Market Power in International Commodity Processing Chains: Preliminary Results from the Coffee Market

Transcription

1 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Ben Sheherd March 2004 The vies exressed in his aer are hose of he auhor. They do no necessarily reflec hose of he Groue d Economie Mondiale de Sciences Po.

2 MARKET POWER IN INTERNATIONAL COMMODITY PROCESSING CHAINS: PRELIMINARY RESULTS FROM THE COFFEE MARKET Ben Sheherd * Groue d Economie Mondiale (GEM) Insiu d Eudes Poliiques de Paris 2, Square de Luynes Paris, France This Draf Daed: 10 March 2004 Absrac: Vecor auoregressions are used o model rice ransmission hrough he coffee rocessing chain, from roducers o he orld marke and from he orld marke o consumers. A comarison is made of rice dynamics agains a backdro of o very differen marke srucures: re1989, roducers exered marke oer hrough exor quoas and saeconrolled markeing channels; os1989, governmen inervenions are minimal, bu rivae acors a inermediary, rocessing and reailing levels have become quie concenraed. Ineresingly, he analysis shos ha liberalisaion has no imroved rice ransmission as significanly as execed and in some resecs aears o have orsened i noiceably. One ossible exlanaion is marke oer amongs rivae acors a inermediae levels in he rocessing chain. JEL Codes: C32; L11; L66; Q13; Q17. Keyords: Vecor Auoregression; Marke Poer; Inernaional Trade; Commodiy Marke Liberalisaion; Coffee. * Encouragemen and helful commens on revious drafs are graefully acknoledged from Profs. Richard Baldin & John Cuddy (Graduae Insiue of Inernaional Sudies, Geneva), Mr. Parick Lo (WTO) and Prof. Parick Messerlin (SciencesPo). Remaining errors are my on resonsibiliy. Conac: benamin.sheherd@scienceso.org.

3 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Marin Luher used o onder ha eole acually do in heaven. For mos aricians in he inensely comeiive food manufacuring indusry, conemlaion of Neslé s soluble coffee business mus seem like he commercial equivalen of Luher s siriual mediaion. (Oxfam (2002) ciing a Deusche Bank reor) 1. Inroducion The coninuing srong erformance of Neslé and oher gians of he rocessed beverage orld is in sriking conras ih he aarenly ever increasing imoverishmen of ordinary coffee farmers a a ime of hisorically lo green coffee rices; he oin is ell encasulaed in Oxfam s (2002) image of enniless farmers versus rofieering roasers. While he bulk of recen ork suggess ha he sources of he curren crisis are essenially srucural (see e.g., Varangis e al. (2003) and ICO (2002, 2003d)), here is noneheless an undercurren of susicion born of ha conras: is here somehing secial abou he srucure of he coffee rocessing chain ha could be a conribuing cause o his aaren decouling of forunes? The resen sudy is an aem o bring some analyical rigour o bear on ha quesion, hich is deal ih only arially and inadequaely by he very small body of exising lieraure. My saring oin is ha marke liberalisaion in he lae 1980s and early 1990s as a defining even in erms of marke srucure: rior o mid1989, roducer counries generally used exor quoas and saeconrolled markeing channels in an aem o exercise marke oer and kee orld rices high. 1 In July 1989, he firs elemen of ha aroach he economic clauses of he Inernaional Coffee Agreemen (ICA) effecively broke don and has no been relaced in any serious ay since (see Akiyama (2001) and Gilber (1996)). In he years folloing he ICA s collase, mos roducer counries also dismanled heir cenralised markeing sysems, meaning ha ha had once been a highly regulaed roducer s marke is no a relaively free marke in hich he vas maoriy of aciviy is lef o he rivae 1 In erms of he roducer counries examined here, Guaemala is he only exceion o he rule: i has alays had a free coffee markeing sysem and rior o July 1989 as consrained only by he ICA s exor quoas. Ben Sheherd Page 2 10/3/2004

4 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke secor (for an overvie, see Akiyama (2001) and ITC (2002)). 2 Wih hese oins in mind, he research quesion can no be reosed: has realisaion of he execed benefis of liberalisaion been hamered by he aciviies of rivae agens large enough o influence marke oucomes? I call his he rofieering roasers hyohesis. 3 Afer revieing some basic informaion on firm concenraion and assessing he adequacy of he exising lieraure in Secion 2, I resen he daase and model o be used in Secion 3. In a nushell, vecor auoregressions (VARs) are used o analyse he dynamics of rice ransmission hrough he coffee rocessing chain in maor roducer and consumer counries, boh rior o and folloing liberalisaion. Cerain characerisics of rice ransmission ha ould be execed o change folloing liberalisaion seed and comleeness, direcion, symmery and rice differenials are examined by esing hyohesised model resricions ha reflec he characerisics in quesion, as ell as hrough he analysis of imulse resonse funcions (IRFs). Secion 4 resens he resuls obained using his aroach, hile Secion 5 concludes ih some elemens of an agenda for furher research. 2. Firm Concenraion in he Coffee Processing Chain The issue of heher firms a inermediae levels in he coffee chain are large enough o have a significan influence on marke rices and quaniies and if so, heher hey in fac use such marke oer o he derimen of consumers and/or roducers has assumed an imoran lace in some of he lieraure on he curren coffee crisis (see e.g., Oxfam (2002), Pone (2002) and Talbo (2002)) and, indeed, has echoes in a leas one governmen osiion aer lodged ih he WTO (Kenya e al, 2003). I is cerainly ossible o marshal a 2 For resen uroses, only Colombia is exceional in his regard: he Naional Federaion of Coffee Groers (Federacafé) sill accouns for around 40% of oal exors. 3 Of course, he quesion oenially alies much more idely han us o he coffee marke: see e.g., Murhy (1999, 2002) & Morisse (1997, 1998). Ben Sheherd Page 3 10/3/2004

