MPRA Munich Personal RePEc Archive Unravelling he underlying causes of price volailiy in world coffee and cocoa commodiy markes Noemie Maurice and Junior Davis UNCTAD 011 Online a hps://mpra.ub.uni-muenchen.de/43813/ MPRA Paper No. 43813, posed 15. January 013 0:05 UTC
Unraveling he underlying causes of price volailiy in world coffee and cocoa commodiy markes Noemie Eliana Maurice and Junior Davis November 01 1
Unravelling he underlying causes of price volailiy in world coffee and cocoa commodiy markes Noemie Eliana Maurice 1 and Junior Davis Absrac: In recen years, Commodiy Dependen Developing Counries (CDDCs) have faced muliple global food, energy and climae crises, compounded by he recen financial and economic crises, which have increased heir vulnerabiliy o excessive price volailiy in commodiy markes. Moreover, srucural vulnerabiliies in mos CDDCs render heir economies more vulnerable o increased commodiy marke urbulence han developed counries, given heir comparaively lower income and high dependence on commodiy expors. This paper aims o empirically examine he paerns and underlying causes of excessive price volailiy for wo major sof commodiies of criical imporance o many of he poores CDDCs: coffee and cocoa. I aims o idenify ineracions, similariies and causaliies beween coffee and cocoa prices on he one hand and, oil and fuures prices on he oher hand. Our analysis of coffee and cocoa hisorical prices shows ha, coffee price volailiy has uneven or varied impac depending on he naure of he marke shock. Oil price spillover effecs on coffee and cocoa markes are also assessed using coinegraion and error-correcion models. Long-run causaliy is found beween oil prices, and coffee and cocoa prices bu, only cocoa has an equilibrium relaionship wih oil in he long-erm. Given he resuls, his sudy proposes some policy recommendaions for managing price risk and addressing regulaion in cocoa and coffee exporing counries. Keywords: Commodiy price volailiy, financializaion, error correcion modelling, coinegraion heory, commodiy dependen developing counries, leas developed counries. JEL classificaion: E30; F4; O11 1 Erasmus School of Economics, Erasmus Universiy, Roerdam, The Neherlands. E-mail: noemiee.maurice@gmail.com UNCTAD, Special Uni on Commodiies Unied Naions. E-mail: junior.davis @uncad.org. The auhors would like o hank Ms. Lauren (SUC) for assisance wih saisical daa collecion.
Conens 1 Inroducion... 5 Overview of he world coffee and cocoa markes... 6.1 Commodiy price volailiy... 10 3 Exploring coffee and cocoa price volailiy... 13 4 Impac of oil spillover effecs and speculaion on coffee and cocoa prices... 1 4.1 Cross commodiy causaliy: Oil vs. Coffee and Cocoa... 4. Coinegraion models and resuls: he effec of speculaion... 6 5 Policy recommendaions and conclusions... 30 6 References... 35 7 Annex... 37 Figure 1 Monhly curren price rends of coffee and cocoa... 7 Figure CDDC Coffee and coca expors as a share of all commodiy expors (%), 009-010... 9 Figure 3. Curren prices of: Arabica, Robusa, Cocoa, and Oil, 1990-011... 10 Figure 4 Coefficiens of variaion for seleced commodiies in he shor and long run, 1960-1970 o 000 010... 1 Figure 5 Percenage variaions in real prices, consumpion and producion, 1990 o 010: (a) coffee and (b) cocoa... 1 Figure 6 Variaion in cocoa, Arabica, Robusa prices vs. oil prices (percen)... 4 Table 1 Main cocoa and coffee exporing counries... 8 Table Specificaion for commodiy prices... 14 Table 3 GARCH (1, 1) ess resuls... 16 Table 4 Wald Tes: Tes of he GARCH hypohesis... 17 Table 5 EGARCH: es resuls for cocoa, Arabica and Robusa... 19 3
Table 6 Arabica, Robusa and cocoa price correlaions in curren and consan prices, 1968-011... 0 Table 7 Granger-causaliy ess resuls... 3 Table 8 Uni roo ess for Arabica Robusa Cocoa fuures prices... 5 Table 9 Ordinary Leas Squares equaions... 6 Table 10 Uni roo ess in levels and firs-difference for Arabica, Robusa and cocoa fuures prices... 7 Table 11 Coinegraion: ADF es on residuals... 8 Table 1 Ordinary Leas Squares equaions... 8 ˆ Table 13 Wald Tes: 1... 9 Table 14 OLS Error Correcion Model... 9 4
1 Inroducion Since 000, Commodiy Dependen Developing Counries (CDDCs) have faced muliple global food, energy and climae crises, compounded by he recen financial and economic crises which have increased heir vulnerabiliy o excessive price volailiy 3 in commodiy markes. Moreover, srucural vulnerabiliies in mos CDDCs render heir economies more vulnerable o increased commodiy marke urbulence han developed counries, given heir comparaively lower income and high dependence on commodiy expors. The World Bank esimaes ha 119 million more people have been pushed ino hunger as a resul of he 008 food crisis. There are now an esimaed 1.0 billion malnourished people worldwide (World Bank 009). Meanwhile, he FAO esimaes ha more han 75 million people were driven ino hunger beween 006 and 010 (FAO 011). The Leas Developed Counries (LDCs) 4 and CDDCs were paricularly harmed by his crisis. The LDCs were paricularly affeced by he 007-008 food crises because he average household spend around 70-80 per cen of heir income on food (UNCTAD 009). Alhough supply and demand fundamenals played a significan role in he food crisis oubreak, many oher facors conribued o he food crisis. For example, large increases in oil prices conribued o rising producion coss and drove food prices higher. The World Bank esimaed ha weakness of he dollar accouned for 15 per cen of he food price increases beween 00 and 008 (Michell 008). Addiionally, over he las decade, major weaher evens such as drough in Russia, excepional fross in Brazil and, excessive rainfall in Canada and Ausralia caused major disrupions o agriculural producion (paricularly for cereals). Price flucuaions are inheren in agriculural 3 Volailiy is a saisical measure of he endency of an asse's price o vary over ime. I is usually capured in he sandard deviaion or variance. 4 Leas developed counries refer o he 48 counries which he Unied Naions recognises as he world s poores and weakes counries, exhibiing he lowes indicaors of social and economic developmen. They have a populaion no exceeding 75 million and a per capia gross naional income (GNI) of less han US$905). See UN-ORHLSS websie: hp://www.unohrlls.org/en/ldc/relaed/59/ (4 January 010). 5
