than agrcultural consumpton n country k, mplyng country k s a net exportng country. Fnally, the parameter k 1 represents the dea that agrcultural producton s equal to agrcultural consumpton n country k, mplyng country k s a zero trade balancng country. Also, mports are defned as n Imports r L, whch s the total value of agrcultural goods that country k consumes from k k k jk kj k abroad. Smlarly, exports are defned as n Exports r L, whch s the total value of agrcultural k k goods that countres abroad purchase from country k. Then, the rent of a zero trade balancng country k, r L k k n k jk kj n k r L k s, (3) r k n k L k L 1 k k kk r. 3. Observatons n Agrcultural Trade Flow Ths study revews the observatons of agrcultural trade data to understand asymmetrc components of trade cost across countres. As mpled by asymmetry n blateral agrcultural trade flows, trade costs could be asymmetrc. As a benchmark, ths study consders a base year of Twenty-three countres and twohundred ffteen ndvdual agrcultural goods are n the sample, whch represents almost 90 percent of total agrcultural expendtures n each country. Trade share,, s constructed as m, and home A e m 7

9 share,, calculated as 1 N j, where m s the value of agrcultural goods that country mports from country j: A s the total value of agrcultural producton n country, e s the total value of agrcultural exports of country to the world: and m s the total value of agrcultural mports of country from the world. Table 1 presents a matrx of trade shares for twenty-three countres. [Place Table 1 Approxmately Here] 3.1 Home Bas Home bas means that countres consume most of ther domestcally produced agrcultural goods,.e., data. Home bas n the data s seen by consderng the large values lyng along the dagonal of table 1 relatve to off-dagonal entres. Home bas for agrcultural goods s shown to decrease wth an ncrease n ncome. Ths s dfferent than Waugh s fndngs for manufactured goods. In the agrcultural trade data, the home bas of hgh, mddle, and low ncome countres are on average 0.79, 0.84, and 0.91, respectvely. A regresson of the logarthm of on the logarthm of per capta GDP n 2001 has a slope coeffcent of and s dfferent from zero statstcally. The statstcs show that hgh ncome countres purchase more from foregn producers than do low ncome countres Asymmetry n the Agrcultural Trade Matrx Table 1 shows that the mport share of country from country j s not equal to the mport share of country j from country,.e.,. For example, the mport share of the U.S. from Canada s 5.01% whle the j mport share of Canada from the U.S. s 39.76%, ndcatng asymmetrc trade flows. Also, most blateral trade takes place between hgh ncome countres. Table 1 shows that blateral trade share of agrcultural goods between hgh ncome countres s 88.42% whle blateral trade share of agrcultural goods between mddle (low) ncome countres s reduced to 6.17% (1.68%). Table 1 also shows asymmetrc trade flows n these three classfcatons. For example, the average mport shares of hgh ncome countres from mddle and low ncome countres are relatvely smlar 0.39% and 0.32%, respectvely, whle average mport shares 8

10 of mddle and low ncome countres from hgh ncome country are relatvely more dvergent wth 0.96% and 0.75%, respectvely. 3.3 Prces Dfference between Hgh and Low Income Countres 215 dsaggregated observatons of prce data of 23 countres were obtaned from the Food and Agrculture Organzaton of Unted Naton (FAO, 2001). The baskets of these goods are the same across 23 countres. Accordng to per capta GDP, 23 countres are dvded nto three dfferent categores.e., hgh ncome country, mddle ncome country, and low ncome country. Average prces of each agrcultural good are calculated n three groups. Sxty-two of the orgnal 215 agrcultural goods are elmnated because the prces of the agrcultural goods are not obtaned n all of the three groups. Usng 153 average prces n each of the three groups, fgure 1 s obtaned. Fgure 1 shows that aggregate agrcultural prces are dfferent between the relatvely rch and the relatvely poor countres. [Place Fgure 1 Approxmately Here] 3.4 Implcaton of Agrcultural Trade Data The three observatons make a straght forward mplcaton n modelng trade cost. Dvdng trade share by home share yelds the followng relatonshp between trade share normalzed by home share, margnal cost of producton, and trade cost: (4) 1 1 mc j mc. Equaton (4) s essentally an arbtrage condton. It ndcates that f mc mc, then country has the j ncentve to purchase relatvely more goods from country j because they are cheaper. Alternatvely, f trade costs between country and j are large, then country has fewer ncentves to purchase a good from country j. In order to descrbe where volatons of symmetrc trade flows come from, equaton (4) can be dvded by the opposng expresson relatng countres and j as follows: (5) j jj j 1 mc j mc 2. 9