5 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke resecable number of basic saisics in suor of he vie ha some inermediae agens are big enough o make such rice maniulaion lausible: In 2000, he o en roasers had a combined share of 63% of he global rocessed coffee marke (ITC, 2002). The o five roasers accoun for 80% of he marke in he USA and 84% in Germany, figures ha are by no means exceional amongs imoring counries (ITC, 2002). The Briish comeiion achdog found ha Neslé had a 56% share of he naional marke and a reurn on caial emloyed in is soluble coffee business of over 100% in 1989 (MMC, 1991). In he end, hoever, i concluded ha his as no agains he ublic ineres. In Colombia, five comanies accoun for around 70% of all rivae secor exors. (See ITC, 2002; LMC Inernaional Ld, 2000b; and USDA, 1999.) In Uganda, hreequarers of all ne enrans en bankru ihin o years of liberalisaion, leaving he o en comanies ih 80% of he oal exor marke (Akiyama e al., 2001). In Guaemala, around 100 exorers are acive in he marke, bu he o five have a combined marke share of aroximaely 50%. Six comanies are linked o mulinaionals and ogeher accoun for around 35% of he exor marke (see ICO, 2002; LMC Inernaional Ld, 2000c; and Varangis e al, 2003). In Mexico, around 200 comanies are acive in he exor marke, ih he o 15 having a combined marke share of 67.5% in 1997 (see ITC, 2002; ITF, undaed; and USDA, 1995). Ben Sheherd Page 4 10/3/2004

6 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke On he oher hand, here are over 200 regisered exorers in Brazil and no single firm has a marke share greaer han 10%. Similarly, over 100 exorers are acive in he Indian marke (see Akiyama, 2001; ITC, 2002; and LMC Inernaional Ld, 2000a). These basic saisics sugges ha concenraion is a serious henomenon in mos, bu no all, roducer and consumer markes. Alhough a deailed exloraion of is causes is no ossible here, he exising lieraure idenifies a number of facors ha migh be a ork: srong roduc differeniaion in consumer markes folloing he lae revoluion (Fier & Kalinsky (2001) and Pone (2002)), imoran scale economies in roducion (ITC (2002) and UNCTAD (1999)), he difficulies of ne firms in accessing finance (Akiyama e al. (2001)) and he imorance of disribuion neorks (cf. Dolan & Humhrey (2000) in he vegeable markeing conex) Previous Quaniaive Work Hoever, he fac ha here is some basic saisical evidence suoring he lausibiliy of he marke oer sory does no mean ha he rofieering roasers hyohesis is necessarily rue. More formal analysis is required, in aricular focusing on he dynamics of rice formaion and ransmission, hich can be seen as one of he rincial exernal exressions of marke oer. Hoever, i is us his ye of research ha has, o dae, lagged behind (see WTO (2003) for a selecive overvie). In aricular, he siuaion in roducer counries hich is he main moivaing force behind Oxfam (2002) and similar NGO lieraure has been largely ignored. A rare exceion is Moss & Guerra Galindo (2001), 4 in hich he auhors invesigaed marke oer amongs Mexican rocessors. In a sandard regression of he orldroducer rice 4 To addiional aers should be noed in assing. WinerNelson & Temu (2002) used survey daa from he Tanzanian marke o conclude ha hile he markeing margin for coffee seemed o have fallen folloing Ben Sheherd Page 5 10/3/2004

7 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke sread, neiher rocessor concenraion nor oal exors ere saisically significan, leading he auhors o conclude ha he evidence did no indicae eiher marke oer amongs rocessors or Mexico s having he abiliy o affec orld marke rices. Alhough heir inerreaion of hese resuls seems correc in as far as i goes, i is unforunaely difficul o ell heher he mehodology adoed as aroriae or no, as basic diagnosic saisics and es resuls are no rovided. The daase is also roblemaic, as he concenraion measure used in fac changed very lile during he samle eriod ( ), so is aaren lack of exlanaory oer is no surrising. Aar from ha isolaed examle, mos oher quaniaive sudies focus on consumer markes. Beendorf & Verboven (1997, 1998) and Koerner (2002a, 2002b) looked a he Duch, US and German markes, using reasonably similar srucural modelling echniques. In all cases, numerous a riori resricions ere imosed in erms of assumed funcional forms and cos srucures. Evidence as found of oligoolisic ineracions in all hree markes, hough he elfare imlicaions aeared more serious in he US han in Germany or he Neherlands. Indeed, Koerner (2002a, 2002b) suggesed ha a rice ar as in fac underay in Germany, ih ricing belo marginal cos. Feuersein (2002) ook a fundamenally differen aroach, using a vecor errorcorrecion model (VECM) o invesigae he relaionshi beeen green coffee rices and reail rices in Germany. She found ha changes in he former ere fully ransmied o he laer in he long run, bu ha adusmen as relaively slo. She also found evidence of asymmeric ransmission of changes in green coffee rices. liberalisaion, he gains o farmers ere offse o some exen hough usually no fully by increased ransacion coss in inu markes. Loez & You (1993) assumed he exisence of oligosony oer amongs coffee exorers in Haii and roceeded o invesigae he facors deermining i. Ben Sheherd Page 6 10/3/2004

8 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Gómez & Casillo (2001) 5 used he BoxJenkins mehodology and a VECM o analyse he US marke. Their orking hyohesis as ha liberalisaion of he orld coffee rade in he 1990s did no resul in an undisored marke, bu raher in a ransfer of marke oer from roducer counries o inernaional holesalers. They found ha he ga beeen orld and US coffee rices had idened folloing liberalisaion (aken as July 1989); hoever, heir resuls mus be reaed ih a cerain sceicism, as he reliberalisaion orldreail rice sread as in fac found o be negaive, hich seems imlausible as a longrun coinegraing relaionshi. Finally, hey concluded on he basis of eak exogeneiy ess ha in he eriod rior o liberalisaion, orld marke rices ended o be se indeendenly ih subsequen adusmen by reail rices, hereas he reverse as rue folloing liberalisaion. This as argued o be consisen ih he acquisiion of marke oer by inernaional holesalers a he exense of roducers. A comaraive analysis of he US, French and German markes as underaken by Gómez & Koerner (2002), again using ime series raher han srucural echniques. Evidence of shor erm asymmeric rice ransmission as found for all hree markes, using an asymmeric VECM. Hoever, differences in erms of marke srucure ere also aaren among he hree counries, ih faser rice adusmen observed for Germany han for eiher France or he USA Consolidaion and Moivaion of he Presen Sudy As can be seen from his brief revie, he exising quaniaive lieraure on he srucure of he coffee marke is achy and does no rovide srong evidence one ay or he oher in erms of he rofieering roasers hyohesis. This is due rimarily o he fac ha exising sudies focus only on he links in he rocessing chain ha mediae beeen he orld marke 5 Sincere hanks o Dr. Miguel Gómez boh for sulying a coy of he aer and encouraging my on research. Ben Sheherd Page 7 10/3/2004

