Moneary Policy Imacs on Cash Cro Coffee and Cocoa Using Srucural Vecor Error Correcion Model By Ibrahim Bamba Michael Reed Preared for resenaion a he Meeing of American Agriculural Economiss Associaion, Aug. 1-4 2004 in Denver, Colorado Ibrahim Bamba is a graduae Suden and Michael Reed is a rofessor. All are in he Dearmen of Agriculural Economics a he Universiy of Kenucky Coyrigh 2004 by Ibrahim Bamba and, Michael R. Reed. All righs reserved. Readers may make verbaim coies of his documen for non-commercial uroses by any means, rovided ha his coyrigh noice aears on all such coies.
Moneary Policy Imacs on Cash Cros Coffee and Cocoa Using Vecor Error Correcion Model Absrac The inernaional marke for he roical cros coffee and cocoa is marked by high rice insabiliy. This aer invesigaes wheher moneary olicy disurbances conribue o cocoa and coffee rice insabiliy. The economeric evidences oin oward high flexibiliy of he rices of cocoa, arabica coffee, and robusa coffee relaive o he indusrial rice and o he exchange rae. Money suly shock has ersisen imac he roical cro rices and exlains an economically significan roorion of heir rices variabiliy. Key words: agriculural rices, volailiy, overshooing, coinegraion Inroducion Coffee and cocoa are wo agriculural commodiies roduced mainly in develoing counries, exored, and consumed almos enirely in high-income indusrialized counries. In several develoing counries, cocoa and coffee are he main deerminans of aggregae exors and overall economic erformance. Saisics from UNCTAD 1 hel ell he sory well. Landlocked African counries such as Burundi and Rwanda rely on coffee for more han 80 ercen of heir oal exors earnings. In Ehioia, coffee s share of oal exor is as high as 79 ercen. The economy of Coe d Ivoire is heavily secialized on cocoa and coffee. Cocoa alone reresens 15 ercen of Coe d Ivoire s GDP and more han 35 ercen of her oal exors. In Cenral and Souh 1 Unied Naion Conference on Trade And Develomen
America, coffee and cocoa reresen he majoriy of exors for counries such as Columbia, Cosa Rica and Haii. The inernaional marke for cocoa and coffee is marked by high rice insabiliy. From January 1990 o December 2003, he coefficien of variaion for cocoa rice, robusa coffee and arabica coffee were, resecively, 22.5%, 51.1% and 42.8%. The high rice volailiy of hese commodiies is exlained generally by real economic facors such as roducion deendence on variable biohysical elemens, inu subsidies favoring excess suly, irreversible invesmen due o heir erennial naure, low-income elasiciy, and inelasic demand. The rice volailiy of hose rimary commodiies can be aribued o moneary and financial imacs. Changes in moneary olicy can affec nominal commodiy rices and ossibly real commodiy rices. This has been he subjec of recen lieraure. A focus of he lieraure on moneary imacs has been on overshooing of commodiy rices. Frankel (1986) adaed he overshooing hyohesis firs inroduced by Dornbusch (1976) o agriculure and analyically derived he dynamics of commodiy rice in a closed economy. By subsiuing agriculural rices for he exchange rae in Dornbusch s overshooing model, Frankel (1986.345) reored ha wih an unaniciaed and ermanen one ercen dro in he suly of money in he long run we would exec all rices o fall by one ercen in he absence of new disurbance. Bu, in he shor run indusrial good rices are fixed ( ) o equilibrae money demand, ineres rae rise. Bu we have an arbirage condiion ha mus hold in commodiy markes: since commodiies are sorable, he rae of reurn on Treasury bill can be no greaer han he execed rae of increase of he commodiy rices minus sorage coss 2. This means ha he so rice of commodiies mus fall oday and mus fall by more han he one ercen 2 In his aricle, Frankel (1986) wroe ha he execed increase of commodiy rices lus sorage coss could no be greaer han he ineres rae. Gordon (1987) correced ha flaws and quesioned he arbirage condiion in commodiy markes.
ha i is execed o fall in he long run. Frankel (1986) modeled analyically he dynamic ah of commodiy rices relaive of heir real or fundamenal long-run equilibrium, subsequen o a change in he money marke. Objecive of he Sudy The objecive of his secion is o invesigae wheher moneary facors conribue o coffee and cocoa rice insabiliy using srucural vecor error correcion models. A modified emirical framework develoed by Saghaian, Reed, and Marchan (2002) is used o es he imlicaion of he overshooing hyohesis for cocoa and coffee subsequen o a moneary shock in he commodiy imoring counry, such as he Unied Saes. If cocoa and coffee rices overshoo following a change in he moneary olicy of he Unied Saes, some evidence of he ransmission of moneary olicy disurbances from develoed counries oward develoing counries would have been unraveled. This asserion assumes ha develoed counries rimarily imor commodiies and develoing counries rimarily exor commodiies counries. Secifically he link beween changes in he US moneary olicy shocks and he volailiy of roical commodiy rices is invesigaed. Saghaian, Reed, and Marchan (2002) used a four-variable ime series model in heir emirical invesigaion of he overshooing model. The four variables were agriculural rices, indusrial rices, money suly, and exchange rae. This analysis adds adds a fifh variable o heir model, he ineres rae, which is naurally ar of he sysem. Unlike Saghaian, Reed, and Marchan (2002), he overshooing of he roical cros rices is invesigaed wih and wihou he assumion of money neuraliy. Wih he money neural assumion, he money suly, he agriculural rice, and he indusrial rice are assumed o move roorionaely in he long run. The rices of cocoa and coffee being deermined in fuures and oions exchange marke; heir relaive flexibiliy and he relaive sickiness of indusrial good rice is examined. The rice of indusrial good is
resumed sicky given he revalence of longer-erm conracual arrangemens in he indusrial secor. Daa and Mehod 3.1. Mehod The alicaion of he overshooing model consiss of hree ses. In he firs se, The saionariy roery of each variable is examined using he univariae augmened Dickey Fuller (1979) uni roo es and he Philis-Perron (1988) uni roo ess. Due o he low reliabiliy of he es for a uni roo, addiional saionariy ess such as he behavior of he univariae auocorrelaion funcion and ime lo are reored. In he second se, he exisence of coinegraing relaionshis among he variables is esed using he Johansen-Juselius rocedure. In he hird and las se, he vecor error correcion models are esimaed under alernaive assumion on he long-run imac of money suly. The validiy of he overshooing hyohesis of Dornbusch is assessed by comaring he sign and magniude of he adjusmen arameers for indusrial good and he agriculural commodiy of ineres. The macroeconomic variables, he sicky indusrial rice, and he agriculural rice are all reaed as endogenous. Daa Monhly ime series daa were colleced from 1981:01 o 2003:12. All U.S macroeconomic daa are ublicly available a he Inerne sie of he Federal Reserve Bank of S Louis. The conceual variables, ineres rae, exchange rae, money suly, and indusrial rice, are reresened by, resecively, he 3-monh Treasury bill rae, he rade-weighed exchange value of U.S. dollar versus currencies of major rading arners 3, he M1 money sock, and he roducer rice index for finished goods (finished goods excluding foods). The level daa are shown in Figure 1 hrough Figure 7. The rices for coffee and cocoa were rerieved, resecively, from he Inernaional Coffee 3 The exchange rae is refereed o exacly as weighed average of he foreign exchange value of he U.S. dollar agains a subse of he broad index currencies ha circulae widely ouside he counry of issue.
