Modelling the World Oil Market Assessment of a Quarterly Econometric Model

Transcription

1 Modellng the World Ol Market Assessment of a Quarterly Econometrc Model Stéphane Dées a1, Pavlos Karadeloglou a, Robert Kaufmann b, Marcelo Sánchez a a European Central Bank b CEES, Boston Unversty May Abstract Ths paper descrbes a structural econometrc model of the world ol market ntended to be used to assess ol prce developments and to provde quanttatve analyses of ol related rsk assessments. In ths model, ol demand s explaned by behavoural equatons that relate demand to domestc actvty and the real prce of ol. Ol supply for non-opec producers s derved from a compettve behavour, takng nto account the effect of geologcal and economc varables. Ol prces are defned by a prce rule based on changes n market condtons and OPEC behavour. In partcular, OPEC acts accordng to a co-operatve behavour and ensures the global equlbrum at the prce determned by the prce rule. Insample smulaton results show that the model satsfactorly reproduces past developments n ol markets. Polcy smulatons show that the responses of demand and non-opec supply are rather nelastc to changes n prce. Fnally, although OPEC s assumed to close the model by absorbng any excess n supply or demand, the model shows that OPEC decsons about quota and capacty utlsaton have a sgnfcant, mmedate mpact on ol prce. Keywords: Ol market; Econometrc modellng; Forecastng JEL codes: C51, Q41. 1 Correspondng author. European Central Bank, Kaserstrasse 29, Frankfurt-am-Man, Tel.: , Fax: , e-mal: stephane.dees@ecb.nt. 2 We are grateful to Marc Dean, Conall MacColle, Ivan Odonnat and Anne-Mare Reu and colleagues from Natonal Central Banks for helpful comments on a prevous verson. The vews expressed n ths paper are those of the authors and do not necessarly reflect the poston of the European Central Bank.

2 1 Introducton It s a standard modellng practce to represent the world ol market n terms of a supply-demand equlbrum schedule (e.g. Bacon, 1991, Al Fars, 1991). Stll, the specfc characterstcs of the ol markets make ts modellng a partcularly complex endeavour. More specfcally, although a demand curve that relates quanttes to prces can be consdered to correctly reflect the ol demand condtons, the modellng of supply s extremely dffcult as ol markets are charactersed by the exstence of a cartel (OPEC) along wth the presence of ndependent producers. Moreover, ol prces react n a complex fashon to changes n market condtons and OPEC behavour. These partculartes of the ol market are ncluded n a full-fledged ol prces model, presented n ths paper, whch ncludes a prcng rule n addton to demand and supply schedules for dfferent regons of the world. To model supply, non-opec and OPEC behavours need to be dstngushed. Whle we assume non- OPEC behavour to be compettve, OPEC behavour s however much more complex to defne, as shown also by the extensve lterature on the subject. Some studes have questoned OPEC s ablty to affect real prces (e.g. Belton, 1998), pontng out that the declne n real ol prces snce the early 1980 s concdes wth a reducton n OPEC s market share. Others, n the academc lterature, have examned the nature and goals of OPEC by testng competng hypotheses for dfferent behavours (Grffn, 1985). Among the behavours envsaged, two can be dentfed as corner solutons: a cartel model, where OPEC s prce maker and a compettve model, where OPEC s prce taker. Efforts to choose among these models focus n part on the dentfcaton of the slope of OPEC s supply curve. A negatve relaton between prce and producton s nterpreted as a backward bendng supply curve, whch ndcates that OPEC sets producton based on some type of non-compettve behavour. However, the analyss of causalty presented n Kaufmann et al. (2003) ndcates that OPEC producton 'Granger causes' prces but prces do not 'Granger cause' producton ndcatng that there s no backward bendng supply curve. In other words the functonng of world ol market s n realty between the two extreme cases of a compettve or completely domnant OPEC. As a result, an ncrease n demand can be consdered to be followed by ncreases n both OPEC producton and the world ol prce. The soluton usually retaned to model such a behavour s to estmate ths ntermedary degree of market control wth a prce rule, whch depends on both demand condtons and OPEC varables. Such a model would gve a prce at whch OPEC s ready to act as a swng producer, gven new demand condtons and would nclude market ndcators that reflect the effect of behavour by the domnant producer. The above consderatons are taken nto account n the proposed model. Gven a certan prce, demand determnes the optmal quantty of ol sold. Non-OPEC countres adapt ther producton to ths new prce Page 2 of 24