markes parly due o he supply-demand dynamics and he unpredicabiliy of weaher paerns and harves yields. There are also debaes as o he exen o which aciviy in fuures rades and over he couner markes (OTC) for agriculural commodiies impac on his volailiy. Whaever he cause, exreme volailiy in food prices deers producers from making he necessary invesmens for increasing produciviy and producion: his is one of he underlying causes of coninued worldwide food insecuriy. This sudy inends o explore he graviy of he commodiy rade and developmen problemaique vis-à-vis high food, energy prices and volaile markes for he world s mos vulnerable CDDCs. I aims o empirically explore underlying price behavior and volailiy in he coffee and cocoa markes, and also o idenify ineracions, similariies and causaliies beween coffee and cocoa prices on he one hand and, oil and fuures prices on he oher hand. This sudy will firs provide an overview of he world coffee and cocoa markes. Nex, we inroduce he daa employed for use in he empirical analyses. The Generalized Auoregressive Condiional Heeroskedasiciy (GARCH) models for Arabica coffee, Robusa coffee and cocoa are hen esimaed and inerpreed. We hen empirically consider he price-effecs of boh energy and financial producs using Granger-causaliy and coinegraion mehods o explore poenial long-erm rend similariies. Las, we consider he empirical resuls o formulae a few policy recommendaions aimed a reducing risks associaed wih commodiy price volailiy in CDDCs. Overview of he world coffee and cocoa markes Coffee and cocoa are boh ropical commodiies mainly produced in CDDCs and have experienced exreme variabiliy in heir prices over he las 40 years. In fac, coffee and cocoa price variaions have proven very large compared o cereals or mea. In his paper we differeniae beween Arabica and Robusa coffee as hey are differen varieies of coffee and raded on separae exchange markes. Coffee and cocoa have similar long-run price rends (see Figure 1). 6
Figure 1 Monhly curren price rends of coffee and cocoa (1960-011) Source: UNCTADSTAT daabse, accessed July 011. Mos of he producion of hese commodiies is locaed in LDCs and developing counries in Africa, Souh America and Souh Asia (see Table 1 and Figure ). Thus, coffee and cocoa price volailiy is of acue economic imporance for CDDCs. As coffee and cocoa are wo major Sub-Saharan African (SSA) expor crops, hey represen a major source of income for many developing counries ha have a srong commodiyexpor dependence. For example, cocoa crop expors in Ghana during 009-010 accouned for 55 per cen of oal commodiy expors. Similarly, cocoa crop expors provide a livelihood for 5 per cen of he Coe d'ivoire's populaion and 4 per cen of is commodiy expors. During 009-010, in Burundi he share of coffee represened 63 per cen of oal commodiy expors and 30 per cen in Ehiopia (UNCTADSTAT 01, FAO 006). For coffee and cocoa exporing CDDCs, price volailiy is a major cause of concern while i is a relaively minor concern for mos imporing counries. For he former, significan flucuaions in world prices may have dramaic effecs boh a he naional and producer levels as exreme volailiy in prices deers producers from making he necessary invesmens for increasing produciviy and producion. For mos imporing counries, changes in coffee or cocoa prices would mos likely resul in relaively minor changes in consumpion habis. 7
Involving over fify producing counries, of which hiry are imporers, coffee is one of he mos widely raded commodiies. Coffee is a perennial crop ha is produced from he same roo srucure for wo or more years. As a seasonal crop; varying from counry o counry, supply for he mos par is ofen unpredicable. For many developing counry governmens, and he privae secor coffee producion, rade and consumpion is a criical conribuor o socio-economic developmen. Source: FAO (011) Table 1 Main cocoa and coffee exporing counries Cocoa exporing counries Coffee exporing counries Brazil Angola Cameroon Brazil Côe d'ivoire Burundi Dominican Republic Cenral African Republic Ecuador Colombia Gabon Cosa Rica Ghana Coe d'ivoire Malaysia Cuba Nicaragua Ecuador Nigeria El Salvador Papua New Guinea Ehiopia Sierra Leone Gabon Togo Ghana Trinidad and Tobago Guaemala Venezuela Honduras India Indonesia Kenya Liberia Mexico Nicaragua Panama Papua New Guinea Philippines Sierra Leone Tanzania Thailand Timor-Lese Togo Uganda Vienam Yemen The Inerconinenal Exchange (ICE) which is par of he New York Board of Trade (NYBOT) governs he world Arabica price hrough Fuures U.S. Coffee "C" conracs 8
while Robusa coffee has been raded for over weny years on he London Inernaional Financial Fuures Exchange (LIFFE) 5. Figure CDDC Coffee and coca expors as a share of all commodiy expors (%), 009-010 Source: UNCTAD, UNCTADsa (accessed Sepember 01). Noe(s): * CDDCs in Asia does no include Oceania * CDDCs in Lain America (incl. Cenral America, Souh America and he Caribbean) SITC codes: Coffee and coffee subsiues [071]; Cocoa [07]. Primary commodiies, precious sones and non-moneary gold (SITC 0 + 1 + + 3 + 4 + 68 + 667+ 971). Cocoa, alhough produced and expored in smaller volumes, has many similariies wih coffee. Niney per cen of he cocoa producing counries also produce coffee (see Table 1). While primarily consumed in Organizaion for Economic Cooperaion and Developmen (OECD) counries, cocoa is exclusively produced in developing counries; which makes cocoa price volailiy an imporan issue for CDDCs. Cocoa harvess and hus produciviy levels are highly dependen on prevalen weaher condiions. Since 5 The Inernaional Coffee Organizaion (ICO) is he main inergovernmenal organizaion in charge of collecing and sharing informaion on coffee and of esablishing inernaional cooperaion in he coffee secor. In 188, wih is enry ino he Coffee Exchange of New York (laer par of he Coffee, Sugar and Cocoa Exchange), coffee prices became more volaile. The mandaes of he Inernaional Cocoa Organizaion (ICCO) focus on enhancing he economic, social and environmenal susainabiliy of he world cocoa economy. 9
195, cocoa has been raded on he New York Cocoa Exchange before joining he Coffee, Cocoa and Sugar Exchange and laer he ICE, as par of NYBOT 6..1 Commodiy price volailiy Commodiy prices have shown considerable volailiy over he pas decade. 7 The price boom beween 00 and 008 was he mos pronounced in several decades in magniude, duraion and breadh. Moreover, he price decline following he onse of he recen global crisis in mid-008 sands ou boh for is sharpness and for he number of commodiies affeced (UCDR, 01). Since mid-009, and especially since he summer of 010, global commodiy prices have again been seadily rising. (see Figure 3). Figure 3. Curren prices of: Arabica, Robusa, Cocoa, and Oil, 1990-011 (in logarihms) Source: UNCTADSTAT and World Bank Commodiy Price Daa (Pink Shee) (accessed April 011). There are many explanaions for he apparen volailiy in commodiy markes, including he so-called financializaion of commodiies as an asse class. The high prices across a broad range of commodiies -- and he poenial diversificaion benefis of a 6 Cocoa fuures conracs are primarily raded and denominaed in UK pounds. 7 Price volailiy is a measure of price variaion from one period o he nex. 10