Organizaion and he Inernaional Cocoa Organizaion. Given ha coffee is a heerogeneous rimary commodiy, he rice for he mos recognized commercial varieies, which are arabica and robusa, are used. The coffee rices are secifically he rice of oher mild arabica and he rice of robusa coffee 4. The choice of using he U.S. macroeconomic daa is dicaed by several consideraions. The world rices of cocoa and coffee are denominaed in U.S. dollars. The U.S. is he leading imorer of boh green coffee beans and cocoa beans wih, resecively, almos 25 ercen and 20 ercen of world oal imors in 1998 (Food and Agriculure Organizaion, FAO). The U.S. ranks high in world consumion for coffee and cocoa. In 1998, er caia consumion of coffee and cocoa was, resecively, 4.1 and 2.42 kilograms. Procor & Gamble, Phili Morris, Mars, and Hershey, all U.S. comanies, are major layers in he oligoolisic marke of goods derived from coffee and cocoa beans. Finally, he New York Board of Trade (NYBOT) 5, wih London based LIFFE 6, is one of wo erminal markes for cocoa and coffee. The coffee raded in NYBOT is he arabica variey and he robusa coffee is mainly raded in LIFFE. Coffee is quoed in U.S. dollars on boh markes. Uni roo ess The augmened Dickey-Fuller ess and he Philis-Perron ess are used o deec he resence of a uni roo in a ime series. The augmened Dickey-Fuller (ADF) assumes ha he errors are saisically indeenden and have a consan variance. To remove he ossibiliy of serial correlaion in he residuals when erforming he augmened Dickey Fuller es, he lieraure recommends regressing he deenden variable on a sufficienly 4 The inernaional coffee organizaion classifies coffee deending on he counry of origin as: Colombian mild arabicas, oher mild arabicas, Brazilian naural arabicas and robusas 5 The New York Coffee, Sugar and Cocoa Exchange (CSCE) was he radiional exchanges marke for cocoa and coffee in he Unied Saes: since 1998 CSCE is a subsidiary of NYBOT 6 LIFFE is for London Inernaional Fuures and Oions Exchange
large number of lags in order o remove he serial correlaion exising in he residuals. Six lags are included in each of our univariae ess for saionariy. The ADF ess involve leas square regression of he firs difference of he series agains he series lagged one eriod and five oher lag difference erms. Moreover, adding five lagged firs difference erms in he uni roo regression minimized he Schwarz informaion crierion. The ADF es secified wih an inerce is reresened as: n i, = 0 + α1z i, 1 + α i (1 B) j= 1 Z α Z + e (2.23) i, j where is he difference oeraor and e is whie noise. The Philis-Perron es is erformed o rovide addiional robusness o he uni roo es. Philis-Perron (1988) develoed a generalizaion of he Dickey-Fuller wih less resricive assumions regarding he disribuion of he error erms. The ossibiliy of heeroskedasic errors is accommodaed in he Philis-Perron es. n i, = 0 + α1z i, 1 + α i (1 B) j= 1 Z α Z + e (2.24) i, j The augmened Dickey-Fuller uni roo es resuls and he Philis-Perron resuls (able 1) reinforce each oher. All series are inegraed of order one. The uni roo ess sugges he naural logarihms of he original series are no saionary bu heir firs differenced are saionary. Because of he low ower of he uni roo ess, addiional invesigaions of he auocorrelaion funcion and ime lo of each series were erformed. For a series o be saionary, he auocorrelaion funcion should converge quickly o zero. The auocorrelaion funcion and he visual insecion of he grahical reresenaion of each series suor he conclusion reached using he augmened Dickey-Fuller and he Philis-Perron uni roo ess. I is concluded ha all series are I (1) rocess and hey saisfy he necessary condiion for a coinegraed relaionshi. Coinegraion ess
To es for he exisence of coinegraion beween each agriculural rice (arabica, robusa, and cocoa) and he U.S. economic variables (indusrial rice, money suly, exchange rae, and ineres rae), he Johansen-Juselius maximum likelihood rocedure is used o es he exisence of coinegraion. The Johansen-Juselius es is a mulivariae rocedure ha can be reresened wih he following error correcion reresenaion: Z = α + ΠZ i + 1 0, 1 Γ Z + ε (2.25) j= 1 where Z is a vecor conaining he five endogenous variables, which include he agriculural rice (eiher cocoa or coffee), he money suly, he exchange rae, he ineres rae and he indusrial rice. The marix Γ i conains he shor-run coefficiens among he variables in he sysem, ε is whie noise and he marix Π conains he informaion on he evenual coinegraion relaions among he series in he vecor Z. Equaion 2.25 can be rewrien as: Z = α αβ Z i + 1 0, 1 Γ Z + ε (2.26) j= 1 The reduced rank marix Π caures he long-run saionary relaionshi among he variables. When he marix Π has a reduced ranked here is a facorizaion Π = αβ, where he marix β conains he r coinegraing relaions and he marix α conains he adjusmens arameers in he vecor error correcion model (VECM), or he shor-run overshooing arameers in he model. Boh β and α are 5 x 2 marices in his alicaion. The saionary error correcion erms are ΠZ -1 = αβz -1. The error correcion ar and he VAR ar can accommodae differen rend secificaions. Regarding he coinegraing equaion ( αβ Z i, 1 ) and he deerminisic comonens (α 0 ) o include in he error correcion reresenaion, Johansen (1995) considered five ossible rend secificaions. Johansen (1994) demonsraed ha he inclusion of inerces in he vecor Z of a differenced series imlies ha he nonsaionary level