3 and OPEC acts as a swng producer to equlbrate supply and demand consstent wth the optmal prce/quantty levels (an ncrease n demand rases prce to P 1 * R and the total supply curve moves from S 0 to S 1 leadng to a new producton level at Q 1 * see Fgure 1 ). Fgure 1: Model wth prce rule P D 0 D 1 S 0 S 1 OPEC P 1 * R (rule) P 0 * Q 0 * Q 1 * Ol The model of the world ol market presented here possesses several of the above desrable features. Frst, ol supply for non-opec producers s derved from equatons ncludng the effect of geologcal and economc varables. Second, ol demand s rather comprehensvely modelled through a set of behavoural equatons. Thrd, ol prces are defned by a prce rule based on changes n market condtons and OPEC behavour. Fnally, two dfferent OPEC behavours are envsaged (co-operaton and competton) resultng to two alternatve closures of the model. Ths rest of ths paper s organsed as follows. Secton 2 descrbes the general structure of the model and gves estmaton results for the demand, supply and prce equatons used n the model. Secton 3 presents an assessment of the model n terms of forecast performance and smulaton propertes. The fnal secton summarses the major fndngs. Page 3 of 24

4 2 General Structure of the Model Ol demand The demand for ol for each country/regon s a functon of economc actvty, the real prce of ol and a tme trend representng technologcal developments lnked to energy effcency. Equatons have been estmated for the man tradng partners of the euro area namely Unted States, Japan, Unted Kngdom, Euro area, Swtzerland, other developed economes, non-japan Asa, Transton economes, Latn Amerca and rest of the world. The estmaton sample spans over the perod, except for some countres and regons for whch, due to data constrants, the sample starts later. Although ths may be problematc for the robustness of the results, especally for the transton economes, whose sample perod ncludes only 24 observatons, we have chosen to adopt the same methodology for all equatons n order to get smlar propertes across countres and regons. The general specfcaton of the econometrc equatons used n the model s: (1) DEM POIL Y E tme D P,., where DEM s ol demand n physcal unts for each country/regon, Brent ol prce n USD, E the exchange rate vs-à-vs the USD and Y s real GDP, POIL s the D P s the CPI ndex (the GDP deflator was also used, but dd not yeld statstcally dfferent results). In addton to the real ol prce varable n equaton (1) whch does not nclude taxes and corresponds to prces traded n the futures market-, an alternatve, tax ncludng measure (calculated by the Internatonal Energy Agency) has been used for the Unted States, Japan, the Unted Kngdom and Swtzerland. Although such a specfcaton neglects any substtuton effects between dfferent energy products, t captures the man factors, whch nfluence the demand for ol 3. The equaton has been estmated as an error correcton specfcaton defnng the dynamc adjustment to a long-run equlbrum relatonshp (see Table 1 for estmates and Table 6a n Appendx for detaled 3 As we focus on macroeconomc aspects, we have delberately excluded any dstncton between sectors, although the sectorrelated behavours mght be strongly dfferentated. We have also excluded any forward-lookng varables (e.g. expectatons about ol prces) n order to keep the model as smple as possble n ths frst verson. Page 4 of 24