11 In a symmetrc world, the term / / always equals one. Therefore, t can be nferred that j jj the devaton from symmetrc trade flows occurs for two reasons: (1) margnal costs of producton of agrcultural goods are dfferent, or (2) trade costs between the two countres are dfferent. If each country has smlar margnal costs of producton,.e., mcj / mc 1, asymmetrc trade flow comes from trade cost. As dscussed n the prevous secton, trade costs mght be asymmetrc because behnd and beyond the border costs are dfferent and relatve magntude of between the border cost s decreasng wth advance n technology. If the behnd the border cost s a major factor n asymmetrc trade flow, then asymmetrc trade flow comes from the exporter. In contrast, f the beyond the border cost s a key factor n asymmetrc trade flow, then asymmetrc trade flows come from the mporter. The next secton wll seek to determne where asymmetrc trade flows come from by examnng agrcultural trade data. 4. Asymmetry n Agrcultural Trade Flow 4.1 Emprcal Examnaton In ths secton, we provde three examples of fttng the blateral trade share nto trade data. The frst s a standard gravty approach n whch trade costs are assumed to be symmetrc. The other two examples are related to asymmetrc approaches. The second s an exporter approach by whch we assume trade costs contngent on behnd the border cost. The last approach s an mporter approach by whch trade costs are assumed to be contngent on beyond the border costs. For llustratve purposes, let us assume that there are only three countres and that country H s a hgh ncome country and countres M and L are a mddle and low ncome country, respectvely. Also, assume throughout the example that () the hgh ncome country exports more to other relatvely hgh ncome countres than to relatvely low ncome country, () the hgh ncome country, however, equally mports from both countres, and () countres M and L do not trade wth one another. These assumptons are generally consstent wth the observed agrcultural trade data. For further convenence, we assume that () all countres have the same land endowment wth mc H mc M mc, () land s the only factor of L producton, and () country H s land rent and yeld are normalzed to one,.e., r H 1 and 1. Then, 10 H

12 the matrx depcts the set of blateral trade shares normalzed by the mportng countres home share and can be expressed as: HM HL 1 mc mc 1 M 1 L mc MH M LH mc L L mc and where HM mc M LH mc MH M 1 LH mc L 1. In the frst approach, symmetry n trade flow restrcts the parameter space so {,, { HL, LH } L, and M L HM MH }. Snce country H exports more to country M than to country L, MH LH, and mports the same shares from both countres, HM HL, the symmetry assumpton on trade flow mples that the trade cost for country L to mport from country H must be greater than that M of country M,.e.,. Usng equaton (1), the aggregate prce n countres M and L s MH LH p M 1 / 1 / 1 / 1 / mc p mc M MH L L LH. The relatvely lower ncome country must have a hgher aggregate prce relatve to the relatvely hgher ncome country. Ths example s quanttatvely nconsstent wth the data: comparable aggregate prce for agrcultural goods s p p. M L The second approach (exporter approach) restrcts the parameter space so { MH, LH },.e., countres M and L face the same cost to mport from country H. Snce MH LH and HM HL, the trade cost of country L to mport from country H must be same wth country M,.e.,. Therefore, MH LH the aggregate prce n countres M and L s p p. The aggregate prce of agrcultural goods n the M L relatvely hgher ncome country must be equal to that of the relatvely lower ncome country. Ths example s also quanttatvely nconsstent wth the data. The thrd approach (mporter approach) restrcts the parameter space so { HM, HL },.e., country H faces the same cost to mport from countres M and L. Snce MH LH and HM HL, the 11

13 trade cost of country M to mport from country H must be less than that of country L,.e.,. MH LH Therefore, the aggregate prce n countres M and L s p p. The relatvely hgher ncome country must M L have a hgher aggregate prce of agrcultural goods as compared to the relatvely lower ncome country. Ths result s consstent wth the data. Therefore, the trade data suggests that the asymmetry of agrcultural trade flows comes from beyond the border costs ncurred by mporters. 4.2 Asymmetrc Components of Trade Costs Based on ths fndng, we seek to dentfy unobservable margnal cost, behnd the border costs, and beyond the border costs. In an attempt to do ths, we redefne equaton (4) as follows: 1 j (6) log S S logt b,x j, where S and S j are destnaton and source-country dummy varables, and logmc logt,x represents the effect on logarthmc normalzed trade share of margnal cost of and beyond the border cost j 1 of the mporter, and logmc logt j,x j represents the effect on logarthmc normalzed trade θ share of margnal cost of and behnd the border cost of the exporter, and logarthmc between the border 1 θ cost s a functon of dstance and border share, logt ( b,x ) d b. k as follows: j Subtractng from, we obtan an equaton representng asymmetrc components of trade cost j (7) log T j,x log T, x. Accordng to the mporter approach, we assume that behnd the border cost s a fracton of beyond the border cost. Then, behnd the border cost s defned as follows: (8) T j,x T,x, 12