variables have a linear rend. When he vecor Z of differenced series and he coinegraing equaions have linear rends, he nonsaionary variables have quadraic rend behavior. Below are he five ossible secificaions: 1. The vecors of level daa, Z, have no deerminisic rends and he coinegraing equaions do no have inerces: Z = Π Z i + 1, 1 Γ Z + ε, where Π =αβ (2.27) j= 1 2. The vecors of level daa Z have no deerminisic rends and he coinegraing equaions have inerces: Z = Π Z i + 1, 1 Γ Z + ε, where Π = α(β Z -1 + ρ 0 ) (2.28) j= 1 3. The vecors level daa Z have a linearly rend and he coinegraing equaions have only inerces: Z = α + Π Z i + 1 0, 1 Γ Z + ε, where Π = α(β Z -1 + ρ 0 ) (2.29) j= 1 4. The vecors of level daa Z have and he coinegraing equaions have linear rends: Z = α + Π Z i + 1 0, 1 Γ Z + ε, where Π= α(β Z -1 + ρ 0 + ρ 1 ) (2.30) j= 1 5. The vecor Z of level daa have quadraic rends and he coinegraing equaions have linear rends: Z = α + α + Π Z i + 1 0 1, 1 Γ Z + ε, where Π= α(βz -1 + ρ 0 + ρ 1 ) (2.31) j= 1 The Johansen-Juselius maximum likelihood rocedure ess he rank of Π by esing he number of non-zero eigenvalues or characerisic roos. Conceually, four ossible scenarios can arise.
1- The rank of Π = zero, he variables in he vecor Z is inegraed of order one; hey are I (1) bu no coinegraed. 2- The rank of Π = five, his is he case of full rank marix which, imlies all variables are I (0). 3- The rank of Π = one, here is only one linear indeenden row in Π or one coinegraing vecor. 4- The rank of Π = r, 1 < r < 5, here are r linearly indeenden rows in Π or r coinegraing vecors. Two null hyoheses are emloyed o es for he exisence and he number of coinegraed relaionshis in he Johansen mehodology. They are referred o in he lieraure as λ-race or he race saisic es and λ-max es or he maximal eigenvalue es 7. Boh he race es and he maximal eigenvalue es are alied using a sequenial rocess. The null hyohesis for he race saisic is he exisence of r coinegraing relaions versus he absence of r coinegraed relaionshis. The rocedure begins by esing wheher here is no coinegraion (r = 0). The rejecion of he null hyohesis leads o esing higher orders of coinegraion. The second es saisic (λ-max es) is used o es he null hyohesis ha here are a mos r coinegraing relaionshis versus he alernaive of r + 1 coinegraing relaionshi. Prior o erforming he Johansen-Juselius maximum likelihood es for coinegraion, he oimal lag order o include mus be deermined. The deerminaion of he lag lengh is carried ou done by using numerous diagnosic saisics such as he final redicion error (FPE), Schwarz informaion crierion (SC), he sequenial modified likelihood raio (LR) es, he Akaike informaion crierion (AIC), and he Hannan-Quinn informaion crierion (HQ). The diagnosic saisics are resened in able 2, 3, and 4. They indicae ha he lag order hree resuls in oimal values for four ou five 7 The asymoic disribuions of he race and maximal eigenvalue es deend on he secificaion used for he daa generaing rocess.
mulivariae diagnosic saisics for he sysem conaining he given roical cro rice (cocoa, arabica or robusa) and he U.S. macroeconomic variables (i.e., money suly, ineres rae, exchange rae, indusrial rice). The coinegraion ess are erformed using each of he five ossible deerminisic rend secificaions of Johansen (1995). The summary for all secificaions of coinegraion ess is resened in able 5. Alhough he numbers of coinegraing vecors deermined wih he race es and he maximal eigenvalue es aear sensiive o he secificaion of he daa generaing rocess, a leas one coinegraing relaionshi is found a he 5% confidence level, regardless of he mulivariae secificaion used for he Johansen rank es. The secificaion reresened by he equaion 2.31 exceed, he secificaions reresened by equaions 2.27, 2.28, 2.29 and 2.30 sugges he resence of wo coinegraing vecors. The resence of more han one coinegraing relaionshi linking he U.S. macroeconomic variables (i.e. money suly, ineres rae, exchange rae, indusrial rice) and each of he commodiy rice of ineres (i.e. robusa coffee, arabica coffee, or cocoa), using he secificaion sugges he resence of 2 linear combinaions of nonsaionary series yielding a saionary relaionshi beween he variables. The exisence of mulile coinegraing vecors imlies ha i may no be ossible o idenify he behavioral relaionshis from he reduced-form relaionshis (Enders, 2003.323). For examle, wih wo coinegraing vecors, here is an indeenden coinegraing vecor for each combinaion of four variables. To rovide meaningful economic inerreaion of he resuls, he sysem needs o be idenified. According o Johansen and Juselius (1994), for a number r of coinegraing vecors, r² resricions are needed o idenify he coinegraing vecor β wihou changing he likelihood funcion. To idenify a sysem conaining wo coinegraing vecors, here mus be a leas wo resricions imosed on each coinegraing vecor.