wide array of invesmen opporuniies -- has araced speculaive invesors (e.g. hedge funds, commodiy index and exchange-raded-funds) ino commodiy markes. Beween 003 and 008, speculaive invesmen in commodiy indexes was esimaed o have increased from $15 billion o around $00 billion (see UCDR, 01). Long-erm comparisons show ha recen price volailiy is no unprecedened for individual commodiies. 8 For example, oil price volailiy in 008, while remarkable, remained well below is spike of he early 1970s. Therefore, examining he shor-erm consan prices provides a beer insigh wih regard o recen food price developmens. The char below presens he coefficiens of variaion (CV) for various food commodiies and oil. CV (1) The CV (1) connecs he sandard deviaion ( ) o he mean ( ) so ha he mean of he daa is considered allowing for cross-commodiy comparisons. CV is a basic measure of price dispersion; i serves o compare he degree of variabiliy from one daa series o anoher. Long-erm comparisons show ha recen price volailiy is no unprecedened for individual commodiies (see Jacks, Rourke and Williamson, 011). Figure 4 presens he coefficiens of variaion for various food commodiies and oil (for comparison purposes). 9 I shows he long-erm volailiy of commodiies prices using annual consan prices for six commodiies over he period 1960-010 and indicaes ha he more recen price flucuaions during 1990-010 are unexcepional for some commodiies (Calvo-Gonzales, Shankar and Trezzi, 010). The volailiy of coffee prices was similar o ha of mos agriculural producs over he pas 50 years. Peroleum and sugar prices were he mos volaile during he period 1960-010. However, i should be noed ha he volailiy esimaes below do no ake ino accoun rends which could be 8 Jacks DS, O Rourke KH and Williamson JG (011), and Calvo-Gonzales O, Shankar R and Trezzi R (010). 9 The coefficien of variaion is a basic measure of price dispersion; i serves o compare he degree of variabiliy from one daa series o anoher. 11
imporan in he conex of a commodiy super cycle, as for example in he case of real meals prices (Cuddingon and Jerre, 008). More specifically, he magniude of he mos recen upswing of food and meals prices was above he hisorical average, while he magniude of he price rebound for oil was similar o hisorical averages, bu occurred more rapidly (Baffes and Haniois, 010). Figure 4. Coefficiens of variaion for seleced commodiies in he shor and long run, 1960-1970 o 000 010 10 Noe: The coefficien of variaion (raio) is based on annual consan dollar values (000=100). The ime series covers he period 1960 010. Annual variaion in seleced real commodiy prices, by decade. The coefficien of variaion is very sensiive o ouliers hence; for example, he large ampliude of price swings ha occurred during he 1979-1981 financial crisis 11 for a broad range of commodiies may bias he indicaor. Alhough he CV does no reach is 1970-1980 hisorical peak, for mos of he commodiies' volailiy has risen significanly over he las decade. We explore some of hese issues empirically in secions 3 and 4 of he paper. 10 The coefficien of variaion is based on annual consan dollar values (000=100). The ime series covers he period 1960-010. 11 The financial crisis of 1979-1981 had many similariies o he recen global financial crisis of 009-010. For example, he US dollar was falling, inflaion in he USA was approaching 13 per cen and a high level of unemploymen a 13 per cen was exacerbaed by a concomian energy crisis in 1979 which le o rapidly escalaing energy food prices. On commodiy markes, precious meals again became a safe haven for invesors wih gold reaching $850 and silver $50 an ounce. 1
3 Exploring coffee and cocoa price volailiy In his paper, coffee and cocoa price volailiy is empirically invesigaed using GARCHype models (comprising a sample size consiss of 49 observaions). We use logarihmic ransformaions of monhly consan prices of Arabica and Robusa from January 1990 o Sepember 010 (1 monhs*0 years+9 monhs= 49 monhs) 1. For secion 4 of he paper, we use he logarihms of monhly curren prices for Arabica, Robusa, cocoa and oil. Daily fuures prices of Arabica, Robusa and cocoa were colleced from Bloomberg. Monhly averages were compued in order o conduc a causaliy analysis. Cocoa fuures prices are exraced from he London Inernaional Financial Fuures and Opions Exchange (LIFFE) and are convered from UK ( ) pounds serling o US dollars using he monhly average of he Bank of England s spo exchange rae saisics. Table liss he commodiy price series, sources and unis of measuremen uilized in his paper. The deflaor ha is used o compue consan prices from curren price ( Cons an Curren / MUV * 100 ) is he UN Uni Value Index of Manufacured (MUV) goods expors. Food price variaions are ofen large and unpredicable. Greaer price unpredicabiliy and uncerainy abou fuure developmens, ofen leads o higher price risks being borne by producers, exporers, imporers and sock holders who are hen very likely o review heir invesmen decisions. To reduce disrupion in boh coffee and cocoa markes will require an empirically accurae measure of volailiy ha akes ino accoun specificaions relaive o each commodiy and allows he predicion of fuure price developmens. ARCH and GARCH processes defined as "mean zero, serially uncorrelaed processes wih non-consan variances ha are condiioned on pas informaion" (Aradhyula and Ho, 1988) are useful economic analysis ools wih srong forecasing accuracy. 1 The 1990-010 period corresponds o he free marke period on commodiy markes. 13