14 where represents the relatonshp between behnd and beyond the border costs of agrcultural good x exported by country j to country. Next we calbrate margnal cost, behnd the border cost, and beyond the border cost as follows: (9) j 1 mc e, (10) T j,x (11) T,x e j 1 j e 1,, and where s a parameter representng the effect of the mporter on trade share (we call as gross j openness to mports) and s a parameter representng the effect of the exporter on trade share (we call j as gross compettveness for exports). 1 j 1 and logmc logt j,x Snce logmc logt,x θ θ j, we can separate net T,x mc openness nto mports,, and agrcultural compettveness of mporter,, from gross openness parameter,. Also, we can separate nfrastructural compettveness, T j,x, and agrcultural compettveness of exporter, mc j j, from gross compettveness parameter,. Then, the effects on the trade share of agrcultural compettveness, net openness to mports, and nfrastructural compettveness for exports can be obtaned by usng the defnton of sem-elastcty n equaton (6) as follows: 0 mc e 1 mporter effect of agrcultural compettveness on trade share, (12) 0 (13) 0 T, x e 1 0 mporter net openness effect on trade share, 13

15 0 mc j (14) e 1 0 exporter effect of agrcultural compettve on trade share, and (15) 0 T j, x e 1 0 exporter nfrastructural compettveness effect on trade share, where 0 and 5. Emprcal Analyss 0 represent trade and home share when S = 0 and S j =0, respectvely. If data s avalable for trade cost,, margnal cost, mc, and trade share,, equaton (4) can be used to estmate. The dffculty encountered here s that trade cost,, and margnal cost, mc, are unobserved. Eaton and Kortum, however, suggest that an estmate of s possble wth use of a smple arbtrage argument and dsaggregated prce data. So, we use a proxy for trade costs obtaned as: (16) logτˆ max 2log p (x) log p (x), x where max 2 denotes the second hghest value n a set of dsaggregated prces. 6 j Wth ˆ calbrated from equaton (16), we assume that overall margnal cost s the same across all 23 countres n estmatng n equaton (4). 7 To construct estmates of equaton (16), producer-level prce data for the year 2001 are obtaned from the Food and Agrculture Organzaton. For tradable agrcultural goods, ths dataset has two hundred ffteen categores for the twenty three countres n the dataset. The blateral trade data are also from 2001 and constructed n the same manner as descrbe n the prevous secton. To summarze, equaton (4) s estmated wth a proxy for trade cost drawn from equaton (16) utlzng ordnary least squares (OLS) omttng the ntercept term. The regresson yelds an estmate of Ths mples a ˆ of whch s smlar to that estmated by Waugh (0.1818) but s much less than those estmated by Eaton and Kortum (from 2.86 to 3.60) and Remer and L (from 2.83 to 4.96). These smlartes and dfferences come from the dfference n defnton of prce (equaton 1) and trade share (equaton 2). Ths study used the same defntons of prce and trade share as Waugh to determne the degree of comparatve advantage. Ths estmated value s used throughout the remander of ths paper. 14