Emirical Resuls 4.1. VECM Resuls using he Unresriced SRM mehod Similar o he coinegraion es, he esimaion of he error correcion model requires one o secify he deerminisic comonens in he mulivariae differenced ime series and in he coinegraing relaionshis. In he absence of rior heoreical guidance, his is done by choosing he secificaion ha maximizes he log-likelihood funcion. The secificaion (2.31), where he variables in he mulivariae nonsaionary ime series, have quadraic rends is found o maximize he log-likelihood funcion. Hence, he esimaion of he error correcion models is carried ou using he secificaion ha include in linear rend in boh he coinegraing relaionshis and he firs differenced vecor auoregressive comonen. Furhermore, given ha all oher deerminisic rend secificaions oulined by Johansen (1944) are nesed in he secificaion wih quadraic rends, missecificaion bias should be minimized by using ha secificaion. The coinegraion ess wih he quadraic rend secificaion are resened in he able 6. Wih he quadraic rend secificaion, only one coinegraing vecor is found a he higher 5% confidence level. There is hus no need for arbirary idenificaion and he only requiremen is o normalize he esimaes wih one of he variables. In his invesigaion, all esimaes are normalized wih resec o he agriculural commodiy rice. The unique coinegraing vecor is inerreed as he long-run equilibrium relaionshi (Engle and Granger, 1987). The esimaes of he unresriced adjusmen coefficien for he agriculural commodiy and he indusrial good he normalized coinegraing vecor are hen used o assess he inference of he overshooing hyohesis. Table 7 resen he resuls obained wih he idenificaion mehod of SRM. In he model conaining robusa coffee rice, he long-run arameers esimaes for he imac of U.S money suly on robusa rice, arabica rice and cocoa rice are, resecively,
6.20%, 0.99%, and 2.43% 8. The hyohesis ha money is neural is rejeced for robusa coffee a he 5% significance level. The hyohesis of money neuraliy is no rejeced for models ha include cocoa and arabica coffee as he agriculural rice. A 1% increase in money suly leads o a long-run increase of cocoa rice, robusa coffee, and arabica coffee rice by, resecively, 6.20%, 0.99%, and 2.43%. The long-run relaionshi beween money suly and he nominal rice on he agriculural commodiies seems o be sensiive o he good considered. The sign of he coefficiens measuring he long-run imac of money suly on cocoa, robusa or arabica coffee are consisen wih our execaions. An exansionary moneary olicy osiively affecs he rices of robusa coffee, arabica coffee, and cocoa. Ye he effec of money suly on robusa coffee rice is unusually large. The adjusmen arameer for he agriculural commodiy rice in he coinegraing vecor is viewed as he overshooing arameer for he commodiy rice because i reresens he deviaion from he long-run equilibrium relaionshi. Likewise, he adjusmen arameer for he indusrial roduc rice is is overshooing arameer. The agriculural rice overshoos in he shor-run when he absolue value of is adjusmen arameer is greaer han he adjusmen arameer for he indusrial good. The adjusmen arameers are resened in able 8. In he cocoa model, he overshooing coefficien is -1.79% and he overshooing coefficien for he indusrial good rice is 0.08%. For robusa coffee he overshooing coefficien is -1.86% and he coefficien for indusrial good rice is 0.04%. Finally, for arabica coffee, he overshooing coefficien is 3.22% and he coefficien of he indusrial good is close o zero. These emirical findings conform o he overshooing findings of Frankel and SRM for agriculural goods. The overshooing arameers for all hree agriculural goods are saisically differen from zero and greaer han he overshooing 8 Johansen (2002) demonsraed ha he long-run coefficiens in a coinegraing relaionshi could be called elasiciies if he variables are measured in logarihms.
arameers for indusrial good rice. All of he coefficiens for indusrial good rice are insignifican. Arabica coffee, robusa coffee rice, and cocoa rices are herefore flexible and reac faser o macroeconomic disurbances, while he indusrial good are sicky. To summarize he resuls, he signs of he overshooing arameers for he imored commodiies are negaive as execed; hey indicae ha rice mus fall afer a macroeconomic shock o reesablish he long run equilibrium among he variables. Even hough he resuls for cocoa and arabica coffee aear reasonable, he lausibiliy of he long run imac of macroeconomic variables such as money suly and exchange rae on robusa rice is quesionable. For examle, he high imac of money suly on robusa rice does no aear o mach he observed downward rend. To imrove on he SRM model, he assumion of money neuraliy is imosed and he behaviors of he agriculural commodiy rice and indusrial good rice are revisied. 