9 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke and he consumer; ineracions beeen he roducer and he orld marke have received scan aenion and here has been no aem o include all hree levels in a single analysis. The resen sudy is an aem o fill ha ga in he lieraure and o rovide a more comlee icure of he underlying marke srucure han exiss a resen. The focus ill be exclusively on he ay in hich informaion on he dynamics of rice ransmission can be used o dra inferences regarding marke srucure. A number of differen ess (see belo) ill be combined, so as o srenghen he resuling inferences. Time series mehodology ill be used in reference o a more radiional, srucural aroach. This is because he former requires only an absolue minimum in erms of a riori assumions and noably does no involve he imosiion of aricular funcional forms designed o model roducer and consumer behaviour. This is an imoran benefi hen analysing marke srucure, as i avoids he rincial shorcoming of srucural models, namely ha a single hyohesis regarding marke srucure canno be esed indeendenly of he hyohesis ha roducer and consumer behaviour are accuraely caured by he funcional forms used in he model (cf. Sexon & Lavoie (2001)). 3. Overvie of he Daa, Model and Tesing Sraegy No ha some benchmarks regarding he aims of his research have been se, his secion inroduces in more deail he mehodology o be adoed. The daase is revieed and he resuls of reesing are summarised. The modelling aroach is hen discussed, including roblems of esimaion and hyohesis esing The Daase I use an Inernaional Coffee Organisaion (ICO) daase of monhly rices running from January 1982 o December All are exressed in nominal US dollars and have been colleced and/or calculaed by he ICO. The daa are sli ino o suberiods, denoed reliberalisaion and osliberalisaion resecively. In he absence of any addiional Ben Sheherd Page 8 10/3/2004

10 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke consideraions, July 1989 is used as he breakoin. For models involving cerain roducer counries, hoever, a laer breakoin is used if usified by he differen ace of domesic reform in ha counry, in hich case a dummy variable is included o searae ou domesic and inernaional liberalisaion effecs. 6 World rices (Figure 1) are roxied using he ICO s Indicaor Prices: Colombian Mild Arabicas (CMA), Oher Mild Arabicas (OMA), Robusas (RG) and Brazilian Naurals (BNG). Each indicaor rice is a eighed average of exdock rices for green coffee in he maor inernaional markes (he USA, France and Germany). 7 In addiion, he ICO also mainains a Comosie Index (CI), hich is iself a eighed average of revailing rices for hose four coffee yes (ICO, 2003b). The ICO mainains roducer rice daa (Figure 2) for all exoring member counries and reail rices (Figure 3) for all imoring members, bu o faciliae he analysis, i as decided o use only a subse of ha daa. A selecion as iniially made of he en larges exoring counries based on 2001 exor volumes. Limied daa availabiliy or concerns as o is reliabiliy in resec of four of hose counries Côe d Ivoire, Indonesia, Peru and Vienam led o heir exclusion from he samle. The analysis as herefore underaken using he six remaining counries (2001 exor ranking in brackes): Brazil (1), Colombia (3), Guaemala (6), India (7), Mexico (8) and Uganda (9), hich ogeher accouned for over 50% of ICO members exors in For similar reasons, only he USA and Germany (45% of 2001 imors) ere considered on he consumer side. 6 March 1990 coinciding ih full liberalisaion of he domesic marke is hus used as he breakoin for Brazil. Seember 1996 is used for India, January 1993 for Mexico and January 1996 for Uganda. 7 ICO (2001, 2003a & 2003b) se ou he saisical definiions and rocedures alied by he ICO in calculaing is indicaor rices. Hoever, no recise exlanaion of he erm exdock is given. General rade usage suggess ha exdock rices include ransor coss and rade measures such as ariffs imosed by he imoring counry. A search of he TRAINS daabase mainained by UNCTAD revealed ha boh he USA and he EU have bound duyfree access for green coffee, meaning ha e need no ake any furher accoun of he oenial effecs of rade measures on he ICO s indicaor rices. Ben Sheherd Page 9 10/3/2004

11 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Preliminary Tess As summarised in Table 1, revious research has ended o rea coffee rices a leas a he orld and consumer levels as inegraed of order one (, or difference saionary). Furher reesing is usified in he resen case for hree reasons. Firsly, roducer rices have no been considered a all in revious ork, hile orld and reail rices have been considered only arially. Secondly, samle eriods and frequencies for hose series hich have been used reviously are ofen differen from hose used here, and saisical roeries migh herefore differ. Thirdly, insufficien aenion has been aid o he ossibiliy of a srucural break in he daa coinciding ih marke liberalisaion. Only Gómez & Koerner (2002) esed for he exisence of such a break (and found i), hough Gómez & Casillo (2001) also uncovered evidence of an imoran change in he daa generaing rocesses coinciden ih liberalisaion. The oin is imoran, as convenional saionariy ess can ofen misake a srucural break for a uni roo (Perron, 1989), oenially leading o model missecificaion. All series have herefore been reesed, using hree common mehodologies (ADF, KPSS and Perron) and considering boh he full samle eriod and he o searae suberiods. 8 Resuls (based on ess a he 10% level of significance) are resened in summary form in Tables 24 and exhibi considerable differences from revious ork. 9 Firsly, here is lile uniformiy amongs he hree ses of series in erms of saionariy. Raher, he daa aear o include a mixure of saionary and nonsaionary rocesses, suggesing ha he modelling mehodology ill need o be flexible enough o deal ih his imoran fac. 8 Boh he Perron (1989) and ADF aroaches (see Greene, 2000; Hamilon, 1994) es he null hyohesis of a uni roo, he rincial difference beeen he o being ha he former exlicily akes ino accoun he ossibiliy of a srucural break and is herefore aricularly ell adaed o use here. KPSS (Kiakoski e al., 1992), on he oher hand, ess he null hyohesis of saionariy. 9 Full resuls are available from he auhor on reques. Ben Sheherd Page 10 10/3/2004