16 To dentfy the effect on trade share of destnaton, source-country, and geographc barrer, we start at equaton (6). For estmaton, we use proxes for between the border costs because between the border costs are related to proxmty factors, such as dstance and shared border. 8 Dstance s accounted for by usng sx dummy varables representng dfferent ntervals of great-crcle dstance between captals. The assocated coeffcent, d k wth k = 1,2,,6, s the effect of dstance between and j lyng n the k th dstance nterval. Intervals are n mles: [0, 375); [375, 750); [750, 1500); [1500, 3000); [3000, 6000); and [6000, maxmum]. The varable b captures the effect of a shared border n whch b = 1 f and j share a common border and zero otherwse. Substtutng these for T b,x equaton as follows: j (17) log S S j dk b. log n equaton (6), we formulate the regresson The error term has mplcatons for potental recprocty n geographc barrers,.e., the possblty that the dsturbance related to shpments from j to s postvely correlated to the dsturbance concernng shpments from to j. 8 Therefore, we estmate equaton (17) by generalzed least squares (GLS). In estmaton, we mpose S and S j j 0 wthout overall ntercept to avod the dummy varable 0 trap. Trade share and home share data are constructed from a detaled trade matrx database for twentythree countres (FAO, 2001). Among these countres, mports from the other twenty-two countres as a share of total mports are 57% on average. Table 2 summarzes the emprcal results estmated by generalzed least squares wth 515 observatons. In terms of fttng blateral trade flows, ths model performs well. For example, the model s adjusted R 2 s 0.96 and most of p-values of the estmated coeffcents are sgnfcant at the 1% level. The coeffcents n the upper porton of table 2 ndcate how dstance and border affect normalzed trade share. As wth the gravty lterature, dstance s an mpedment whle a common border serves as a catalyst for blateral trade flow. The estmates reported n table 2 are generally consstent wth those estmated by Eaton and Kortum (2002), Waugh (2010), and Remer and L (2010). However, our results ndcate that a dstance 15

17 of less than 375 mles has a relatvely small mpact on trade share. Trade share s affected when tradng partners are separated by more than 375 mles. For example, dstances less than 375 mle reduces trade share by 39% whle dstances between 375 and 750 mles reduces trade share by 312%. The negatve sgn and successvely larger magntudes on the dstance dummes suggest that freght costs and possbly other fxed components of transport costs (nsurance premum, holdng costs for agrcultural goods n transt, nventory cost due to bufferng the varablty of delvery dates, and preparaton costs assocated wth shpment sze and so on) may be a partcularly mportant mpedment to trade n global agrcultural markets. The border share dummy varable coeffcent s postve whch mples that f tradng partners share a common border, trade share ncreases by 8%. However, the estmated coeffcent s not sgnfcant at 10% level. For destnaton-country effects (representng gross openness to mports), our results show that the countres most open to mports are Canada, Australa, and the Unted States wth a respectve 62%, 31%, and 29% hgher trade share than average. Ths mples that fnal users of agrcultural goods n these countres lkely pay less than fnal users n the other countres. The countres least open to mports are Inda, Ethopa, and Chna wth a respectve 51%, 45%, and 40% lower trade share than average. Ths mples that fnal users n these countres lkely pay more than fnal users n the other countres. In order to llustrate the relatonshp between gross openness and economc development, we scatter estmates of destnatoncountry aganst per capta GDP. Fgure 2.1 shows a proportonal and strong relatonshp between two parameters. Also, the logarthm of was regressed on the logarthm of y. Here y s per capta GDP n The ntercept s and the slope coeffcent s Both are estmated at α = The regresson llustrates that the larger the per capta GDP, the larger the gross openness to mports, whch could reflect a varety of factors, ncludng lower tarffs and non-tarff barrers. In consderng source-country effects (representng gross compettveness for exports), the most effcent exportng countres are the Unted States and Brazl wth a respectve 184% and 122% hgher than average trade share. Ths mples that the agrcultural ndustry n these countres works more effcently as 16

18 compared to those of the other countres. The least effcent exportng countres are Ethopa, Morocco, and Bulgara wth a respectve 56%, 54%, and 54% lower trade share than average. Ths mples that the agrcultural ndustry n these countres works less effcently as compared to those of the other countres. The relatonshp between gross compettveness and economc development s presented n Fgure 2.2. Ths shows a proportonal but not strongly relatonshp between the two parameters. In addton, the logarthm j of s regressed on the logarthm of y. The ntercept s and the slope coeffcent s However, nether s sgnfcant at the 10% level. Ths result can be more specfed by solatng margnal cost and nfrastructural effcency parameters from the gross compettveness parameter. Ths ssue s dscussed n the followng secton. 5.1 Estmaton of Margnal Cost, Behnd and Beyond the Border Costs The regresson formulated by equaton (17) provdes us wth values of parameters j and. These estmated parameter values are put nto equatons (9), (10), and (11). Followng ths route, we successvely estmate unobservable margnal cost, behnd the border costs, and beyond the border costs. Table 3 shows estmated values of margnal cost by whch an nverse ndex (mc -1 ) can be obtaned to represent agrcultural compettveness. Lower margnal cost represents hgher agrcultural compettveness. For example, the agrcultural competveness of the Unted States s hgher than that of Japan because the margnal cost of the Unted States (0.55) s lower than that of Japan (1.37). The countres havng the lowest margnal costs are the Unted States and Brazl wth and 0.551, respectvely. The countres havng the hghest margnal costs are Morocco and Russa wth and 1.556, respectvely. Fgure 3.1 plots estmated agrcultural competveness (mc -1 ) aganst per capta GDP. A relatonshp between agrcultural competveness and economc development s dffcult to fnd. For example, the margnal cost of a hgh ncome country (Japan) s hgher than that of a low ncome country (Brazl). However, agrcultural competveness ncreased wth land sze. For example, the margnal cost of relatvely large countres such as the Unted States, Brazl, Argentna, Chna, and Inda s lower than those of relatvely small countres such as Japan, Korea, and Greece. In order to determne the relatonshp between agrcultural competveness 17