4.2. Resuls wih Long-Run Neural Money Resricions In he SRM model esimaed earlier, money suly is allowed o have real effecs on goods rice in boh he shor-run and in he long run. An alernaive emirical verificaion of he shor-run effec of money on indusrial and agriculural rices in he overshooing framework requires one o osi exlicily ha money is neural in he longrun (Roberson and Orden, 1990). Hence, he long-run imac of money suly on indusrial good rice and agriculural rices (robusa coffee, arabica coffee, and cocoa) are resriced o uniy for his ar of he analysis. The aroriaeness of he resricion will also be esed. The ess for he long-run neuraliy of money suly on he roical beverage rices are conduced using he likelihood raio saisic. The resuls resened in able 9 show ha he hyohesis of money neuraliy canno be rejeced a he 5% level of significance for all hree sysems. The likelihood raio es is erformed using he asymoic chi-square disribuion. The sizes of he imacs of oher macroeconomic
variables, such as he exchange rae, on agriculural good are sensible, secifically in he robusa coffee model. For insance, in he unresriced SRM model, an increase of he exchange rae by one ercen increases he rice of robusa rice by more han 7.45%. Wih he long-run money neuraliy assumion, he same augmenaion of he exchange rae leads o a 3.30% increase in he rice of robusa coffee. In all he hree models, he long-run imacs of exchange rae and ineres rae on he roical commodiy rices are saisically significan. The long-run inerce and he ime rend coefficien are negaive. These sugges he ossible exisence of negaive rend beween he roical cro rices and he Unied Saes economic variables included here. The adjusmen arameers obained when long-run money neuraliy is imosed are resened in able 10. The resuls sill indicae ha he rices of cocoa and boh yes of coffee overshoo. Their overshooing arameers are significanly differen from zero and heir magniudes are greaer han he overshooing arameer for he sicky indusrial rice. The adjusmen coefficiens for he indusrial good rice remain insignifican in all hree models. Therefore, cocoa rice and coffee rices reurn o heir long-run equilibrium faser han he indusrial good rice. Arabica coffee rice reacs more srongly o he Unied Saes marke informaion han robusa coffee or cocoa rice. The monhly overshooing arameer for arabica coffee and cocoa are resecively -3.41% and -2.57%. This migh be due o he fac ha arabica rice is aken from he NYBOT whereas robusa is aken from he LIFFE in London, England. The overshooing coefficiens obained when he assumion of neural money is imosed are similar o he ones obained wih he SRM mehod. The degree of overshooing of boh yes of coffee is slighly higher when money is assumed neural and he arabica rice aears o overshoo more han he robusa rice o change in he US macroeconomic variables. For cocoa, he degree of overshooing is marginally lower wih money neuraliy han wih he SRM model. As execed, he signs of he
adjusmen coefficiens on money suly and ineres are differen. The ineres rae, being he ooruniy cos of money, should move rices in he oosie direcion of money suly changes. In he unresriced SRM model, his did no hold for all cases, bu he ineres rae coefficien was always he oosie sign from he money suly coefficien in he second exerimen. Therefore, he resriced SRM reresenaions imrove on he esimaion of he dynamic relaionshi beween he Unied Saes macroeconomic variables and he rice of roical beverage coffee and cocoa by yielding lausible elasiciy esimaes. 4.3. Imulse Resonse and Variance decomosiion Nex, he imac of money suly on he volailiy of cocoa and coffee rice is invesigaed using innovaions accouning echniques such as he imulse resonse funcions and he variance decomosiion. The imulse resonse funcion measure he effec of a shock o one variable anoher variable for a number of eriods ahead wih oher variables held consan. Examinaion of he imulse resonse hels race he effecs of moneary shocks on curren and fuures value of coffee and cocoa rices wih and wihou he assumion money neuraliy. The resonses funcions are obained using he generalized imulse resonse echnique of Pesaran and Shin (1998), which is available in he Eviews sofware. Pesaran and Shin adaed he Cholesky orhogonalizaion echnique o obain imulse resonse funcions ha are indeenden of he variables ordering, and herefore, giving a unique dynamic behavior. The Cholesky decomosiion echnique allows one o single ou he individual shock effecs when he elemens in he residuals covariance marix are conemoraneously correlaed. Because he imulse resonse funcion o a sandard deviaion of money suly innovaion on he roical commodiies rices are nearly analogous wih and wihou he money neuraliy assumion, only he resonse funcion for he resriced model is resened. The dynamic resonses of he agriculural and indusrial good rices o one