5 results) 4. Because unt root tests shows that the real prce of ol s I(0) for all countres, ths varable has been excluded from the long-run relatonshp and ncluded only among the dynamc terms. Ths means that the real prce of ol whch dsplays a hgh level of volatlty - appears statonary, mplyng that t does not affect the long-run determnaton of ol demand. The results show that the error correcton term n the ol demand equaton s sgnfcantly dfferent from zero n all countres/regons but Latn Amerca. The sgnfcance of ths term ndcates that the real prce of ol s contegrated wth ts long-run determnants. The long-run elastcty of ol demand to real GDP s lower than one for all countres. It s very close to one n the US and Latn Amerca, less than 0.2 n the UK and Swtzerland, and between ½ and ¾ n all other countres/regons. Ths elastcty of ol demand to real GDP s smaller n the short run than n the long run - thereby contrbutng to a smoother adjustment to changes n the latter varable - n the US, the Euro area, Other Developed Economes, and Transton Economes. Ths elastcty s larger n the short run than n the long run - thereby dsplayng a temporary over-reacton to changes n real GDP n Japan, the UK, Swtzerland, and Non-Japan Asa. Fnally, the short run and long run elastctes are smlar n the cases of Latn Amerca and the Rest of the World. Table 1: Results of ol demand estmatons Adjustment Estmaton Long term coeffcents Short-term dynamcs coeffcent of the ECM perod Real GDP Tme trend Real GDP Real ol prce 5 Unted States :1-2002: Japan :1-2002: Euro area :1-2002: Unted Kngdom :2-2002: Swtzerland :1-2002: Non-Japan Asa :1-2002: Other Dev. Eco :1-2002: Transton eco :1-2002: Latn Amerca :1-2002: Row :1-2002: The results usng the tax-ncludng end-use prce varable (for the cases of Unted States, Japan, Unted Kngdom, Euro area, and Swtzerland) are reported n Table 6b n Appendx. These results are roughly the same as those obtaned usng the real prce of ol excludng taxes. 5 Defned as the ol prce n natonal currency deflated by a domestc prce ndex for each country. CPI has been used as a proxy for domestc prce ndex. It can been shown that the results obtaned wth CPI are more satsfactory than the ones obtaned wth other domestc prce proxes (GDP deflator n partcular). Page 5 of 24

6 Ol supply We dstngush dfferent supply behavour for OPEC and non-opec countres. The former s modelled based on a co-operatve behavour, n whch OPEC matches producton to demand. Non-OPEC producton has had a sgnfcant effect on OPEC s share of the world ol supply and, as a consequence on OPEC s ablty to nfluence prces. Producton from non-opec countres s modelled usng a technque that quantfes the effect of geologcal and economc factors. Non-OPEC supply Although most producers outsde OPEC can be consdered as prce takers and proft maxmsers, economc models of non-opec producton generally have proved unrelable, because of the lack of a smple relaton between real ol prces and producton (Cleveland and Kaufmann, 1991). Indeed, to model non-opec producton, a complex nterrelaton between resource depleton, techncal change, economc ncentves and poltcal consderatons has to be taken nto account. The relaton among these forces s modelled usng a methodology developed by Kaufmann (1991) that combnes the curve fttng technque developed by Hubbert (1962) wth the econometrc model of supply, poneered by Fsher (1981) and called hybrd methodology (Kaufmann, 1995). Ths hybrd methodology s estmated n three steps. Frst a logstc curve s estmated for cumulatve ol producton accordng to the method developed by Hubbert. (2a) Q t n whch Q ( 1 a exp( b( t t0 )) Q s the ultmate recoverable supply of ol, Qt s cumulatve ol producton at tme t, and t 0 s the start date of the analyss. The frst dfference of the logstc curve gves an estmate for the annual rate of producton (Qt). Ths s the Hubbert s bell shaped curve for the producton cycle of a non-renewable resource, whch s called a producton curve. Because the physcal characterstcs of the ol felds do not entrely determne producton, the hybrd methodology also ncorporates the effects of economc and poltcal varables. In the second step, the annual rate of producton generated by the producton curve (Qt) s used as an explanatory varable n a contegratng relaton that s gven as follows: (2b) PROD Q ROIL Dummy t 1 t 2 t 3 4 Asym n whch PRODt s ol producton, ROIL s the real prce of ol, Dummy s a dummy varable that may affect local producton (e.g. the Trans Alaskan Ppelne System), and Asym s a varable desgned to test the symmetry of the producton curve. Asymt s Q tmes a dummy varable that equals one after the peak of the producton curve. As such, the Asym varable can be used only for regons where producton has contnued beyond the peak of the producton curve. t t Page 6 of 24