19 and economc development, the former varable s regressed on the latter. Ths regresson result dd not ndcate a sgnfcant relatonshp between the two varables. However, when agrcultural competveness was regressed on land sze, the results ndcate that land sze and agrcultural compettveness are postvely correlated. Ths mples that agrcultural compettveness s closely related to economes of scale. Table 3 shows estmated values of behnd the border cost by whch an nverse ndex (T (j,x)) -1 can be obtaned to represent nfrastructural compettveness for exportng agrcultural goods. Lower behnd the border costs represent more effcent nfrastructure for agrcultural exports whch wll promote agrcultural exports. For example, although margnal costs of the Unted States (0.550) and Brazl (0.551) are smlar, the average mport share of 22 countres from the Unted States (5.29) s hgher than that from Brazl (2.10). Ths may mply that nfrastructural compettveness of the Unted States s stronger than that of Brazl because behnd the border costs of the Unted States (0.640) are lower than those of Brazl (0.820). The countres havng the lowest behnd the border costs are the Unted States and Canada wth and 0.678, respectvely. The countres havng the largest behnd the border costs are Ethopa and Bulgara wth and 1.453, respectvely. Fgure 3.2 plots estmated nfrastructural competveness (T (j,x) -1 ) aganst per capta GDP. Hgh-ncome countres appear to have a more effcent nfrastructure for agrcultural exports whch could reflect a varety of factors, ncludng better developed transportaton and nsttutonal networks. In order to statstcally measure the relatonshp between nfrastructural competveness and economc development, the former varable s regressed on the latter. Ths regresson result shows that when ncome ncreased by 1%, nfrastructural competveness ncreased by 0.83% (a result that was statstcally sgnfcant at the 1% level). As shown n the prevous secton, gross compettveness exhbted no correlaton wth economc development. However, when agrcultural compettveness and nfrastructural effcency are separated from gross compettveness, nfrastructural effcency s closely related to economc development whle agrcultural compettveness s closely related to economes of scale,.e., land sze. Ths result, then, mples that economc development s more closely related to mprovement of market nfrastructure for agrcultural exports than to mprovement of agrcultural compettveness whch s related to economy of 18

20 scale. As a result, we can conclude that agrcultural exports n hgh ncome countres are promoted by better market nfrastructure. Table 3 also shows estmated values of beyond the border costs by whch an nverse ndex (T (,x)) - 1 can be obtaned to represent net openness to mports of agrcultural goods. Lower beyond the border costs represent more access of exporters to the mportng country. As a result, hgher net openness wll promote agrcultural mports. For example, although agrcultural competveness of the Unted States mc s hgher than that of Chna mc , the average mport share of the Unted States from 22 countres us (0.61) s hgher than that of Chna (0.17). Ths may mply that the mport barrers of Chna are more restrctve than those of the Unted States because the beyond the border costs of Chna (0.960) are hgher than those of the Unted States (0.409). The countres havng the lowest beyond the border costs are the Unted States and Canada wth and 0.460, respectvely. The countres havng the greatest beyond the border costs are Ethopa and Bulgara wth and 2.112, respectvely. Fgure 3.3 plots estmated net openness to mports (T (j,x) -1 ) aganst per capta GDP. Hgh-ncome countres appear to be more open to mports, whch could reflect a varety of factors ncludng lower tarffs and non-tarff barrers. In order to statstcally measure ther relatonshp, net openness was regressed on economc development. Ths regresson result shows that when ncome ncreases by 1%, net openness ncreases by 1.67% (a result that was statstcally sgnfcant at the 1% level). In ths study, we nclude four EU member countres such as France, Italy, Span, and Greece. Although EU member countres have common economc polcy, the behnd and beyond the border costs of Greece s dfferent from those of the other three member countres. As dscussed, behnd and beyond the border costs can be affected by not only polcy-nduced factors (such as tarffs and non-tarff barrers) but also market-nduced factors (such as marketng costs and retaler/wholesaler margns). Accordng to the results, Greece shows an neffcent market structure compared wth the other EU member countres because the behnd and beyond the border costs are very hgh compared wth those of France, Italy, and Span. us 19