sandard deviaion moneary shock are resened from figure 8 o figure 10 9. The resonses funcions confirm he saisical resuls obained reviously. An exogenous one sandard deviaion moneary suly shock has a long lasing, volaile, negaive hen osiive imac on each he roical commodiy rice. The iniial imac of money suly on each he agriculural rice is a negaive jum ha srech over hree, four, and five monhs, resecively for cocoa, robusa coffee, and arabica coffee. The rices hen rebound afer reaching a minimum level and increase seadily oward a osiion long-run equilibrium. A longer-erm imac of he money suly shock on he roical commodiies rices is relaively small, less han 2.00% in each case, bu he moneary shock a ersisen effec on commodiy rice. For examle, none of he resonses funcions reaches a sable level rior o a 50-monh forecas horizon. The resonses of he indusrial good rice o a moneary shock are clearly negligible in he hree models. Thus, he agriculural commodiy rice resonds faser o money suly shock han he indusrial good rice. Overall, imosing he assumion of money neuraliy does no noably change he dynamic ah of he commodiies rices. Using he case of robusa coffee, for which he hyohesis of money neuraliy was rejeced wih he SRM model, as illusraion, a one sandard deviaion innovaion originaing from he money suly resuls in a negaive dro in rice which reaches a minimum level a -2.40% afer four monhs (See Figure 11 for an amlified lo). Then a reversal occurs, he resonse become osiive afer 29 monhs and ke increasing unil a new long-run osiive equilibrium level is reached a 1.36%. To assess he ercenage variaion in he roical good rices ha originaes from he oher variables, heir forecas error is decomosed. The mulivariae covariance marix in VAR being in general non-diagonal, he Cholesky lower riangular ordering is used o comue he variance decomosiion. The ordering imosed goes from he money suly o he agriculural rice. Secifically, he money suly is followed by he ineres 9 Money suly forecas error are assimilaed o money suly innovaion
rae, which is followed by he exchange rae, which is followed by he indusrial rice, which is followed by he agriculural rice. The variance decomosiion resuls under money neuraliy are resened in ables 11, 12, and 13. In all cases, he variaion in each of he variable forecas error is exlained by heir own shock a 80% or higher wih a 12 monhs forecas horizon. This suggess ha he dynamics of cocoa or coffee rices are mainly funcions of heir resecive marke fundamenals such as suly and demand condiions. A he 12-monh forecas horizon, he Unied Saes moneary insrumens, such as he money suly and he ineres rae, joinly affec coffee rices more han hey affec cocoa rice. The conribuion of he Unied Saes moneary insrumens in he variance of he roical cro is higher for arabica han for robusa coffee rice and cocoa rice. Wih a 24-monh forecas horizon, he join conribuion money suly and ineres exceed 10% for each of he agriculural rice. Summary and Conclusion In his aer, he vecor error correcion mehodology is used o analyze he imacs of Unied Saed moneary olicy on coffee and cocoa rices using a modified conceual and analyical framework of Saghaian, Reed, and Marchan. The resuls of his analysis shed ligh on overshooing of agriculural rices for wo imoran cros for less develoed counries, cocoa and coffee. The economeric evidence oins oward overshooing of cocoa and coffee rices in he shor-run. The rice of cocoa and coffee resond faser o money suly shock han he indusrial good rice. Thus, arabica coffee rice, robusa coffee rice and cocoa rice aear more flexible han indusrial good rice. The roical cro rices also resond faser han he exchange rae, which is also raded in ransaren aucion marke. The imulse resonse funcions of arabica coffee rice, robusa coffee rice and cocoa rice o money suly change are marked by heir iniial negaive imac and heir ersisen osiive long-run effec. The imulse resonses funcions signal ha he long-run imacs of money suly sock on he roical beverage cro rices are ermanen. The dynamic ahs of he roical commodiy rices
obained wih and wihou he money neuraliy assumion were very similar. The variance decomosiion resuls indicae ha sochasic changes in he fundamenal marke condiions each roical cro are he main sources of volailiy. However, Unied Saed moneary olicy insrumens are found o exlain an economically significan roorion of he variaion in he forecas error of coffee and cocoa rice. The flexibiliy and overshooing of he roical cros rices have imoran imlicaion for develoing counries. In he 1970 s, inernaional commodiy agreemens and markeing boards have unsuccessfully aemed o sabilize he rices of coffee and cocoa rice. Nowadays, marke-based insrumens such fuures and oions hedging, are advocaed as efficien and effecive alernaive o miigae he rice insabiliy of hese roical beverage cro. Besides of challenges such ransacion coss, exchange rae risk, and he basis risk, arising when one aem o use markes locaed in New York and London o hedge her ouus rice from a develoing counry, here is a need o accoun for he role of changing macroeconomic olicy.