Source: ICO, ICCO Bloomberg, he World Bank Table Specificaion for commodiy prices Commodiies Period (mm/yyyy) Price Specificaions Source Uni Arabica (A) 01/1990-09/010 Monhly average ICO US /kg consan prices Robusa (R) 01/1990-09/010 Monhly average ICO US /kg consan prices Cocoa (C) 01/1990-09/010 Monhly average ICCO US /kg consan prices Arabica (A) 01/1990-04/011 Monhly average ICO US /kg curren prices Robusa (R) 01/1990-04/011 Monhly average ICO US /kg curren prices Cocoa (C) 01/1990-04/011 Monhly average ICCO US /kg curren prices Peroleum Crude 01/1990-04/011 Monhly average prices Bloomberg $/bbl Of Bren, Dubai and World Bank Wes Texas (A) fuures prices 01/1990-04/011 Daily curren prices Bloomberg US$/lb (R) fuures prices 11/1991-01/009 Daily curren prices Bloomberg US$/MT (C) fuures prices 01/1990-04/011 Daily curren prices Bloomberg GBP/MT GARCH models use pas prices o model and forecas condiional variances. They also allow a wide range of possible specificaions o boh model volailiy and examine volailiy persisence and asymmery in coffee prices over ime. Any GARCH model assumes ha prices have a ime-varying (non-consan) variance which means ha in some periods, markes are more volaile han in ohers. The objecive of his secion of he paper is o characerize he condiional variance of Arabica, Robusa and cocoa price series. Le us assume ha he Arabica prices series auoregressive process: A P 13 are generaed by he P A c p i1 P i A 1 (4.1) 13 R P sands for Robusa price and C P for cocoa price. 14
While he condiional variance is presened in a GARCH (1, 1) model wih a consan, pas informaion abou volailiy ( 1 ) and pas forecas variance ( h 1 ): 1 ~ N(0, h ) h 1 h 1 (4.) The condiional variance h of he informaion se available a ime -1 1 considers varying confidence inervals of volailiy. Table 3 presens univariae GARCH (1, 1) parameers for he mean and he variance equaions of boh coffees and cocoa. The preferred regression has he AR order p and he moving average (MA) order q ha minimize he Schwarz informaion crierion (SIC) 14. In addiion, regressions are esimaed using a range of {1; 5} for p and {0; 5} for q and he combinaion of p and q wih he lowes SIC is he preferred model. 14 The Schwarz informaion crierion (SIC) is a crierion for model selecion among a finie se of models. I is based, in par, on he likelihood funcion, and i is closely relaed o Akaike informaion crierion (AIC). Alhough he original derivaion assumes ha he observed daa is independen, idenically disribued, and arising from a probabiliy disribuion in he regular exponenial family, SIC has radiionally been used in a much larger scope of model selecion problems. 15
Table 3 GARCH (1, 1) ess resuls Cocoa: AR (1) Arabica: AR (1) Robusa: ARMA (1,1) Cocoa c 1 p 1 A c R 1 p 1 c 1 p 1 1 1 Condiional variance h h 1 1 Cocoa Arabica Robusa ARMA c 4.940 5.60 4.610 (0.158) (0.13) (0.06) φ 0.976 0.969 0.97 (0.011) (0.015) (0.014) γ 0.41 (0.075) GARCH δ 0.001 0.00 0.00 (0.000) (0.001) (0.001) α 0.47 0.178 0.144 (0.080) (0.067) (0.067) β 0.6 0.505 0.55 (0.11) (0.10) (0.44) α+β 0.870 0.68 0.669 Schwarz -.74 -.64 -.418 Adjused R^ 0.947 0.940 0.968 The Arabica resuls show ha AR(1) is he specificaion ha maximizes he bes qualiy of fi. Robusa on he oher hand is bes approximaed wih he model ARMA(1,1) and, boh he AR and he MA coefficiens are significanly differen from 0. Finally, cocoa is bes approximaed by an AR(1) model. All he coefficiens in Table 3 are significan and he regressions show a high adjused R-squared, meaning ha he esimaed parameers of he condiional mean have a srong explanaory power of hisorical price movemens. Given he high adjused R-squared, i would seem ha GARCH models perform well a modelling condiional variance. Noneheless, his is no guaranee ha he GARCH process is a saisically valid improvemen over he AR(MA) process (Aradhyula and Hol, 1988). Therefore, we es he GARCH hypohesis ha he condiional variances are in fac, no consan using he following hypohesis: H 0 : 0, 0 H1 : 0or 0 16
A Wald es of he join significance of α and β is conduced for he hree commodiies in Table 4. The saisics used in a Wald es is he Chi-squared; if he p-value of he chisquared exceeds he significance level (0.05) he null hypohesis of saionariy in he volailiy canno be rejeced. Resuls indicae ha p-values of he Chi-squared disribuions of Arabica, Robusa and Cocoa are all equal o 0, hus, we rejec he null hypohesis of saionariy in he condiional forecas variances; GARCH is an improvemen over he AR process for he hree ropical commodiies. Table 4 Wald Tes: Tes of he GARCH hypohesis : 0, 0 Wald Tes: H 0 Tes Saisic Value df Probabiliy Equaion: COCOA_GARCH F-saisic 53.76003 (, 43) 0.000 Chi-square 107.501 0.000 REJECT Equaion: ARABICA_GARCH F-saisic 31.58837 (, 43) 0.000 Chi-square 63.17674 0.000 REJECT Equaion: ROBUSTA_GARCH F-saisic 15.88593 (, 4) 0.000 Chi-square 31.77186 0.000 REJECT From he GARCH analysis, i is possible o infer ha shocks in prices are refleced in volailiy, bu one migh also consider how changes in variabiliy evolve when shocks are posiive or negaive. Undersanding volailiy in response o posiive or negaive shocks is crucial for CDDC producers so hey can predic fuure volailiy in commodiy prices wih more accuracy and hus, improve he esimaion of fuure revenue sreams. Also, Nelson (1991) and Schwer (1989) mainain ha sock volailiy is higher during recessions and financial crisis. We aemp o model how changes in variabiliy evolve when shocks are posiive or negaive by inroducing symmery or leverage effecs in he variance o GARCH models. The Exponenial Generalized Auoregressive Condiional Heeroskedasiciy (EGARCH) is used o esimae he logarihm of condiional variance in order o deermine wheher or no he observed volailiy reacs asymmerically o good and, or bad news. Good news in he case of a commodiy migh be favourable weaher forecass for coffee and cocoa crops or policies ha promoe agriculural developmen and growh; whils bad news may for example be a naural disaser or calamious weaher even (hurricane, ornado, flooding ec) or for example sharp rises 17
in oil prices. In order o assess his for cocoa and coffee we esimae he following EGARCH: log( h ) 1 1 h 1 h 1 1 log( h 1 ) (4.3) In his model he effecs of residuals is exponenial and no quadraic. The asymmery is measured by he coefficien ; if i is negaive and significan, as for many financial asses, here is posiive asymmery and negaive price shocks have a sronger impac on price volailiy han posiive shocks. The impac of posiive shocks (good news) is measured by ( 1 ) h 1 whereas he impac of negaive shocks is capured by ( 1 ) h 1. The hypohesis esed wih he EGARCH model is he following: H H 0 : 0 : 0 0 The resuls in Table 5 show he EGARCH is preferred for cocoa, Arabica and Robusa regressions wih regard o he SIC. Resuls show ha none of he asymmeric coefficiens are negaive and, only for cocoa is approximaely equal o zero ( =0.035) meaning ha, posiive and negaive shocks have approximaely he same impac on is volailiy. In addiion, he GARCH (1, 1) model has a smaller SIC han he EGARCH model and hus, cocoa volailiy is beer approximaed wih he asymmery specificaion. On he oher hand, he asymmery coefficiens for arabica and robusa are large and significan: for arabica, 0. 4, and for Robusa 0. 351 and, boh p- values are equal o zero. The SIC indicaes ha he EGARCH describes he volailiy in world coffee prices beer han he GARCH (1, 1). Posiive shocks have a more prominen effec on he observed volailiy han negaive shocks. An empirical examinaion of he varying volailiy of coffees and cocoa enables he esimaion of bes fi for he modelling of hese hree commodiies. In he case of cocoa, prices follow an auoregressive process of order one AR(1) and is condiional variance is a GARCH (1,1) process. Arabica and robusa prices follow an ARMA model of order 18