12 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke The srucural break issue is aarenly crucial, noably because he daa someimes aear o be saionary in he re1989 suberiod considered on is on, hile in he os1989 suberiod hey ofen (hough no alays) aear o be. I ould herefore seem quesionable o rely solely on resuls for he full samle eriod, hich seem o sho ha he series are all. Moreover, hese resuls ould seem o add eigh o he argumens in favour of a modelling sraegy ha slis he samle ino o suberiods, and hich herefore leaves room for differen saionariy characerisics o be a ork rior o and folloing liberalisaion The Basic Model My modelling aroach dras heavily on he ime series models discussed in Secion 2, as ell as on he groing quaniaive lieraure analysing rice ransmission in verically inegraed suly chains (e.g., Chavas & Meha (2002), Harmann e al. (undaed), Guilloreau e al. (2003), Zachariasse & Bune (2003)) and sudies of marke inegraion (e.g., Baffes & Aad (1998), Bukenya & Labys (2002) and Sanuán & Gil (undaed)). I ado ha migh be ermed an eclecic aroach o examining marke oer: raher han esimaing a model designed o es for one or anoher indicaor, I insead consruc a more general saisical reresenaion of he daa, hich is hen used o es for a number of differen indicaors, all on he basis of a single saisical model. 10 Raher han he VECM aroach alied by a number of oher researchers, simle VARs in levels and firs differences ill be used in his case, aking he general form of sysems (1) and (2) and esimaed on he basis of he full samle (ih dummy variables) or one or oher subsamle, deending on he es being erformed. This choice as made for o reasons. 10 This aroach oes a grea deal o Sims (1980) and Johansen (1995) as ell as o a draf manuscri eniled The Coinegraed VAR Model: Economeric Mehodology and Macroeconomic Alicaions, kindly made available by Prof. Kaarina Juselius on her ebsie. Ben Sheherd Page 11 10/3/2004

13 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Ben Sheherd Page 12 10/3/2004 Firsly, he saionariy roeries of he daa series are neiher easy o idenify ih ceriude nor uniform over ime and sace, so here are oenial benefis o adoing a modelling sraegy ha is no deenden on he sric assumion ha all series are (or saionary) a all oins in ime. Secondly, and relaed o he firs oin, i as found ha sable coinegraing relaionshis he basis of VECMs ere difficul o esablish ih confidence over he full ime eriod under consideraion. Comarison of re and osliberalisaion resuls as aricularly roblemaic and made he use of more general models all he more aracive. 11 Φ Φ k k k k DU DU ε µ ε µ (1) Φ Φ k k k k DU DU ε µ ε µ (2) The variables and refer o orld and roducer rices resecively, 12 and ε is a sochasic i.i.d. (0, Ω) error erm. In models here reail rices and no roducer rices are used, r is subsiued for. The model is adused accordingly hen more han one orld rice series is included. All rice series are reaed as endogenous a maor dearure from he srucural radiion, in hich assumions are generally made regarding exogeneiy hile he deerminisic erms in sysem (1) are limied o a consan and an inervenion dummy se o zero rior o July 1989 and uniy elsehere; in sysem (2), aking firs differences of he 11 Full coinegraion and VECM resuls are available from he auhor uon reques. 12 For roducer counries, he orld rice is used hich mos closely mirrors he ye of coffee roduced and here more o yes are roduced, o orld rice series are used. For consumer counries, CI is used.

14 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke inervenion dummy resuls in an imulse dummy ha is zero everyhere exce July Obviously, he dummy is deleed from all osliberalisaion models as ell as hose reliberalisaion models for hich he dae used as he breakoin beeen he re and osliberalisaion subsamles is in fac July As noaion becomes comlicaed hen more han one orld rice series is involved, he folloing convenion is adoed: for examle, RG BNG 3 refers o he hird lagged erm of he RG series in he BNG equaion Making he Basic Model Talk Wihou more, models in he form of sysems (1) and (2) are no erribly informaive. The key is herefore o ake hem for ha hey are: nohing more han convenien mahemaical reresenaions of rices a o differen marke levels, hich are believed a riori o be linked in some ay, even hough he recise naure of he link is unknon. Wha is imoran in erms of economic insigh is o use hese reresenaions o es hyoheses ha are economically meaningful. The nex secion resens he recise ess ha are used, bu i is useful a his oin o inroduce informally he asecs of he linkages beeen he o rice series i is roosed o invesigae: 1. Seed and comleeness of rice ransmission: The move from saedominaed markeing channels and exor quoas in roducing counries o a relaively free marke sysem is execed o make rice ransmission faser and more comlee in boh direcions. On he oher hand, consumer markes can be considered as liberal in boh eriods and here should herefore no be any significan change in he seed and comleeness of rice ransmission folloing liberalisaion. 2. Direcion of rice ransmission: Prior o liberalisaion, roducer counries made a concered effor boh o exercise marke oer (by using exor quoas o resric suly and kee rices u) and o insulae farmers from exernal (orld marke) rice shocks. In erms of he coffee rocessing chain, rice ransmission ould herefore be Ben Sheherd Page 13 10/3/2004