References: Dornbusch, R. Execaions and Exchange raes Dynamics. Journal of Poliical Economics 84 (1976): 1161-76. Enders W. Alied Economeric Time Series Second ediion 2003, John Wiley, and Sons, Inc. Eviews User s Guide Version 4 for Windows 9x, NT 4.0, 2000 and XP. Quaniaive Micro Sofware Irwin CA 2000. Fisher L., P. L. Fackler and D. Orden. Long-run Idenifying Resricion for an Error Correcion Model of New Zealand Money, Prices and Ouu. Journal of Inernaional Money and Finance 14 (1995): 127-147. Frankel, A. J. Execaion and Commodiy Price Dynamics: The Overshooing model American Journal of Agriculural Economics 68 :( 1986): 343-348. Johansen, S. The Role of he Consan and Linear Terms in Coinegraion Analysis of Nonsaionary Variables. Economerica Reviews 13 (1994): 205-229. Johansen, S. Likelihood-based Inference in Coinegraed Vecor Auoregressive Models. Oxford Universiy Press 1995. Johansen, S. The Inerreaion of Coinegraing Coefficiens in he Coinegraed Vecor Auoregressive Model. Working Paer No 14, Dearmen of Theoreical Saisics, Universiy of Coenhagen, Oc. 2002. Johansen, S. and K. Juselius. Tesing Srucural Hyohesis in a Mulivariae Coinegraion Analysis of he PPP and UIP for he UK. Journal of Economerics 53 (1992). Pesaran, M.H. and Y. Shin. Generalized imulse resonse analysis in linear Mulivariae models Economics Leers 58 (1998):17-29.
Saghaian, S. H., M. Reed and M. A. Marchan. Moneary Imacs and Overshooing of Agriculural Prices in an Oen Economy. American Journal of Agriculural Economics 84 1 (2002): 90-103.
Tables and figures Table 1. Uni Roo Tess Philis-Perron Augmened Dickey-Fuller Level Firs Level Firs Robusa -3.46-8.72-1.40-12.46 Arabica -1.85-7.10-2.10-12.90 Cocoa -2.04-7.76-2.04-13.14 M1-3.19-6.44-2.78-18.56 PPI -0.94-7.12-1.59-12.76 Ineres rae -0.37-5.66 0.23-10.99 Exchange rae -1.27-7.30-1.26-11.49 reresens he difference oeraor MacKinnon 1% criical values for rejecion of hyohesis of a uni roo is -3.45 Table 2: VAR Lag lengh selecion crieria Variables Arabica PPI Exchange rae Ineres rae M1 Lag LogL LR FPE AIC SC HQ 0 539.19 NA 0.00-3.99-3.92-3.96 1 3272.62 5344.47 0.00-24.20-23.79* -24.04 2 3332.54 114.91 0.00-24.46-23.72-24.16 3 3387.57 103.49* 1.31E-17* -24.68* -23.61-24.2* 4 3401.30 25.32 0.00-24.60-23.19-24.03 5 3417.88 29.94 0.00-24.54-22.79-23.84 * indicaes lag order seleced by he crierion LR: sequenial modified LR es saisic (each es a 5% level) FPE: Final redicion error AIC: Akaike informaion crierion SC: Schwarz informaion crierion Table 3: VAR Lag lengh selecion crieria Variables Robusa PPI Exchange rae Ineres rae M1 Lag LogL LR FPE AIC SC HQ 0 508.65 NA 0.00-3.76-3.69-3.73 1 3310.93 5479.08 0.00-24.48-24.08* -24.32 2 3371.82 116.78 0.00-24.75-24.02-24.46 3 3424.56 99.18 9.96E-18* -24.96* -23.89-24.53* 4 3440.52 29.43 0.00-24.89-23.49-24.33 5 3457.00 29.77 0.00-24.83-23.09-24.13 * indicaes lag order seleced by he crierion LR: sequenial modified LR es saisic (each es a 5% level) FPE: Final redicion error AIC: Akaike informaion crierion SC: Schwarz informaion crierion HQ: Hannan-Quinn informaion crierion
Table 4: VAR Lag lengh selecion crieria Variables Cocoa PPI Exchange rae Ineres rae M1 Lag LogL LR FPE AIC SC HQ 0 625.19 NA 0.00-4.63-4.56-4.60 1 3364.78 5356.50 0.00-24.89-24.48* -24.72 2 3425.70 116.85 0.00-25.15-24.42-24.86 3 3480.24 102.56* 6.57E-18* -25.37* -24.30-24.94* 4 3490.62 19.13 0.00-25.27-23.86-24.70 5 3510.74 36.34 0.00-25.23-23.49-24.53 * indicaes lag order seleced by he crierion LR: sequenial modified LR es saisic (each es a 5% level) FPE: Final redicion error AIC: Akaike informaion crierion SC: Schwarz informaion crierion HQ: Hannan-Quinn informaion crierion Table 5: Summary of all Five ses of Johansen assumions Daa Trend: None None Linear Linear Quadraic VAR No Inerce Inerce Inerce Inerce Inerce EC No Trend No Trend No Trend Trend Trend Variables ARABICA PPI EXCH T3BILL M1 Trace 2 2 2 2 1 Max-Eig 2 2 2 2 1 Variables ROBUSTA PPI EXCH T3BILL M1 Trace 2 2 2 1 1 Max-Eig 2 2 1 1 1 Variables COCOA PPI EXCH T3BILL M1 Trace 2 2 2 2 1 Max-Eig 2 2 2 1 1 Number of Coinegraing Relaions a 5% significance level Max-Eig=maximum eigenvalue VAR=Vecor auoregressive and EC=error correcion Table 6: Coinegraion es using he Quadraic deerminic rend Trace saisic Model Robusa + Arabica + Cocoa + 5 % CV 10% CV None 89.38** 93.70** 95.19** 77.74 73.4 A mos 1 42.44 51.32* 49.56 54.64 50.74 A mos 2 16.63 21.23 23.45 34.55 31.42 A mos 3 5.54 7.01 7.89 18.17 16.06 A mos 4 1.84 3.36 3.21 3.74 2.57 Maximum eigenvalue saisic Model Robusa + Arabica + Cocoa + 5 % CV 10% CV None 46.94** 42.38** 45.63** 36.41 33.74 A mos 1 25.81 30.08* 26.11 30.33 27.76 A mos 2 11.09 14.22 15.57 23.78 21.53 A mos 3 3.70 3.65 4.68 16.87 14.84 A mos 4 1.84 3.36 3.21 3.74 2.57 (**) and (*) denoe significance a 5% and 10% CV is for criical value from Oserwald-Lenum (1992) + denoes he US variables, PPI, EXCH, T3BILL, M1