p=4, q= for arabica and p=1, q=1 for robusa. Boh coffees condiional variances are beer esimaed wih he EGARCH model. Cocoa: AR (1) Arabica: ARMA (4, ) Robusa; ARMA (1, 1) Table 5 EGARCH: es resuls for cocoa, Arabica and Robusa Cocoa c 1 p 1 A c p R 1 c p p p 1 3 3 4 4 1 1 1 p 1 1 1 EGARCH: log( h ) 1 1 h 1 h 1 1 log( h 1 ) Coefficien Cocoa Arabica Robusa ARMA c 4.911 5.410 4.747 AR 0.139 0.85 0.58 0.974 1.48 0.980 MA EGARCH 1 0.010 0.075 0.010 - -1.048 - - 0.096 - - 1.037-3 4 1 1 * Noe: Only Cocoa coefficien is significanly equal o 0. - 0.080 - - -0.69 - - 0.069 - - -0.088 0.3-0.09 0.067-0.931 - - 0.03 - -.073-3.178 -.308 0.710 0.574 0.777 0.54-0.036 0.015 0.135 0.141 0.146 0.035* 0.4 0.351 0.090 0.104 0.086 0.71 0.40 0.579 0.117 0.110 0.138 SIC -.71 -.80 -.466 Alhough he price correlaions beween he hree commodiies is very high (0.8 in he long-run) (see Table 6), specificiies in erms of heir price volailiy are less clear. Volailiy, expressed by he condiional variance of he price series, is modelled wih differen feaures for arabica, robusa and cocoa, and suggess ha here may be persisence in volailiies and ha he price series are bes esimaed wih a varying 19
variance. We find differen resuls for each of he hree ropical commodiies. The price model AR(1) is used for he cocoa price series, Robusa's prices are modelled wih ARMA(1,1) process and, Arabica prices follow a ARMA(4,) process. The condiional variance definiion follows an EGARCH process wih similar coefficiens and a posiive and significan for boh coffees, which suggess ha, heir volailiy is more affeced by posiive shocks in prices han by negaive price shocks. Moreover, a large increase in oil prices (considered a negaive shock) will have a lower impac on coffee price variabiliy han a seep decline in oil prices (posiive shock) of a similar magniude. Cocoa, on he oher hand does no show any asymmeric paern in is varying volailiy. Thus, in a world of high oil prices, coffee price volailiy is no as excessive as in a conex of low oil prices; whils cocoa price volailiy is largely unchanged. Table 6 Arabica, Robusa and cocoa price correlaions in curren and consan prices, 1968-011 SHORT RUN: curren prices 1968-1990 Cocoa Arabica Robusa 56 obs. Cocoa - Arabica 0.84 - Robusa 0.90 0.96-1990-011 Cocoa Arabica Robusa 56 obs. Cocoa - Arabica 0.6 - Robusa 0.36 0.77 - SHORT RUN: consan prices 1990-010 Cocoa Arabica Robusa 49 obs. Cocoa - Arabica 0.9 - Robusa 0.09 0.76 - LONG RUN 1960-010 Cocoa Arabica Robusa Cocoa - Arabica 0.908 - Robusa 0.418 0.91 0
4 Impac of oil spillover effecs and speculaion on coffee and cocoa prices This secion addresses wo of he main underlying causes of coffee and cocoa price volailiy. Commodiy price variabily mainly resuls from changes in heir fundamenals namely, supply and demand. Figure 5 shows ha for non-essenial goods, variaion in fundamenals do no necessarily reflec he exen of he price surges ha have occurred over he pas 0 years. Figure 5 Percenage variaions in real prices, consumpion and producion, 1990 o 010: (a) coffee and (b) cocoa Source: Auhors calulaions based on ICO and ICCO daa accessed July 011. One of he reasons for he disconnecion beween producion and prices in commodiy markes may be explained by he Separaion heorem according o which "when a fuures marke exiss, he opimum producion of he firm does no depend upon he (subjecive) disribuion of he random price nor upon he firm's aiude oward risk" (Broll and Zilcha, 199). Thus whenever a fuures marke is available, he price and producion of he commodiy may grow independenly. Therefore, we do no dwell upon an empirical analysis of he fundamenals for coffee and cocoa, bu raher focus on wo exernal drivers of hese commodiy prices namely, he energy secor represened by crude oil prices and he financial secor which is refleced by fuures prices. In his 1
secion, all he commodiy prices are denominaed in curren dollar prices as only curren prices are raded in he financial markes 15. Barnard (1983) highlighed he poenial for fuels o be disrupive o agriculural commodiy prices. Aciviies such as: planing, he applicaion of ferilizer, harvesing, sorage and ransporaion require an imporan amoun of diverse fuels; he mos usual being crude oil, coal, gas, and more recenly biofuels. Also, i has been argued ha he prices of boh coffees and cocoa are influenced by oil prices (Baffes J. 007), and ha curren prices have been volaile in recen years hence providing raders wih significan rend-following opporuniies (ICE 011). We uilize Granger-causaliy ess o assess he long-erm causaliy links beween oil and commodiies prices while coinegraion mehods are used o assess he long-run relaionship beween cash and fuures prices of cocoa and coffee. 4.1 Cross commodiy causaliy: Oil vs. Coffee and Cocoa In sub-saharan Africa, cocoa is mainly grown by smallholder farmers ( 1 hecare) and ofen on a subsisence basis (ITC, 001). Larger cocoa planaions exis in Brazil, Ecuador and Malaysia. Alhough cocoa is paricularly sensiive o weaher condiions and diseases ha may negaively affec producion, relaively lile ferilizer is uilized (FAO 006). On he oher hand, coffee producion is increasingly mechanized and uses various chemical ferilizers (e.g. nirogen, poassium ec.) which are by-producs of he peroleum indusry. Here, we only consider he indirec effec of ferilizers prices on coffee and cocoa prices hrough he oil price. Fuels are also required for sorage and ransporaion hus direcly enhancing he poenial ransmission effec of oil prices on coffee and cocoa prices. Graph 8 (Annexes), shows ha coffee and cocoa price changes were ofen preceded by variaions in he oil price of a similar magniude over he pas fify years. Therefore, we aim o deermine wheher causaliy beween oil prices and, coffee and cocoa prices holds in he long-run considering he ime-horizon: 1990-010 15 However, consan dollar prices provide a beer fi for esimaing hisorical volailiy.