7 In the thrd step, the short-run dynamcs of the supply equatons are estmated usng an error correcton model that has the followng specfcaton: (2c) PROD t 1 Q ROIL t1 2 t 3 t 4 Asym Equatons (2a)-(2c) are estmated usng a range of assumptons for the start date of the producton curve (t 0 ), the value for Q and the start date of the data sample used to estmate the contegratng relaton. The results reported n Tables 2a and 2b are chosen usng the followng crtera. Frst, the combnaton of t 0 and Q s chosen n order to generate a statonary resdual n equaton (2b). Ths s done to dentfy the form of the producton curve that contegrates wth producton (and the other varables). Next, we retan combnatons n whch the coeffcents have the correct sgn and are statstcally sgnfcant n both the long-term relaton (equaton 2b) and the error correcton model (equaton 2c). Fnally, we chose the combnaton that has the hghest R 2. These last two crtera dentfy an equaton that s lkely to generate an accurate forecast. t Because the geologcal and economc envronments dffer among non-opec countres, equatons are estmated for nne regons: Lower 48 states (US), Alaska (US), Canada, Mexco, Brazl, Non-OPEC Latn Amerca, Europe, Non-OPEC Afrca, and Non-OPEC Asa. Other non-opec regons nclude the Former Sovet Unon, Chna, and non-opec Mddle East. Producton by these regons s forecast exogenously due to the dffcultes assocated wth market structure (Former Sovet and Chna) and/or geographcal dsparty (e.g. non-opec mddle East ncludes non-contguous natons such as Syra, Oman, etc.). Table 2a: Results for equaton (2a) Start date (t 0 ) Q a b R 2 Lower 48 (US) Alaska (US) Canada Western Europe Non-OPEC Asa Non-OPEC Afrca Non-OPEC Latn Amerca Mexco Brazl Page 7 of 24

8 Table 2b: Results for equaton (2b) Lower 48 (US) 0.62 (23.32) Alaska (US) 0.53 (20.26) Canada 0.50 (10.32) Western Europe 0.84 (47.29) Non-OPEC Asa 0.67 (53.19) Non-OPEC Afrca 1.04 (67.17) Non-OPEC Latn Amerca 0.94 (53.93) Mexco 0.15 (9.04) Brazl 0.30 (24.36) Q t ROIL Cost Dummy Asym R (2.69) (2.71) 1.77 (2.45) 6.03 (4.72) 1.30 (2.69) 0.84 (2.59) 7.18 (9.71) 0.79 (2.25) (-7.01) (18.0) (14.55) 0.17 (10.43) (-3.57) 0.37 (14.56) OPEC supply OPEC producton s generally consstent wth a co-operatve behavour and ths s retaned as the closure of the model (see Kaufmann, 1995). Accordng to that, OPEC restrans producton from exstng capacty to match demand. The demand for ol s equal to the dfference between demand and non-opec supply: opec oecd non-opec (3) PROD DEM Stocks - NGLS - PROD - PG oecd n whch Stocks s the level of stocks reported by OECD, NGLS s non gas lqud and PG net processng gans. The model can also adopt the compettve behavour as a closure rule, mplyng that OPEC countres compete not only among themselves but also wth non-opec producers for market share leadng OPEC to ncrease producton to levels that are consstent wth operable capacty. To smulate the lack of cooperaton, the model sets OPEC producton to 95% of capacty: j (3 ) opec opec PROD 0.95* Capacty Ol produced n excess of demand s put nto stocks. Page 8 of 24