Table 7. Long-Run Parameers wih SRM Model PPI EXCH T3BILL M1 TREND C ARABICA 2.67 2.87** 1.22** 0.99 0.013014 14.21853 (4.45) (1.18) (0.29) (1.29) PPI EXCH T3BILL M1 TREND C ROBUSTA -0.80 7.46** -1.81** 6.20** -0.03-60.57 (6.82) (1.78) (0.44) (1.96) PPI EXCH T3BILL M1 TREND C COCOA -1.93 4.06** -1.16** 2.43** -0.01-14.94 (3.45) (0.91) (0.21) (0.98) ( ), **, and C denoe he sandard error, he significance a 5% and he inerce Table 8. Error Correcion wih SRM Model ARABICA PPI EXCH T3BILL M1 ε arabica,-1-3.22%** 0.00% 0.55%** 0.51% -0.59%** (0.009) (0.001) (0.002) (0.006) (0.001) ROBUSTA PPI EXCH T3BILL M1 ε robusa,-1-1.86%** 0.04% 0.45%** 0.13% -0.29%** (0.004) (0.00) (0.001) (0.003) (0.001) Cocoa PPI EXCH T3BILL M1 ε cocoa,-1-1.79** 0.08% 0.89%** -0.40% -0.71%** (0.008) (0.001) (0.002) (0.007) (0.003) ( ) and ** denoe he sandard error and he significance a 5% Table 9. Long-Run Parameers wih Resriced Model PPI EXCH T3BILL M1 TREND C ARABICA 1 2.58** 0.86** 1-0.01-15.83 (0.69) (0.26) LR es for M1 = ARABICA = PPI = 1 chi-square = 0.069 Probabiliy =0.97 PPI EXCH T3BILL M1 TREND C ROBUSTA 1 3.31** -1.37** 1-0.02-18.09 (0.88) (0.32) LR es for M1 = ROBUSTA = PPI = 1 chi-square = 4.978 Probabiliy = 0.08 PPI EXCH T3BILL M1 TREND C COCOA 1 3.15** -1.05** 1-0.01-15.49 (0.49) (0.21) LR es for M1 = COCOA = PPI = 1 chi-square = 1.29 Probabiliy = 0.52 ( ), **, and C denoe he sandard error, he significance a 5% and he inerce
Table 10. Error Correcion wih SRM resriced Model ARABICA PPI EXCH T3BILL M1 ε arabica,-1-3.41%** -0.01% 0.59%** 0.52% -0.63%** (0.006) (0.001) (0.002) (0.006) (0.001) ROBUSTA PPI EXCH T3BILL M1 ε robusa,-1-2.57%** 0.02% 0.52%** 0.81% -0.44%** (0.006) (0.004) (0.001) (0.003) (0.001) Cocoa PPI EXCH T3BILL M1 ε cocoa,-1-2.43%** 0.09% 0.95%** 0.13% -0.94%** (0.009) (0.006) (0.002) (0.008) (0.002) ( ) and ** denoe he sandard error and he significance a 5% Table 11: Variance Decomosiion for Arabica Period ARABICA PPI EXCH T3BILL M1 1 96.92% 0.02% 0.46% 1.22% 1.38% 2 96.02% 0.07% 0.93% 1.53% 1.45% 3 94.91% 0.16% 0.63% 2.35% 1.95% 4 93.15% 0.18% 0.51% 2.74% 3.42% 5 91.32% 0.29% 0.41% 3.46% 4.52% 6 89.88% 0.44% 0.38% 4.45% 4.85% 7 88.63% 0.55% 0.34% 5.43% 5.05% 8 87.64% 0.68% 0.31% 6.17% 5.20% 9 87.05% 0.81% 0.29% 6.59% 5.26% 10 86.75% 0.88% 0.27% 6.81% 5.29% 11 86.63% 0.92% 0.27% 6.86% 5.32% 12 86.62% 0.96% 0.29% 6.81% 5.32% 13 86.66% 1.00% 0.35% 6.68% 5.32% 14 86.72% 1.04% 0.44% 6.51% 5.29% 15 86.78% 1.08% 0.56% 6.33% 5.26% 16 86.81% 1.11% 0.73% 6.13% 5.21% 17 86.80% 1.14% 0.95% 5.95% 5.16% 18 86.73% 1.18% 1.21% 5.79% 5.09% 19 86.59% 1.21% 1.52% 5.66% 5.03% 20 86.37% 1.24% 1.88% 5.56% 4.95% 21 86.06% 1.26% 2.29% 5.51% 4.87% 22 85.68% 1.29% 2.74% 5.50% 4.79% 23 85.21% 1.31% 3.22% 5.55% 4.71% 24 84.65% 1.34% 3.75% 5.64% 4.62%
Table 12: Variance Decomosiion for Robusa Period ROBUSTA PPI EXCH T3BILL M1 1 98.43% 0.00% 0.27% 0.23% 1.07% 2 97.46% 0.03% 0.13% 0.41% 1.95% 3 96.41% 0.06% 0.30% 0.58% 2.66% 4 94.91% 0.08% 0.30% 0.89% 3.82% 5 94.12% 0.11% 0.24% 1.18% 4.35% 6 93.81% 0.15% 0.20% 1.39% 4.46% 7 93.60% 0.17% 0.19% 1.48% 4.57% 8 93.52% 0.18% 0.20% 1.45% 4.65% 9 93.56% 0.19% 0.25% 1.35% 4.65% 10 93.61% 0.20% 0.33% 1.24% 4.62% 11 93.60% 0.20% 0.46% 1.15% 4.59% 12 93.54% 0.20% 0.62% 1.10% 4.54% 13 93.40% 0.20% 0.83% 1.10% 4.47% 14 93.16% 0.20% 1.09% 1.17% 4.39% 15 92.80% 0.20% 1.38% 1.31% 4.30% 16 92.34% 0.19% 1.72% 1.54% 4.21% 17 91.75% 0.19% 2.10% 1.85% 4.11% 18 91.05% 0.18% 2.52% 2.24% 4.01% 19 90.23% 0.18% 2.97% 2.72% 3.90% 20 89.29% 0.17% 3.46% 3.28% 3.79% 21 88.25% 0.17% 3.97% 3.93% 3.68% 22 87.11% 0.16% 4.51% 4.64% 3.57% 23 85.88% 0.16% 5.07% 5.43% 3.47% 24 84.56% 0.15% 5.65% 6.27% 3.36% Table 13: Variance Decomosiion for Cocoa Period COCOA PPI EXCH T3BILL M1 1 97.86% 0.00% 0.60% 1.01% 0.53% 2 98.75% 0.02% 0.52% 0.41% 0.30% 3 98.31% 0.22% 0.50% 0.51% 0.46% 4 98.52% 0.21% 0.43% 0.47% 0.37% 5 98.76% 0.18% 0.36% 0.40% 0.31% 6 98.88% 0.15% 0.32% 0.38% 0.27% 7 98.82% 0.14% 0.32% 0.49% 0.23% 8 98.58% 0.12% 0.38% 0.71% 0.22% 9 98.20% 0.11% 0.49% 0.99% 0.21% 10 97.67% 0.10% 0.66% 1.37% 0.20% 11 96.97% 0.09% 0.88% 1.84% 0.21% 12 96.14% 0.09% 1.16% 2.39% 0.22% 13 95.19% 0.08% 1.48% 3.01% 0.24% 14 94.13% 0.08% 1.84% 3.68% 0.27% 15 92.98% 0.08% 2.23% 4.41% 0.30% 16 91.76% 0.07% 2.65% 5.17% 0.34% 17 90.49% 0.07% 3.10% 5.96% 0.38% 18 89.17% 0.07% 3.55% 6.78% 0.42% 19 87.83% 0.07% 4.02% 7.61% 0.47% 20 86.48% 0.07% 4.50% 8.44% 0.52% 21 85.12% 0.07% 4.98% 9.27% 0.56% 22 83.76% 0.07% 5.46% 10.10% 0.61% 23 82.42% 0.07% 5.93% 10.92% 0.66% 24 81.09% 0.07% 6.40% 11.73% 0.71%
Figure 8: Imulse resonse o one sandard deviaion of M1 shock using Money Neural Model 0.01 0-0.01 0 20 40 60 80 100 120-0.02-0.03 ARABICA Figure 9: Imulse resonse o one sandard deviaion of M1 shock using he money neural PPI 0.02 0.01 0-0.01-0.02-0.03 0 20 40 60 80 100 120 ROBUSTAS PPI Figure 10: Imulse Resonse o one sandard deviaion of M1 shock using he Money Neural model 0.018 0.013 0.008 0.003-0.002-0.007 0 20 40 60 80 100 120 COCOA PPI Figure 11: Imulse Resonse o one sandard deviaion of M1 shock using he Money Neural model 0.015 0.01 0.005 0-0.005-0.01-0.015-0.02-0.025-0.03 0 10 20 30 40 50 ROBUSTAS