15 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke execed o be boom u a firs, in he sense ha changes in farm gae rices ranslae ino changes in orld marke rices and reail rices, bu he reverse is no necessarily rue (aricularly regarding he roducerorld marke inerface). Liberalisaion ould be execed o change his dynamic, by romoing rice ransmission simulaneously in boh direcions (boom u and o don). 3. Symmery of rice ransmission: Price Transmission Asymmery (PTA) migh be execed a he orldroducer rice inerface rior o liberalisaion, in line ih he suly and rice managemen olicies folloed in mos roducer counries. Hoever, his ould be execed o change folloing liberalisaion and here is no reason no o exec a move oards more symmeric rice ransmission. Similarly, rice ransmission from he orld marke o consumers ould be execed o be reasonably symmeric in boh suberiods. 4. Changes in markeing margins: One of he avoed goals of marke liberalisaion as o reduce he markeing margin facing coffee roducers (e.g., Akiyama e al. (2001)). I is herefore execed ha he ga beeen orld and roducer rices should be smaller folloing liberalisaion han before i. On he oher hand, no aricular change ould be execed in he ga beeen orld and reail rices. I can readily be seen ha oucomes conrary o execaions in each of hese areas could be consisen ih an exlanaion in erms of marke srucure, namely ha increased concenraion a inermediae levels in he rocessing chain has revened liberalisaion from having is inended effecs (cf. Gómez & Casillo, 2001). While i could be argued ha one such oucome is no necessarily a smoking gun in erms of he rofieering roasers hyohesis, such an obecion ould lose some of is oency if, for examle, all four oucomes ere o be conrary o execaions: he analogy ould hen be closer o ha of observing a loud bang, he smell of gun oder, fingerrins on a gun and a dead body. Tha Ben Sheherd Page 14 10/3/2004

16 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke is, he use of a number of indicaors rovides a sronger circumsanial basis on hich o dra inferences abou he underlying henomena in quesion. 4. Overvie of Resuls This secion resens a summary of he maor resuls obained ih he aroach oulined above. Given he comaraively large number of models involved a oal of 56 including simle and lagaugmened VARs in levels and firs differences, covering he full eriod or one or oher suberiod i is no ossible o resen deailed informaion on each model, so he emhasis here is on draing ogeher hose resuls and deriving imlicaions from hem Esimaion and he Uni Roo Problem Afer firs deermining k (he number of lags of each endogenous variable o be included in he model), sandard OLS regression as used o esimae he various models in he form of sysems (1) and (2). 14 Boh reesing (see above) and insecion of characerisic roos (Johansen, 1995) sugges ha a number of he simle VARs in levels conain uni roos or near uni roos, meaning ha hose models as hey sand canno be relied on for hyohesis esing, due o he robable nonsandard disribuion of common es saisics. 15 The soluion o his roblem is a ragmaic one: hile VARs in levels are used as he saring oin for analysis, VARs in firs differences and lagaugmened VARs in levels (Dolado & Lükeohl, 1994) are used o buress resuls as necessary. This should overcome he difficulies arising from nonsaionariy, as all comonens in firsdifferenced VARs aear saionary, hile 13 All calculaions ere erformed using EVies versions 3.1 and 4.1. Full model resuls including esimaed arameers, diagnosic ess, resuls of hyohesis ess and IRF los are conained in a deailed saisical aendix, available from he auhor uon reques. 14 A number of mehods are available for deermining k (see Lükeohl & Breiung (1996)). Here, i as chosen using sandard model selecion crieria and lag exclusion Wald ess as an iniial guide, bu he final decision as based on a ragmaic radeoff beeen he need for aroximae hie noise residuals and he imorance of conserving degrees of freedom. Diagnosic ess sugges ha he models hus consruced are olerably ellsecified. 15 The modulus of he larges characerisic roo in mos of he VARs in levels is around 0.9. Ben Sheherd Page 15 10/3/2004

17 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Dolado & Lükeohl (1994) have shon ha addiion of an exra lag o a nonsaionary VAR can allo consisen esing of simle hyoheses, such as Granger causaion, rovided ha he exra lag is lef unconsrained (see also Toda & Yamamoo (1995) and Yamada & Toda (1998)) Model Diagnosics Models in levels aear a firs glance o erform very ell, ih adused R 2 coefficiens for individual equaions in excess of 0.8 in almos all cases. Models in firs differences, as ould be execed, exhibi markedly loer adused R 2, more in he region of 0.1 o 0.2. In any case, R 2 should no be overemhasised as i could be inflaed in he models in levels due o nonsaionariy of he underlying series. More general diagnosic ess 16 indicae ha he VARs in levels and firsdifferences aear reasonably ell secified, a leas from he oin of vie of serial correlaion. 17 Alhough LM ess occasionally indicae saisically significan higher order serial correlaion, he roblem is generally no oo serious. Lile eigh is laced on Pormaneau saisics ha aear o indicae more serious roblems han do he LM ess, due o he robable disorion of he es saisic in he resence of a uni roo or near uni roo. In any case, residual correlograms ere checked visually and disclosed in all cases a aern of aroximae hie noise, only occasionally marred by a very small number of significan sikes. Togeher ih he oher evidence, his is suggesive of he fac ha any remaining serial correlaion is minor and ha 16 The resen discussion of diagnosic ess benefied grealy from he draf manuscri made available by Prof. Kaarina Juselius on her ebsie (referenced elsehere). The relevan ess are also oulined in Johansen (1995) and Lükeohl (undaed). 17 Residuals from hese models almos alays exhibi srong evidence of heeroskedasiciy and nonnormaliy; hoever, hese roblems are less imoran in erms of inference and hyohesis esing han is serial correlaion. I ould herefore aear ha he models are acceably ell secified. From he observed aern of he residuals, i is likely ha GARCHye effecs linked o he eaks and roughs caused by Brazilian fross are a he roo of his roblem. Ben Sheherd Page 16 10/3/2004