9 Ol prce Due to a hgh degree of volatlty, the real prce of ol features s dffcult to model. Startng wth Frankel (1946), several other studes assess the dfferent factors that mply fluctuaton n ol prces, the most mportant beng the behavour of a domnant producer. In partcular, when OPEC became the domnant producer n the early 1970s, prce volatlty ncreased tremendously. Although the effect of behavour by a domnant producer on the volatlty of ol prces s relatvely easy to see, modellers fnd t dffcult to smulate ther effect for two reasons: () the nablty to forecast behavour of the domnant producers; and () the nablty to translate a partcular behavour nto a change n real ol prces (Kaufmann, 1995). We have seen n Secton 1 that the ablty of OPEC to affect ol prces les between that of a cartel and that of a prce taker. Emprcal analyses smulate ths ntermedate degree of control wth a prce rule, whch s a functon that relates supply/demand balance to prce (Gately, 1995). Rather than focus on general supply/demand condtons, prce rules nclude market ndcators that reflect the effect of behavour by the domnant producer. We estmate an econometrc equaton for the real prce of ol from quarterly data, 1986:III through 2000:III, usng the followng specfcaton: ROILt 1DAYSt 2Quotat 3Cheat (4) Q Q Q War t t 4 Caputl t n whch ROIL s the US crude ol mport FOB prce and s measured n 1996 US $ per barrel (Monthly Energy Revew, varous months), Days s days of forward consumpton of OECD ol stocks, whch s calculated by dvdng OECD ol stocks by OECD ol demand (Monthly Energy Revew, varous months), Quota s the OPEC producton quota (mllon barrels per day), Cheat s the dfference between OPEC producton (Monthly Energy Revew, varous months) and OPEC quotas (mllon barrels per day), Caputl s capacty utlsaton by OPEC, whch s calculated by dvdng OPEC producton (mbd) by OPEC capacty (mbd), Q1, Q2, and Q3 are dummy varables for quarters I, II, and III, respectvely, and War s a dummy varable for the Persan Gulf War (thrd and fourth quarters of 1990). Dckey-Fuller tests ndcate that the varables n equaton (4), other than the dummes, are non-statonary. We use the dynamc ordnary least squares (DOLS) estmator developed by Stock and Watson (1993) to estmate the contegratng relaton gven by equaton (4) because t generates asymptotcally effcent estmates of the regresson coeffcents for varables that contegrate, t s computatonally smple, and t performs well relatve to other asymptotcally effcent estmators. The coeffcents estmated by DOLS represent the long run relaton among varables. To examne the short run dynamcs n a second step, we use OLS to estmate an error correcton model (ECM). The sgns on the regresson coeffcents estmated (Table 3) are consstent wth prevous results descrbed by Kaufmann (1995) and Balabanoff (1995). The regresson coeffcent assocated wth Days s Page 9 of 24

10 negatve an ncrease n stocks reduces real ol prce by dmnshng relance on current producton and thereby reducng the rsk premum assocated wth a supply dsrupton. Smlarly, an ncrease n the OPEC quota tends to allevate upward pressure on prces. An ncrease the Cheat varable also tends to reduce prce an ncrease n OPEC producton relatve to ther quota ncreases supply relatve to the demand perceved by OPEC when settng the quota (perceved demand may not be the most mportant or only varable used to set the quota). The sgn on the regresson coeffcent assocated wth Caputl s postve, whch s consstent wth those descrbed by Gately and Kyle (1977) and Kaufmann (1995). The postve sgn ndcates that ncreases n capacty utlsaton tend to ncrease prces. Ths effect s consstent wth OPEC s role as the margnal producer durng the 1986:III 2000:III perod. Durng ths perod, OPEC generally set producton to match the expected dfference between non-opec supply, whch s determned largely by non-opec capacty (as prce takers, non-opec producers generally operate at or near capacty), and demand (and to keep prces wthn a desred range). As demand for ol from OPEC ncreases producton relatve to capacty, utlsaton rates rse, whch sgnals a tghtness n the market. The War varable has a postve effect on prces prces rose after the Iraq nvason of Kuwat n antcpaton of a supply dsrupton, but ths effect dsappeared durng the frst quarter of 1991, when t became apparent that the war would have lttle effect on ol supples from the Persan Gulf. Results of the ECM estmate ndcate that prces do not adjust mmedately to the long-term relatonshp. Regresson results ndcate that the error correcton coeffcent s statstcally sgnfcant (Table 4). Ths result s consstent wth the nterpretaton of equaton (4) as a contegratng relaton n whch the rght hand sde varables ' Granger cause' real ol prces. The pont estmate of the error correcton coeffcent s 0.51, whch ndcates that 51 percent of the dsequlbrum n equaton (4) s elmnated after one quarter. Table 3: Estmates for prce equaton Varables Coeffcents Days (8.80) Caputl (10.95) Cheat (6.62) Quota (10.96) Adjustment rate (3.24) Prce(-1) 0.27 (2.65) Prce(-2) (2.35) Prce(-3) 0.36 (1.98) Page 10 of 24