and hen, wheher a similar rend beween oil and, cocoa and coffee is empirically observed. Firs, we conduc Granger causaliy ess 16 for crude oil, Arabica, Robusa, and cocoa using large lag lenghs in order o accoun for a long adjusmen period of he commodiies prices o variaions in he oil price, he resuls of which are presened in Table 7. Table 7 Granger-causaliy ess resuls Null Hypohesis Lags included Observaions F-saisic Prob. LN_OIL does no LN_ARABICA 48 08 1.901 0.003 LN_ARABICA does no LN_OIL 1.15 0.70 LN_OIL does no LN_COCOA 36 0 1.736 0.01 LN_COCOA does no LN_OIL 1.05 0.441 LN_OIL does no LN_ROBUSTA 51 05 1.694 0.01 LN_ROBUSTA does no LN_OIL 1.091 0.349 Source: Annex - Table 1. Table 7 shows ha we canno rejec he hypohesis ha he oil price Granger-causes Arabica, Robusa and cocoa price variabiliy a he 5 percen level (p-values: Prob. > 0.05). However, he oil price is no Granger-caused by Arabica, Robusa or cocoa prices a he 5 percen level. I is imporan o highligh ha he oil-commodiy causaliy conclusions are dependen on he number of lags included. The resuls show ha oil price spillover effecs on Arabica and Robusa ake approximaely 4 years while i akes only 3 years for cocoa; which seems consisen wih observaions oulined in Figure 6. 16 'x is a Granger cause of y if presen y can be prediced wih beer accuracy by using pas values of x raher han by no doing so, oher informaion being idenical' (Charemza and Deadman 199). 3
Figure 6 Variaion in cocoa, Arabica, Robusa prices vs. oil prices (percen) Arabica vs. oil 1.6 1. 0.8 0.4 0.0-0.4-0.8 1964 1969 1974 1979 1984 1989 1994 1999 004 009 Differenced OIL Differenced ARABICA Robusa vs. oil 1.6 1. 0.8 0.4 0.0-0.4-0.8 1964 1969 1974 1979 1984 1989 1994 1999 004 009 Differenced OIL Differenced ROBUSTA Cocoa vs. oil 1.6 1. 0.8 0.4 0.0-0.4-0.8 1964 1969 1974 1979 1984 1989 1994 1999 004 009 Differenced OIL Differenced COCOA 4
The concep of coinegraion enables us o furher deermine he possible relaionship beween he variables. Now ha a long-run causaliy link has been esablished beween oil and beverages, we use coinegraion ess o ascerain he long-run relaionship beween hese variables. Empirically, wo I(1) coinegraed series are defined, herefore if a linear combinaion of boh is saionary I(0); an adjusmen beween hese wo variables prevens errors becoming larger in he long-erm (Balcombe and Davis, 1994). Also, curren coffee, cocoa, and oil prices should follow an I(1) process. The augmened Dickey-Fuller (ADF) ess reveal he presence of uni roos in levels (p-values > 0.05) bu no in firs differences (p-values < 0.05) hence, prices of he sudied commodiies are I(1) (see Table 8). Table 8 Uni roo ess for Arabica Robusa Cocoa fuures prices Fuures Arabica "C" Fuures Cocoa Fuures Robusa Uni roo in firs-differences Lag lengh 1 0 1 -saisic Prob. -saisic Prob. -saisic Prob. ADF saisic -13.451 0.000-1.819 0.000-11.19 0.000 Uni roo in levels Lag lengh 1 0 1 -saisic Prob. -saisic Prob. -saisic Prob. ADF saisic 0.675 0.861 0.78 0.871 0.4 0.755 Criical values: 1% -.574 -.574 -.574 5% -1.94-1.94-1.94 10% -1.616-1.616-1.616 By means of equaion (5.1), Granger coinegraion ess are conduced, generaing he residuals series û and hen, esimaing an ADF uni roo es on hose residuals by means of equaion (5.). Coinegraion of he series implies ha he ADF uni roo es of he residuals û is saionary. C, a c. Oil u, a C, a : Curren price a ime of a : { A, R, Cocoa } p u ˆ, a uˆ a j a uˆ 1,, j, a, a j1 (5.1) (5.) 5
The resuls of equaion (5.1) are presened in Table 1. The repored adjused R- squared provides a firs hin regarding he coinegraion of he variables. In he firs regression, i indicaes ha variaions in cocoa, Arabica and Robusa prices respecively explain 45%, 10% and % of he variaions in oil prices. Tes resuls indicae ha, only cocoa prices are coinegraed wih oil prices a he 5% level. Coinegraion beween oil prices and coffees prices (Arabica and Robusa) is weakly rejeced a he 10% level. This suggess ha alhough coffee producion uses more echnological and pero-chemical ferilizer inpus han cocoa, here is no linear relaionship beween coffee and oil whereas, such a relaionship is observed for cocoa and oil. In fac, cocoa and oil price series may rend ogeher in he long-run. In summary, alhough long-run causaliy from he oil secor o he beverage commodiy secor is a valid assumpion, only cocoa shares he same long-erm rend as oil. Besides, a shor-run analysis confirms he consisency of he long-run equilibrium relaionship beween cocoa and oil prices. As mos coffee and cocoa exporing counries are oil imporing price-akers, here is limied policy space for hem o reduce heir vulnerabiliy o oil price flucuaions, whaever he implicaions for heir commodiy expors. Table 9 Ordinary Leas Squares equaions Mehod: Leas Squares Dependen Variable: LN_COCOA LN_ARABICA LN_ROBUSTA Variable Coef. Sd. Error Coef. Sd. Error Coef. Sd. Error (LN_OIL) 0.368 0.05 0.11 0.037 0.105 0.044 C 3.796 0.087 4.735 0.19 4.539 0.153 Adjused R-squared 0.453 0.11 0.018 4. Coinegraion models and resuls: he effec of speculaion The global economic crises since 008-009 may have alered he naure of he relaionship beween fuures and cash prices of some agriculural commodiies. The 000 deregulaion of financial insrumens (fuures) encouraged speculaors o massively rade commodiies in which hey had no business ineres; and herefore, 6