18 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke he models can safely be used as he basis for inference subec, of course, o he nonsaionariy roblem Seed and Comleeness of Price Transmission Imulse resonse funcion (IRF) analysis is he rimary VAR ool for assessing he dynamics of rice ransmission, as i allos he researcher o race he effec on all endogenous variables of a oneeriod shock o a single variable. 18 Table 5 resens summary resuls of IRF analysis conduced on re and osliberalisaion VARs in levels. One oin ha can immediaely be noed is ha in sie of he concerns raised reviously regarding saionariy, only in o cases (India and Uganda in he osliberalisaion suberiod) did IRFs urn ou o be exlosive; all ohers sele don o zero evenually, hough ofen decaying quie sloly. The inerreaion of simle IRFs is necessarily imressionisic, bu in his case can noneheless offer some valuable informaion on he dynamics of rice ransmission. Firsly, here is a comaraive lack of generalised evidence suggesing ha rice ransmission imroved subsanially folloing liberalisaion. Alhough cumulaive IRFs are larger and closer ogeher for o roducer counries suggesing ha rice shocks are ransmied more fully afer liberalisaion han before he oosie is he case for roducer rice IRFs in o oher counries, suggesing ha innovaions are ransmied less fully o roducers afer liberalisaion han before. Secondly, he IRFs for hree roducer counries eak laer folloing liberalisaion han rior o i, suggesing ha here is in fac a greaer lag in rice ransmission no han in he 1980s. For nearly all roducer counries, IRFs decay noiceably more sloly afer liberalisaion, suggesing ha alhough in some cases rice ransmission may be more comlee, i may also be sloer. 18 Pesaran & Shin (1998) generalised imulses ere used o generae he IRFs. Ben Sheherd Page 17 10/3/2004

19 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Finally, he case of he consumer markes is sriking. Their IRFs sugges ha in boh cases, rice ransmission as considerably faser and more comlee rior o liberalisaion han aferards. Tha is, IRF eaks ere earlier reliberalisaion, decay as faser and cumulaive IRFs ere closer ogeher. This is an imoran resul, as he facors affecing rice ransmission from he orld marke o consumer markes are fundamenally of a rivae naure, governmenal inervenions being essenially nonexisen. I is herefore highly suggesive of he fac ha changes in marke srucure folloing liberalisaion have had an imoran effec on he dynamics of rice ransmission, and no necessarily for he beer Direcion of Price Transmission Wihin he basic VAR frameork (levels or firs differences), he direcion of rice ransmission can be analysed by esing for Granger causaliy amongs he rice series in quesion. The null hyoheses are: firsly, ha lagged values of roducer rices do no ener he orld rice equaion significanly; and secondly ha lagged values of orld rices do no ener he roducer rice equaion significanly. Or in erms of consraining models (1) and (2), he ess are H 0 : 0, 1... k and H 0 ' : 0, 1... k (ih adusmens as needed in he cases of mulile orld rice series or inclusion of reail rices). Given ha crossequaion resricions are involved, he sysem has o be reesimaed by Seemingly Unrelaed Regression (SUR) and he hyohesis esed using Sims (1980) modified likelihood raio (LR) es, hich aears in equaion (3). (T is he oal number of observaions, c is he number of esimaed arameers and he o Σs reresen he residual covariance marices of he resriced and unresriced models resecively.) The LR es saisic is disribued as a chisquare variae ih degrees of freedom equal o he number of resricions imosed. ( log Σ Σ ) λ ( T c) log ~ χ 2 (3) R U Ben Sheherd Page 18 10/3/2004

20 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Reecion of H 0 and nonreecion of ' H 0 means ha revious roducer rices are relevan o deermining curren orld rices, bu ha as orld rices are no relevan o deermining curren roducer rices. Hence, causaion runs exclusively from roducer o orld rices, hich is consisen ih boomu rice ransmission. Similarly, nonreecion of H 0 and reecion of ' H 0 is consisen ih odon rice ransmission. If boh null hyoheses are reeced, hen causaion is bilaeral and rices changes are ransmied in boh direcions. Finally, if neiher null hyoheses can be reeced, here is no Granger causaliy in he sysem and ransmission beeen orld and roducer rices mus be very eak. Table 6 shos he resuls of hese ess in summary form across he o model yes used, namely lagaugmened VARs in levels and VARs in firs differences. The direcion of causaliy is indicaed ih arros. When more han one orld rice is involved, resuls are reored boh for individual Granger causaliy (e.g., BNG causes Brazil ) and oin causaliy (e.g., BNG and RG oinly cause Brazil ); oin causaliy is indicaed ih braces. Unless oherise indicaed, all ess are conduced a he 10% level of significance. As is common, he resuls of he Granger causaliy ess in his case dislay some sensiiviy o model secificaion. To generalise, hoever, i ould aear ha bilaeral ransmission beeen orld and US reail rices as he norm in boh suberiods, bu ha Germany exhibied boom u ransmission. For roducer counries, i seems clear ha boom u rice ransmission as considerably sronger rior o liberalisaion han aferards; insulaing roducers from orld rices as, indeed, one of he obecives of he ICA and so his is no an unexeced resul. In he firs eriod, five of he six roducer counries exhibi credible evidence of eiher bilaeral causaliy or boomu rice ransmission; only Colombia shos no sign hasoever of his mechanism s having been in oeraion. Afer liberalisaion, on he oher hand, only hree counries (Brazil, Colombia and India) seem o be involved in boomu rice ransmission, and in he cases of Brazil and India he evidence is mixed and no a all Ben Sheherd Page 19 10/3/2004

21 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Ben Sheherd Page 20 10/3/2004 robus o model secificaion. I ould herefore seem ha odon rice ransmission is a more imoran driving force in roducer markes afer liberalisaion han rior o i, bu ha no such change is aaren in consumer markes Symmery of Price Transmission VARs in firs differences are used as he basic model for assessing PTA. Hoever, a sligh modificaion is necessary: sysem (2) mus be decomosed o allo for differen coefficiens on osiive and negaive rice changes, hich can hen be esed for equaliy (imlying symmeric rice ransmission). This is done in sysem (4), in hich is zero for firs differences less han or equal o zero and oherise is equivalen o he firs difference oeraor and akes a nonzero value only hen firs differences are negaive. This is a simle adaaion of more radiional, singleequaion PTA models and finds an echo in he asymmeric VECM aroach. (See he overvies in von CramonTaubadel & Meyer (undaed), Gonzales e al. (2003) and Gómez & Koerner (2002).) Φ Φ k k k k k k DU DU ε µ ε µ (4) Sysem (4) can be esed for PTA using he folloing hyoheses. Firsly, he symmeric secificaion in model (2) can be esed as a nesed model ihin he more general asymmeric secificaion in model (4). Formally, e es he oin H0: k Reecion of he null hyohesis imlies ha here is some evidence of PTA on a sysemide basis and ha furher invesigaion is required o uncover is naure and exen. The second sage of PTA analysis is herefore o es each equaion in he sysem searaely, again using he null hyohesis ha rice ransmission is symmeric in ha equaion, agains he alernaive ha i is asymmeric in ha aricular equaion. In formal erms, e sequenially es null hyoheses