21 Up to ths pont the analyss examned destnaton-country effects (gross openness), sourcecountry effects (gross compettveness), the margnal cost effect (agrcultural compettveness), behnd the border cost effects (nfrastructural effcency for agrcultural exports), and beyond the border cost effects (net openness to mports). Ths examnaton shows that economc development s related to reduced trade restrctons and mproved market nfrastructure whle agrcultural competveness s related to land sze. In the next secton, we conduct counterfactual analyss n two dfferent scenaros. In scenaro 1, beyond the border costs are reduced to the level of U.S. beyond the border costs n order to dentfy the effect of an ncrease n market access. Conversely, n scenaro 2, beyond the border costs are ncreased to the level of Ethopan beyond the border costs n order to dentfy the effect of a decrease n market access. 6. Counterfactual Analyss Trade costs have been shown to be the key component n explanng asymmetrc agrcultural trade flows. As dscussed n the prevous secton, counterfactuals are evaluated n terms of two crtera. One crteron utlzed s the ncrease n market access. Lower beyond the border costs postvely affect trade share n ndvdual countres, whch reflects a lower prce of mported agrcultural goods. The other crteron s a decrease n market access. Hgher beyond the border costs negatvely affect trade share n ndvdual countres whch, n turn, reflects a hgher prce of mported agrcultural goods. In counterfactual 1, a change n the trade share s calbrated to determne what happens f each country reduces ther beyond the border costs. In counterfactual 2, trade share s calbrated to determne what happens f each country ncreases ther beyond the border costs. 6.1 Lmtatons of the Model Before dscussng the results of counterfactual smulatons, the parameterzaton of counterfactuals smulaton should be descrbed n order to make t clear that the smulaton results represent trade share mplcaton of net openness to mports gven very restrctve condtons. For example, counterfactual smulatons are conducted as follows. Usng agrcultural trade data, we determned the ntal trade share, ˆ, and net trade parameter, ˆ. Also, trade cost,, as a regressor s obtaned by takng the second 20

22 hghest relatve prce of ndvdual agrcultural good between mportng country and exportng country j. As dscussed, j ˆ and ˆ are estmated by the regresson formulated n equaton (17). Then, the values of margnal cost, behnd the border costs, and beyond the border costs are calculated at ˆ and ˆ 0.5 usng equatons (9), (10), and (11), respectvely. 6.2 Major Results In counterfactual smulatons, a change n net openness has consecutve effects on gross openness and then trade share. For example, a change n beyond the border costs,x nfluences gross openness and trade share. In agrcultural trade, countres mplement trade restrctons most of whch are related to mport restrcton. Ths mport restrcton deters trade flow n nternatonal agrcultural markets. Gven current levels of mport restrctons, the average trade share of twenty-three countres used n ths study s 15.4%. The countres havng the lowest trade shares are Inda, Ethopa, and Chna wth 1.8%, 2.2%, and 3.6%, respectvely. The countres havng the greatest trade shares are Canada, Uruguay, and Mexco wth 51.4%, 33.7%, and 25.6%, respectvely. In ths counterfactual smulaton, the effects of reducton and ncrease of beyond the border cost on trade share are dentfed. In counterfactual 1, beyond the border costs of 22 countres are reduced to the lowest beyond the border cost of the Unted States to quantfy the effect of an ncrease n net openness. Table 4 presents the smulaton results. When beyond the border costs are decreased to the level of the Unted States, average trade share ncreases by 335%. As a result, the average trade share of 23 countres ncreases from 15.4% to 53.3%. Most of the ncrease n trade share appears n low ncome countres, whle the least ncreases occur n hgh ncome countres. Among the hgh ncome countres, the largest ncrease n trade share occurs n Greece, Korea, and Japan wth 436%, 368%, and 302%, respectvely, reflectng the fact that these countres currently mpose restrctve mport barrers. In counterfactual 2, beyond the border costs of 22 countres are ncreased to the hghest beyond the border cost of Ethopa to quantfy the effect of a decrease n net openness. Table 4 presents these smulaton T 21