conribued o he price surges in food and energy secors, desabilizing businesses and producer incomes (Ash e al., 010, Gilber and Morgan 010). In fac, since 1990 cash coffee and cocoa prices and fuures prices have ended o move in a similar direcion, irrespecive of increased speculaion. I could herefore be argued ha fuures markes are quie efficien; as fuures prices and cash prices are convergen and i is also likely ha boh variables are coinegraed. Afer verifying ha fuures prices are I(1) (see Table 10), we conduced Granger coinegraion ess and obained he following resuls (see Table 10 and Table 11) for he equaions (5.3) and (5.4): C, a. F, a u, a C, a : Cash price a ime for commodiy a : { A, R, Cocoa } (5.3) F, a : Fuure price a ime for commodiy a : { A, R, Cocoa } p u ˆ, a uˆ a j a uˆ 1,, j, a, a j1 (5.4) Table 10 Uni roo ess in levels and firs-difference for Arabica, Robusa and cocoa fuures prices Fuures Arabica "C" Fuures Cocoa Fuures Robusa Uni roo in firs-differences Lag lengh 1 0 1 -saisic Prob. -saisic Prob. -saisic Prob. ADF saisic -13.451 0.000-1.819 0.000-11.19 0.000 Uni roo in levels Lag lengh 1 0 1 -saisic Prob. -saisic Prob. -saisic Prob. ADF saisic 0.675 0.861 0.78 0.871 0.4 0.755 Criical values: 1% -.574 -.574 -.574 5% -1.94-1.94-1.94 10% -1.616-1.616-1.616 If he wo price series are I(1) and he linear combinaion of hem is I(0), he variables are said o be coinegraed and hus, bivariae models may be specified o ake ino accoun he linear relaionship beween he wo series in he shor-run. ADF es resuls in Table 11 aes o he rejecion of he null hypohesis of a uni roo in he residuals a he 1% level (Prob. <0.05), hereby fuures series and heir corresponding cash prices 7
series are coinegraed. The coinegraion order (1, 1) and he coinegraing vecor [1, - ˆ ] corresponding o: [1, 0.98] for Arabica, [1, 1.0] for Robusa and [1, 0.95] for cocoa may be posiively acceped (see Table 1). Table 11 Coinegraion: ADF es on residuals Arabica fuures Cocoa fuures Robusa fuures -saisic Prob. -saisic Prob. -saisic Prob. ADF saisic -.789 0.0054-9.139 0.000 -.803 0.005 Criical values: 1% -.574 -.574 -.574 5% -1.94-1.94-1.94 10% -1.616-1.616-1.616 Table 1 Ordinary Leas Squares equaions Dependen LN_COCOA LN_ARABICA LN_ROBUSTA Var.: Variable Coefficien Sd. Error Coefficien Sd. Error Coefficien Sd. Error 0.981 0.006 1.013 0.01 0.95 0.0647 0.0318-0.069* 0.055 0.446 Adjused R- squared 0.989 0.976 0.98 * denoes insignificance a a 5% level 0.0058 0.078 Engle and Granger (1987) noes ha all coinegraion series have an error correcion represenaion. Posiively acceped coinegraion suggess ha an error correcion model (ECM) may be esimaed o assess shor-erm price adjusmens. We esimae he error correcion mechanism wih an unresriced OLS in equaion (5.5): C, a 0 1F, a ( C 1, a. F 1, a ), a (5.5) We replace by is previously compued OLS esimae ˆ so ha C, a, F, a and ( C 1, a ˆ. F 1, a ) are all I (0) (Charemza and Deadman, 1991) and he error is correced ( ~ I (0) ). Given he Wald es resuls (see Table 13), we assume ha ˆ 1, a hence, he Engle Granger equaion is simplified as follow: C, a 0 1F, a ( C 1, a F 1, a ), a (5.6) 8
The Arabica model (see Table 14) suggess ha he predicive power of he model is very high; especially for Arabica and Robusa. Indeed adjused R-squared for Arabica, Robusa and cocoa models are respecively 0.95, 0.90 and 0.70. Wald Tes ˆ Table 13 Wald Tes: 1 Tes Saisic Value df Probabiliy Arabica -saisic.1 54 0.035 F-saisic 4.50 (1, 54) 0.035 Chi-square 4.50 1 0.034 Cocoa -saisic -3.05 54 0.003 F-saisic 9.31 (1, 54) 0.003 Chi-square 9.31 1 0.00 Robusa -saisic -13.04 05 0.000 F-saisic 169.97 (1, 05) 0.000 Chi-square 169.97 1 0.000 Table 14 OLS Error Correcion Model ( C F C, a 0 1F, a 1, a 1, a ), a Dependen Variable : C, a a : Arabica 1 a : Cocoa Variable Coefficien Sd. Error -Saisic Prob. 0-0.001 0.00-0.79 0.466 0.907 0.013 69.790 0.000-0.030 0.018-1.74 0.086 adjused 0 1 a : Robusa 1 R 0.951-0.001 0.003-0.6 0.81 0.800 0.03 4.993 0.000 0.034 0.033 1.018 0.310 adjused R 0.716 0 0.005 0.003 1.445 0.150 0.843 0.01 40.6 0.000-0.059 0.03-1.844 0.067 adjused R 0.89 Despie he low frequency of monhly daa, i is possible o esimae he speed of adjusmen beween fuures and cash prices. An ECM provides a good represenaion of 9
shor-run adjusmens beween cash and fuures markes for Arabica, Robusa and cocoa. Shor-run adjusmens are consisen wih he long-run equilibrium relaionship exising beween cash and fuures series suggesing ha he speed of adjusmen is very fas, and cocoa and coffee fuures markes are reasonably efficien. 5 Policy recommendaions and conclusions Price flucuaions are inheren in agriculural markes parly due o he supplydemand dynamics and he unpredicabiliy of weaher paerns and harves yields. There are debaes as o he exen o which aciviy in fuures rades and over he couner markes (OTC) for agriculural commodiies impac on his volailiy. Whaever he cause, exreme volailiy in food prices deers producers from making he necessary invesmens for increasing produciviy and producion: his is one of he underlying causes of coninued worldwide food insecuriy. Indeed, recen weaher caasrophes, oil price surges, inflaion, declining value of he U.S. dollar and, growing financializaion on fuures exchange markes have grealy led o he unpredicabiliy of food prices and marke developmens. Several inernaional organizaions have invesigaed policy responses in order o miigae he risks associaed wih high prices and volailiy in global food markes. A policy recommendaion pu forward by he G0 17 suggess srenghening he long erm produciviy, susainabiliy and resilience of he CDDCs agriculural secor, hrough enhanced public invesmen and naional food securiy programs. Increasing ransparency in food and fuures markes and, eliminaing domesic rade policies would also reduce rade disorions and markes insabiliies (Saaz and Weber, 011 and, Limao and Panagariya, 003). This paper examined volailiy, oil, and fuures spillover effecs on hree major ropical commodiies: Arabica, Robusa and Cocoa. Volailiy developmens and implicaions were analyzed from he supply-side ha is, exporing LDCs and CDDCs. In his case, large price decreases are simulaneously refleced in he rade balance and in he 17 Policy repors elaboraed by FAO, IFAD, IMF, OECD, UNCTAD, WFP, he World Bank, he WTO, IFPRI, and he UN HLTF (011). 30