22 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke of he form H0:, 1... k and H0: k, Reecion of eiher null hyohesis imlies ha rice ransmission in he equaion concerned is asymmeric. Finally, e es exlicily for shorerm PTA occurring over one eriod only by conducing 2k indeenden ess of null hyoheses of he form H0: and H0:. Once again, reecion of any one of he null hyoheses is indicaive of shorerm PTA a a given lag. Summary resuls from esing a he 10% level aear in Table 7, in hich he conclusion ha PTA exiss in one direcion or anoher includes he siuaion here shorerm PTA is found o exis (i.e., a null hyohesis of symmery is reeced even in resec of one lagged rice change). On ha basis, only one roducer counry (Guaemala) as no subec o PTA in he reliberalisaion eriod. I is noable ha exce in he case of Brazil, PTA in roducer counries rior o liberalisaion ook he form of asymmeric ransmission of orld rices o roducer rices, hile changes in he oher direcion ere ransmied symmerically. Tha is, he resuls in Table 7 are broadly consisen ih he rice mainenance and sabilisaion racices used by roducer counries (exce noably Guaemala) before liberalisaion and hich ere exressly designed o sheler roducers o some exen from changes in orld rices. Wha is far more surrising in erms of roducer counries is ha Table 7 discloses significan evidence of PTA afer liberalisaion as ell; only Mexico and India exhibi symmery of rice ransmission. Moreover, PTA in he osliberalisaion eriod is (exce for Brazil) in boh direcions, ha is from orld o roducer rices as in he reliberalisaion eriod and from roducer o orld rices. This is an exremely significan resul, as i suggess ha far from imroving he qualiy of rice ransmission, liberalisaion has in fac ended o erode i. Given ha he coss of adusing rices a he orld and roducer levels should no be quie Ben Sheherd Page 21 10/3/2004

23 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke small esecially in he general absence of governmen rice conrols in roducer counries osliberalisaion i is difficul o rely on menu coss as an exlanaion of PTA in his case; ha is, menu coss a his level should have decreased subsanially folloing liberalisaion, ye PTA has seemingly orsened. Anoher ineresing resul is ha consumer counries also dislay significan evidence of PTA, running boh from orld o reail rices and from reail o orld rices. This is rue boh before liberalisaion and aferards. While i is ossible ha menu coss are a leas ar of he exlanaion for PTA from orld o reail rices, i is more difficul o see heir role in he oosie direcion, as here is lile reason o exec rice sickiness a he orld marke level Changes in he Markeing Margin To isolae changes in he markeing margin folloing liberalisaion, i is necessary o esimae VAR models over he full samle eriod, using dummy variables o ake accoun of relevan breakoins. One aroach is o es he null hyohesis H0: Φ Φ, reecion of hich ould be consisen ih a change in he markeing margin folloing liberalisaion, as i ould sho level ums of differen magniudes in he o series. Secondly, e make use of he fac ha he consan erm in a VAR in levels can be inerreed as a summary of he condiion of genuinely exogenous variables (Greene, 2000). Prior o liberalisaion, e ould exec hese erms o be differen for roducer and orld rices, o ake accoun of he differen exogenous olicies o hich hey ere subec. Afer liberalisaion, i seems lausible ha hey migh be subec o he same exogenous condiions. Hence, e es H µ 0 : µ and comare he resul ih ha obained under H 0 : µ Φ µ Φ. If he firs null is reeced bu he second is no, his suggess ha he exogenous condiions surrounding rice formaion a he o levels have come closer ogeher folloing liberalisaion. Ben Sheherd Page 22 10/3/2004

24 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke Table 8 resens a summary of resuls from ess of he above hyoheses, again a he 10% level. For four of he six roducer counries, he markeing margin decreased as execed folloing liberalisaion. Hyohesis ess sugges ha i remained unchanged for Guaemala and India: a lausible resul for he former, given he grahical evidence, bu very surrising for he laer. The Indian resul seems o flo from he fac ha none of he dummy variables enered he VAR significanly, erhas indicaing ha he hyohesised daes of srucural breaks should be rehough. The Table also shos ha for four of he six counries, he combined osliberalisaion exogenous erm is indisinguishable as beeen he roducer and orld rice equaions, here i as saisically differen rior o liberalisaion. As ih he aaren generalised reducion in markeing margins, his suggess imroved marke inegraion folloing liberalisaion and is essenially ha is execed on a sandard vie of he benefis from liberalisaion. Finally, i mus be noed ha he markeing margin for Germany has increased according o he hyohesis ess erformed. Indeed, even hough he resul for he USA is unchanged, i should be noed ha he signs and magniudes of he relevan dummy variables sugges srongly ha he margin increased here as ell, even if he effec is no saisically aaren. Alhough here are oher ossible exlanaions for such an exansion of he margin noably relaed o he share of noncoffee coss in roducion of he final roduc he influence of marke srucure cerainly canno be excluded. 5. Conclusions and Agenda for Furher Research The resuls of he resen sudy can be summarised as follos: Liberalisaion has brough abou only limied imrovemens in he seed and comleeness of rice ransmission among roducer counries, he orld marke and Ben Sheherd Page 23 10/3/2004