11 Fgure 2: The model structure Demand sde (1) DEM for Y : Real GDP POIL Y,., E tme D P US, euro area,uk, Swtzerland,Other P D : Domestc tme :tme trend DEM :Ol demand n mbd POIL :Ol prce nusd prce ndex E: Exchange rate vs - a - vs USD Dev. Eco. NJA,Transton, Latn Amerca, Rest of the world Supply sde (2) PROD (3) PROD (3' ) PROD oecd Stocks non-opec j Cost : Measure of Cost DEM NGLS : Non Gas Lqud, ROIL 0.95* Capacty PROD :Ol supply n mbd ROIL : Real prce of ol nusd : Ol stocks reported n mbd for j US (lower 48 Alaska),Canada, Asa, Afrca, Europe, Latn Amerca (Brazl, Mexco and opec opec Stocks producton costs PG :Processng Gans n mbd (net) j opec OECD oecd - NGLS - j PROD non-opec - PG others) Ol Prce (4) POIL oecd Quota Stocks oecd DEM : OPEC quota n mbd Capacty :OPEC capacty n mbd, Quota opec, PROD opec Quota opec, PROD Capacty opec opec Closure of the model The model s closed by OPEC behavour : cooperaton [equaton (3)] or non - cooperaton Note: A bar over a varable means t s determned exogenously. [equaton (3') stocks used as resduals] Page 11 of 24

12 3 Assessment of the model Forecast performance of the ol model To nvestgate the forecast performance of the model we carred out an n-sample statc and dynamc smulaton from 1995q1 to 2000q3. Ol prce $/b: Actual and model projectons Fgure Act ual Dynamc St at c Q1 1995Q3 1996Q1 1996Q3 1997Q1 1997Q3 1998Q1 1998Q3 1999Q1 1999Q3 2000Q1 2000Q3 Source: ECB Fgure 3 dsplays the actual and the ftted seres for ol prce over the smulaton perod. Standard measures of forecast accuracy suggest that the forecast errors mpled by the model are small (see Table 4). The RMSE of the fully dynamc smulaton s less than 2% after 1 quarter, around 2.5% after one year and s slghtly less than 5% after 3 years. Table 4: Computatons of the n-sample forecastng performance for ol prce (Root mean squared errors n percentage of the baselne value) 1 quarter 1 year 2 years 3 years Dynamc smulaton Statc smulaton Assessment of the model propertes: Some basc smulatons The man propertes of the model are presented n ths secton wth the help of four smulatons. The frst one, an exogenous ol prce shock, helps understand the supply/demand behavour wthout any prce feedback. The three followng smulatons are related wth each component of the prce equaton, namely OPEC quota, OPEC capacty and OECD stocks, whle the demand sde of the model and ts nteracton Page 12 of 24

13 wth the other parts are tested by two addtonal smulatons: an ncrease n the US real GDP and a deprecaton of the US dollar. Fnally, an alternatve behavour of OPEC s presented n the last smulaton. In all the smulatons, dependng on the orgn of the shock, the prce rule gves the optmal prce response, determned smultaneously wth demand. Non-OPEC supply also reacts to ths change n prce wth one-quarter lag, whle OPEC produces the necessary quantty of ol, whch together wth non-opec supply, matches demand. Exogenous ol prce shock The fst shock assumes a 50% permanent ncrease n the prce of ol. As depcted n Fgure 4, the 50% ncrease n the prce of ol leads to a decrease n demand by 3% n the long run. Non-OPEC producton tends to ncrease, although the response s qute nelastc to a change n prce (non-opec supply ncreases by 1.75% relatve to baselne). Because demand decreases and non-opec supply ncreases, OPEC needs to cut ts producton to balance supply and demand. Fgure 4 Exogenous shock on ol prce (50% ncrease) WORLD_SUPPLY/ DEM AND OPEC_SUPPLY NON_OPEC_SUPPLY OIL_PRICE Q4 2002Q1 2002Q2 2002Q3 2002Q4 2003Q1 2003Q2 2003Q3 2003Q4 2004Q1 2004Q2 2004Q3 2004Q4 Source: ECB Although the response of non-opec supply appears to be nelastc at an aggregate level, some dfferences across non-opec countres occur (Fgure 5). The bulk of the supply responses to the 50% ncrease n ol prce s around 1% wth two countres departng sgnfcantly from ths average, namely Mexco and the lower 48 US states whose response s around 5.5%. In Mexco, the sgnfcant response of producton s due to the fact that Mexco s a relatve new producer, wth lots of unused capactes and a sgnfcant lack of cash to explot those unused capactes. Hence, an ncrease n ol prce would provde Mexcan producers wth the cash needed to pump more ol. For the US, the hgh response s due to the presence of a large number of producers who are able to ncrease quckly and strongly ther producton n case of a spke n ol prces that would ncrease profts. Page 13 of 24