23 results. When beyond the border costs are ncreased to the level of Ethopa, average trade share decreases by 63%. As a result, the average trade share of the 23 countres decreases from 15.4% to 4.7%. Ths magntude of decrease s relatvely small as compared to the results of counterfactual 1. Ths mples that current levels of beyond the border costs are hgh, whch reflects a currently hgh level of mport restrctons. 7. Conclusons The Rcardan model mples that countres wll tend to produce and export goods n whch they have comparatve advantage (Feenstra, 2003). If every country specalzes n ths manner, the trade share of each tradng country would ncrease to the pont where ts welfare would be maxmzed (Jabara and Thompson, 1980). However, ths mplcaton apples only to a purely ntegrated market n whch there are no trade costs. Anderson and Wncoop (2004), however, ndcate that nternatonal trade costs are qute large and hghly varable across goods and countres. In partcular, trade costs of agrcultural goods are sgnfcantly hgher than those of other commodtes (Khan and Kalrajan, 2011; Kalrajan, 2007; Brenton et al, 2001; and Messerln and Zarrouk, 2000). These hgher trade costs dstort comparatve advantage n the producton and export of agrcultural goods. Accordng to ths study, less developed countres, on average, have sgnfcantly larger trade costs than do more developed countres, whch may explan why the majorty of agrcultural trade occurs between hgh ncome countres. As noted, trade costs comprse a large share of the prce pad by a foregn end user. Trade costs, unfortunately, cannot be observed drectly from trade data. However, agrcultural trade costs can be derved from an economc model lnkng trade flows to observable and unobservable varables. Inference has manly used the gravty approach whch does well n representng the mpacts of economc development (often usng GDP as a prncpal measurement) and dstance between two economc unts n blateral trade flows. However, symmetrc restrctons on trade costs n the standard gravty approach have provoked controversy amongst nternatonal economsts because emprcal examples conflct wth the symmetry assumpton (Balster and Hllberry, 2007). 22

24 Recent studes show that blateral trade can be nfluenced by asymmetrc trade costs. One example s Eaton and Kortum (2001 and 2002). However, ther study dd not separate the effects of margnal cost, behnd the border costs, and beyond the border costs on trade flows. In ths study, we sought to dentfy these effects of margnal cost, behnd the border costs, and beyond the border costs on trade flows. For example, agrcultural compettveness (ndcated by an nverse ndex for margnal cost) s more dependent on land sze and less dependent on economc development. However, nfrastructural competveness (ndcated by an nverse ndex for behnd the border costs) and net openness (ndcated by an nverse ndex for beyond the border costs) are dependent on economc development. Ths mples that most agrcultural trade nduced by hgh ncome countres s closely related to the development of market nfrastructure and the reducton of trade barrers whle agrcultural productvty s closely related to economes of scale. As ndcated n the prevous secton, ths study ndcates that beyond the border costs are a key factor affectng trade flows. Based on ths fndng, two counterfactual smulatons are analyzed to quantfy the effects on trade flow of net openness to mports. The two counterfactual smulatons ndcate that the current level of mport restrctons s restrctve. Fnally, ths model may be used as a novel approach to analyze ndvdual agrcultural goods wth detaled market nformaton n order to dentfy more useful market nformaton. For example, agrcultural trade cost s hghly lnked wth geographcal dstance and mperfect market ntegraton. Techncal trade costs (transportaton cost, storage cost n transt, and so on) are ncurred mostly by the former and nsttutonal and poltcal trade costs are generated manly from the latter. Insttutonal trade costs can be dvded nto drect nsttutonal trade costs (tarffs, quotas, and trade costs assocated wth the exchange rate system) and ndrect nsttutonal trade costs (transport nfrastructure, law enforcement, related property rghts nsttutons, and nformatonal nsttutons). The poltcal trade costs are often ncurred from mport bans and embargos based on natonal nterests. These trade costs of ndvdual agrcultural good can be quanttatvely measured by usng our model supplemented wth specfc market nformaton. Therefore, research can be conducted for a varety of agrcultural goods 23

25 to dentfy the composton of these asymmetrc trade costs and ther mplcatons for nternatonal agrcultural markets. 24