25 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke consumer counries; in some cases, ransmission ould even aear o have orsened in erms of hese crieria. There has been a disinc move oards a sysem of rice ransmission ha is more odon han boomu, a reversal of he siuaion ha exised rior o liberalisaion. Bilaeral ransmission is a subsanially less common henomenon han execed. Asymmeric rice ransmission ersiss a all levels of he marke in sie of liberalisaion, and could even said o be more idesread no han i as before. Alhough liberalisaion seems o have delivered on is romise of reduced markeing margins in mos roducer counries, here is some evidence o sugges ha margins have increased on he consumer side. Wha is sriking abou hese resuls is he raher sombre icure hey ain of he imac of marke liberalisaion in he 1980s and 1990s. The rerea of he sae from he coffee rocessing chain does no ye aear o have led o he esablishmen of a aricularly ellfuncioning marke. In ligh of he increased firm concenraion idenified above, i is cerainly ossible ha marke srucure and marke oer have layed a role in roducing some of he observed resuls. Alhough i is far oo early o rovide a definiive anser o he rofieering roasers hyohesis, I ould sugges ha here is no sufficien evidence o sugges ha he firms involved a leas have a rima facie case o anser and ha furher, more deailed invesigaion is called for. Wih his in mind, here are a number of direcions oen o fuure research, boh in erms of he coffee marke and oher rimary commodiy markes ha exhibi similar characerisics: The aroach aken here could be rendered more consisen and inclusive by using a anel daa VAR, hich ould enable esing of crosscounry hyoheses. This ould also faciliae a more deailed comarison of he links beeen differen marke Ben Sheherd Page 24 10/3/2004

26 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke srucures across roducer and consumer counries and divergences in observed crosscounry oucomes. More deailed rice daa could be used o model consumer markes ih a number of differen rocessing sages, hereby giving a richer icure of rice ransmission dynamics and oenially heling o idenify more recisely he locaion of marke imerfecions in he consumer counries. Mulivariae volailiy (GARCHye) models could be used o model he ransmission of rice volailiy hrough he rocessing chain (cf. Yang e al. (2003) and Weaver & Nacher (2000)). Alicaion of similar mehodologies o differen rocessing chains ould also rovide a useful Given ha commodiy rocessing chains are ofen inernaional, here are boh olicy and oliical imlicaions of his research a he suranaional level. The issue has already been raised in a general ay a he WTO (see Kenya e al. (2003) and WTO (2003)) and he rade olicy communiy is saring o come alive o he issues involved: see e.g., Abbo (1998, 2003), Gilber & Varangis (2003), Josling (1999, 1999a) and MacLaren & Josling (1999). Hoever, his research rogramme is in is infancy arly due o he auciy of suoring emirical ork. To ensure ha he oliical discourse akes lace ihin an aroriae inellecual frameork, i is herefore imoran for researchers o devoe ime o boh he emirical and heoreical issues involved. Ben Sheherd Page 25 10/3/2004

27 Marke Poer in Inernaional Commodiy Processing Chains: Preliminary Resuls from he Coffee Marke References Abbo, P. (1998). Comeiion Policy and Agriculural Trade. OECD Documen COM/AGR/CA/TD/TC/WS(98)106. Abbo, P. (2003). Toards More Socially Resonsible Cocoa Trade. IATRC Working Paer Akiyama, T. (2001). Coffee Marke Liberalisaion Since In Akiyama, T., Baffes, J., Larson D. F. & Varangis, P. (eds). Commodiy Marke Reforms: Lessons of To Decades. Washingon D. C.: The World Bank. Akiyama, T., Baffes, J., Larson, D. & Varangis, P. (2001). Marke Reforms: Lessons from Counry and Commodiy Exeriences. In Akiyama, T., Baffes, J., Larson D. F. & Varangis, P. (eds). Commodiy Marke Reforms: Lessons of To Decades. Washingon D. C.: The World Bank. Baffes, J. & Aad, M. I. (1998). Deecing Price Linkages: Mehodological Issues and an Alicaion o he World Marke of Coon. Mimeo. Beendorf, L. & Verboven, F. (1997). Ho Comeiive is he Duch Coffee Marke? CPB Reor, 97(4), Beendorf, L. & Verboven, F. (1998). Comeiion on he Duch Coffee Marke. Research Memorandum 141. The Hague: Cenral Planning Bureau. Bukenya, J. O. & Labys, W. C. (2002). Price Convergence on World Commodiy Markes: Fac or Ficion. Research Paer 20021, Wes Virginia Universiy. Chavas, J.P. & Meha, A. (2002). Price Dynamics in a Verical Secor: The Case of Buer. Saff Paer 452, Universiy of WisconsinMaddison. Dolado, J. J. & Lükeohl, H. (1994). Making Wald Tess Work for Coinegraed VAR Sysems. Mimeo. Dolan, C. & Humhrey, J. (2000). Governance and Trade in Fresh Vegeables: The Imac of UK Suermarkes on he African Horiculure Indusry. Journal of Develomen Sudies, 37(2), Feuersein, S. (2002). Do Coffee Roasers Benefi From High Prices of Green Coffee? Inernaional Journal of Indusrial Organisaion, 20, Fier, R. & Kalinsky, R. (2001). Can an Agriculural Commodiy be DeCommodified, and if so, Who is o Gain? Discussion Paer 380. Brighon: Insiue of Develomen Sudies. Gilber, C. L. (1996). Inernaional Commodiy Agreemens: An Obiuary Noice. World Develomen, 24, 119. Gilber, C. L. & Varangis, P. (2003). Globalisaion and Inernaional Commodiy Trade ih Secific Reference o he Wes African Cocoa Producers. Working Paer Cambridge, MA: Naional Bureau of Economic Research. Gómez, M. I. & Casillo, M. (2001). Srucural Change, Rens Transferring and Marke Poer in he Inernaional Coffee Marke: A Time Series Analysis. Mimeo. Gómez, M. I. & Koerner, J. (2002). Do Reail Coffee Prices Increase Faser Than They Fall? Asymmeric Price Transmission in France, Germany and he Unied Saes. Mimeo,.fundacion.uc3m.es/earie2002/aers/aer_496_ df. Gonzales, F., Guilloreau, P., Le Grel, L. & Simioni, M. (2003). Asymmery of Price Transmission Wihin he French Value Chain of Seafood Producs. Working Paer Toulouse: Insiu Naional de la Recherche Agronomique. Greene, W. H. (2000). Economeric Analysis. Fourh ediion. Ne Jersey: PreniceHall. Ben Sheherd Page 26 10/3/2004