14 Fgure 5 Exogenous shock on ol prce (50% ncrease) MX_SUPPLY BR_SUPPLY US_SUPPLY CA_SUPPLY AL_SUPPLY AS_SUPPLY AF_SUPPLY EU_SUPPLY LA_SUPPLY Q4 2002Q1 2002Q2 2002Q3 2002Q4 2003Q1 2003Q2 2003Q3 2003Q4 2004Q1 2004Q2 2004Q3 2004Q4 Source: ECB Shock on OPEC quota The ncrease n OPEC quota by 10%, assumed n ths smulaton, can be nterpreted only as an announcement effect. Indeed, OPEC producton n the model s ntally unaffected by the shock per se. Through the prce equaton, the ncrease n quota depresses sgnfcantly the prce of ol (a 10% ncrease n quota leads to a mmedate 6% decrease n ol prce). As a result, demand ncreases and OPEC flls n the gap to balance demand and supply (see fgure 6) causng a further lagged decrease n non-opec supply and prce, whle demand keeps on growng. Hence, OPEC needs to produce even more to absorb the ncreasng excess demand. However, ths excess demand tends to nfluence the prce of ol, whch resumes an upward path. Demand reacts negatvely to ths prce ncrease, whch causes further ol prce decrease n the followng perod. After a successon of oscllatons, the prce of ol reaches a long run effect close to a 10% permanent decrease. Page 14 of 24

15 Fgure 6 Shock on OPEC quota wth OPEC producton unaffected by the shock (10% ncrease) WORLD_SUPPLY/DEM AND OPEC_SUPPLY NON_OPEC_SUPPLY OIL_PRICE Q4 2002Q1 2002Q2 2002Q3 2002Q4 2003Q1 2003Q2 2003Q3 2003Q4 2004Q1 2004Q2 2004Q3 2004Q4 Source: ECB Shock on OPEC capacty A 5% ncrease n OPEC capacty decreases automatcally OPEC s rate of capacty utlsaton, whch acts negatvely on ol prce leavng OPEC producton ntally unaffected by the shock. Only demand and non- OPEC supply, n reacton to prce developments, determne the response of OPEC. As we can see n Fgure 7, the magntude of the response as well as ts dynamc pattern s very smlar to the prevous smulaton. Fgure 7 Shock on OPEC capacty wth OPEC producton unaffected by the shock (5% ncrease) WORLD_SUPPLY/DEM AND OPEC_SUPPLY NON_OPEC_SUPPLY OIL_PRICE Q4 2002Q1 2002Q2 2002Q3 2002Q4 2003Q1 2003Q2 2003Q3 2003Q4 2004Q1 2004Q2 2004Q3 2004Q4 Source: ECB Page 15 of 24