longer-erm has a derimenal effec on growh. On he oher hand, sudden price hikes may encourage producers o increase producion and adjus heir invesmen decisions, which may rigger even more insabiliy in he markes. The resuls of he presened GARCH models provide an accurae assessmen of commodiy price volailiy. The condiional variances are found varian over ime due o volailiy clusering 18, hus revering o he mean raher han remaining consan or moving in a monoonic paern over ime, which jusifies he use of a GARCH model. Furher analysis reveals uneven effecs in Arabica and Robusa price volailiies, which, are more affeced by posiive shocks han negaive shocks. A good harves in coffee crops will rigger more volailiy in is price han a bad harves. However, cocoa volailiy reacs symmerically o he marke shocks wheher posiive or negaive. Cocoa price volailiy is eviden, regardless of wheher here is a good or poor harves. This paper considered poenial causaliy and linkages beween he crude oil price and, boh coffees and cocoa prices in he long-run. I appears ha variaions in coffee and cocoa prices follow oil price variaions wih, respecively 4 and 3-year inervals. Neverheless, he hypohesis of a long-run equilibrium relaionship only holds beween oil and cocoa prices meaning ha, srucural changes in he oil price will be direcly refleced in cocoa prices. Baffes (007) shows ha he average price elasiciy for cocoa; was high and significan while he average coffee elasiciy was paricularly low; in shor a 100 per cen variaion in he oil price causes a 49 per cen shif in cocoa prices, bu does no cause a significan variaion in coffee prices. In summary, oil price developmens have no significan effec on coffee price variabiliy in he shor-run. On he oher hand, policy-makers should closely monior oil price surges as hey appear o srongly influence cocoa prices and heir volailiy in boh he shor and long-run. We also examined he relaionship beween Arabica, Robusa and cocoa cash prices and heir corresponding fuures prices. The deregulaion of financial and physical insrumens in 000, along wih he inroducion of new elecronic rading opporuniies in 007 has raised concerns abou efficiency in he coffee and cocoa 18 In conras o he ofen-assumed log-normal disribuion of asse price reurns, i is ofen observed ha periods of high price volailiy follow periods of low volailiy and vice versa. 31
fuures markes. However, in his sudy, he observed coinegraion beween cash and fuures series beween 1990 and 010 suggess ha boh ICE and LIFFE fuures markes are (saisically) unbiased and herefore, serve as price discovery channels for coffee and cocoa secor paricipans. The very shor adjusmen period noiceable beween fuures and cash prices suggess ha, hedging sraegies miigae price risk only if hey are an immediae reacion o marke aciviy. Noneheless, he lack of saisical bias of fuures markes does no necessarily imply a full-hedging of price risk (Broll and Zilcha 199). In fac, he Separaion heorem saes ha unbiased fuures esimaors of he spo prices do no imply ha price risk is enirely avoided. Recen sudies have shown ha major speculaive aciviy has increased price risk for cash marke paricipans, paricularly commercial raders (Schaffni-Chaerjee, 011 and, Schuer, 010). As a consequence of increasing speculaive aciviy, small farmers growing cocoa and coffee in developing counries are even more exposed o price risk, especially as few alernaives o manage price risk are available o hem. Gabre-Madhin (010) and, Forenbery and Zapaa (004) have proposed he creaion of local commodiy exchanges which are more accessible o commercial hedgers (for example; he Ehiopia Commodiy Exchange which reduces he incenives of speculaors by imposing mandaory delivery and higher margins. Such iniiaives may largely reduce price risk and hus, promoe economic sabiliy in many CDDCs. Commodiy producers in developed counries are increasingly relying on hedging o miigae exposure o price volailiy. However, he exen of hedging in developing counries remains limied. A few counries have used marke-based insrumens o miigae he income risks. 19 The main reason for he low use of financial insrumens is he lack of familiariy on he par of boh privae secor operaors (especially farmers and exporers) and, in a few 19 For example, Mexico hedged, via opions, all of is oil sales for 009 in 008 a a srike price of US$ 70 a barrel when he oil price was US$ 100 a barrel. 19 The cos of purchasing opions a US$ 1.5 billion enabled he programme o make a savings of more han US$ 5 billion.. 3
insances, he lack of ineres from governmen officials. Using financial insrumens in hedging requires echnical and managerial experise and an insiuional framework ha ensures adequae reporing, recording, monioring and evaluaing mechanisms. Furhermore, i is also necessary o esablish inernal conrol procedures ha avoid and proec agains speculaive ransacions. 0 Marke-based insrumens can play a fundamenal role in building ailor-made faciliies o address commodiy price insabiliy. However, i is doubful wheher he fuures markes are as suiable for addressing problems emanaing from price variabiliy as hey are for reducing uncerainy in revenue flows. This nowihsanding, fuures markes do allow Governmens o eliminae uncerainy associaed wih variabiliy. Apar from emergency measures designed o assis he mos vulnerable and he longerm measures designed o ackle excessive commodiy price volailiy on he supply side, here is a need o consider how he funcioning of commodiy derivaives markes could be improved in a way ha would enable hose rading venues o beer fulfill heir role of providing reliable price signals o commodiy producers and consumers. In ligh of he vial role of informaion flows in commodiy price developmens, a se of four policy responses o improve marke funcioning should be considered: Firs, greaer ransparency in physical markes would enable he provision of more imely and accurae informaion abou commodiies, such as spare capaciy and global sock holdings for oil, and for agriculural commodiies, areas under planaion, expeced harvess, socks and shor-erm demand forecas. This would allow commercial marke paricipans o more easily assess curren and fuure fundamenal supply and demand relaionships. Second, a beer flow of and access o informaion in commodiy derivaives markes, especially regarding posiion-aking by differen caegories of marke paricipans, would furher improve marke ransparency. In paricular, measures designed o ensure 0 Claasens S (199). How can developing counries hedge heir bes? Finance and Developmen. Sepember 199. 33