26 Table 1. Agrcultural Trade Share,, n Percent for 23 Countres - FAO, 2001 Hgh Income Countres Mddle Income Countres Low Income Country C01 C02 C03 C04 C05 C06 C07 C08 C09 C10 C11 C12 C13 C14 C15 C16 C17 C18 C19 C20 C21 C22 C23 C01 U.S C02 Japan * 0.00* * 0.00* * 0.01 C03 Canada C04 France C05 Italy C06 Australa C07 Span * C08 Greece * C09 Korea * 0.00* * 0.00* * 0.00* * 0.01 C10 Argentna C11 Mexco C12 Hungary * * 0.00* 0.00* * 0.00* 0.00 C13 Uruguay * * * * * 0.00* * 0.00 C14 Brazl C15 Peru * * 0.00* 0.00* 0.00* 0.00* 0.00* 0.00* 0.00 C16 South Afrca * C17 Russa * 0.00* * 0.00* 0.00* 0.00* * 0.00* C18 Turkey C19 Bulgara * * * 0.00* 0.00* 0.00* * 0.00* 0.00 C20 Morocco * * 0.00* * 0.00* * * 0.00* * 0.00* 0.00 C21 Chna C22 Inda * C23 Ethpa * * 0.00* * 0.00* 0.00* 0.00* * Entry n column, row j, s the fracton of agrcultural goods country mports from country j.. Zeroes wth stars ndcate the value s less than Zeroes wthout stars are zeroes n the data. 25

27 Table 2. Geographc and Country-Specfc Etmates and Ther % Effects on Trade Share Observatons Used TSS SSR MSE Geographc barrers Estmated Parameter p -value % Effect d 1 [0, 375) < d 2 [375, 750) < d 3 [750, 1500) < d 4 [1500, 3000) < d 5 [3000, 6000) < d 6 [6000, maxum) < b Shared border Destnaton Country Source Country Country δ p -value % Effect δ j p -value % Effect Unted States 0.99 < < Japan 0.83 < < Canada 2.10 < < France 0.70 < < Italy 0.89 < < Australa 1.08 < < Span 0.73 < < Greece Korea < Argentna < Mexco Hungary < Uruguay < Brazl < Peru < < South Afrca Russa < Turkey Bulgara < < Morocco < Chna < < Inda < Ethopa < < Note 1: For estmated d, the mpled % effect on trade cost s 100 [Exp(-θ d )-1)] wth θ = Note 2: TSS s total sum of squares, SSR s resdual sum of squares, and MSE s mean square error Note 3: Estmated by GLS wth 515 observatons, Adjusted R 2 s

28 Table 3. Margnal Cost, Behnd and Beyond the Border Cost at θ = and ρ = 0.5 mc T (j,x) T (,x) Unted States Japan Canada France Italy Australa Span Greece Korea Argentna Mexco Hungary Uruguay Brazl Peru South Afrca Russa Turkey Bulgara Morocco Chna Inda Ethpa

29 Table 4. Counterfactuals 1 and 2: Decrease and Increase n Beyond the Border Cost Benchmark Counterfactual 1 Counterfactual 2 0 / 0 1 / 1 Δ( 1 / 1 ) 2 / 2 Δ( 2 / 2 ) U.S Japan Canada France Italy Australa Span Greece Korea Argentna Mexco Hungary Uruguay Brazl Peru S.Afrca Russa Turkey Bulgara Morocco Chna Inda Ethpa Average Note 1. In Counterfactual 1, all countres' beyond the border costs are reduced to Note 2. In Counterfactual 2, all countres' beyond the border costs are ncreased to

30 Prce Hgh Mddle Low Economc Development Fgure 1. Agrcultural Prce and Economc Development 29

31 Gross openess to mports (S ) Canada Australa U.S. Chna Ethopa Inda Per captal GDP, 2001, natural log Fgure 2.1 Gross openess and Per captal GDP 30

32 Gross competveness for exports (S j ) Ethopa Morocco Bulgara Brazl U.S Per capta GDP, 2001, natural log Fgure 2.2 Gross competveness and Per capta GDP 31

33 Agrcultural competveness (mc 1 ) Ethopa Morocco Bulgara Brazl U.S Per capta GDP, 2001, natural log Fgure 3.1 Agrcultural Compettveness and Per capata GDP 32

34 2.0 Infrastructural competveness T (j,x) Ethopa Bulgara U.S. Canada Per Capta GDP, 2001, natural log Fgure 3.2 Infrastructural Compettveness and Per capta GDP 33

35 Ne openess to mports T (,x) U.S. Canada Ethopa Per capta GDP, 2001, natural log Fgure 3.3 Net Openess and Per capta GDP 34

36 Mexco 0.30 Italy 0.25 Japan France Span 0.20 Korea Turkey Russa Hungary 0.15 U.S. Morocco Australa Argentna Greece Peru 0.10 Brazl S.Affrca Bulgara 0.05 Chna Inda Ethopa Fgure 4.1 Relatonshp between Trade Share ( / ) and Beyond the Border Cost (T (,x)) 35