16 Shock on OECD stocks The avalablty of stocks coverng a longer perod of consumpton exerts downward pressures to the prce of ol. Indeed, the prce of ol decreases sharply (by around 25% ntally and by a maxmum of 32% after one year) and demand tends to ncrease, reducng as a consequence the ntal negatve prce mpact (see Fgure 8). However, ths shock has a permanent, negatve mpact on ol prce because demand does not ncrease enough to absorb all the ncrease n stocks. Non-OPEC supply s slghtly reduced by the decrease n prce and OPEC supply has to ncrease to elmnate the excess demand. Fgure 8 Shock on ol stocks (10% ncrease) WORLD_SUPPLY/ DEM AND OPEC_SUPPLY NON_OPEC_SUPPLY OIL_PRICE Q4 2002Q1 2002Q2 2002Q3 2002Q4 2003Q1 2003Q2 2003Q3 2003Q4 2004Q1 2004Q2 2004Q3 2004Q4 Source: ECB Demand shock: an ncrease n US GDP A demand shock due to an ncrease n US real GDP by 1% leads to an mmedate 0.2% rse n ol demand mplyng a 0.4% rse n OPEC producton. Part of the excess demand s also elmnated by an ncrease n ol prce by 1.2% after two quarters, whch causes demand to declne slghtly and non-opec supply to ncrease leadng OPEC to satsfy only half of the ntal need for the subsequent quarters (see Fgure 9). However, the extent to whch the model varables react to ths shock s very small. Although the ncrease n US ol demand s n the long-term equvalent to the ncrease n real GDP (long-run elastcty equal to 0.96), there s n ths model no second-round effect of US GDP ncrease on the rest of the world. Page 16 of 24

17 Fgure 9 Shock on US real GDP (1% ncrease) WORLD_SUPPLY/ DEM AND OPEC_SUPPLY NON_OPEC_SUPPLY OIL_PRICE Q4 2002Q1 2002Q2 2002Q3 2002Q4 2003Q1 2003Q2 2003Q3 2003Q4 2004Q1 2004Q2 2004Q3 2004Q4 Source: ECB USD deprecaton A shock on the USD acts through the demand sde of the model. Indeed, ol demand s modelled as dependent on the real prce of ol n domestc currency. Hence, a US dollar deprecaton tends to lower the prce of ol n domestc currency for the countres other than the US. The shock smulated here s a 10% deprecaton of the USD. Ths ncreases the demand for ol by around 0.7% n the short-run. Because demand rses, the prce of ol ncreases (by 2%), reducng as a second-round effect the ntal response of demand. In the long term, however, demand s permanently affected by the ntal shock (by 0.5%), whle ol prce tends to converge towards a 1.5% permanent ncrease (see Fgure 10). Fgure 10 Shock on USD (10% deprecaton) WORLD_SUPPLY/ DEM AND OPEC_SUPPLY NON_OPEC_SUPPLY OIL_PRICE Q4 2002Q1 2002Q2 2002Q3 2002Q4 2003Q1 2003Q2 2003Q3 2003Q4 2004Q1 2004Q2 2004Q3 2004Q4 Source: ECB Page 17 of 24

18 Alternatve OPEC behavour In the prevous smulatons, we have assumed that OPEC adopted a co-operatve behavour. We have seen, however, that the current producton levels of OPEC countres are less than ther operable capacty (around 85%). In ths last smulaton, nstead of consderng OPEC as a swng producer, we assume that OPEC ncreases producton to levels consstent wth operable capacty (95% n ths case, whch represents a sgnfcant ncrease n OPEC supply). The closure of the model s n ths case realsed through the OECD stocks that absorb the excess of supply, puttng downward pressures on ol prces. Ths prce response acts negatvely on non-opec supply but postvely on ol demand. In the long run, the decrease n prces tend to let demand rse so that to absorb any remanng dsequlbrum between demand and supply. Part of the sharp ncrease n OPEC supply s absorbed by stocks, leadng the prce of ol to declne strongly by 40% on average, through the ncrease n the number of days of forward consumpton (see Fgure 11). However, another force plays n the other drecton through the ncrease n OPEC capacty utlsaton, whch tends to make the prce of ol more costly. Ths explans the very large oscllatons exhbted n Fgure 11. In the long-term, the effect of stock ncrease s larger, leadng to a permanent decrease n ol prce. Fgure 11 Non-cooperatve behavour of OPEC WORLD_SUPPLY OPEC_SUPPLY NON_OPEC_SUPPLY WORLD_DEM AND OIL_PRICE STOCKS Q4 2000Q2 2000Q4 2001Q2 2001Q4 2002Q2 2002Q4 2003Q2 2003Q4 2004Q2 2004Q4 Source: ECB Page 18 of 24

19 3 Concludng remarks Ths paper descrbes a model of the world ol market used for ol prce projectons and to provde quanttatve analyses to ol related rsk assessments. In ths model, ol demand s explaned by behavoural equatons that relate demand to domestc actvty and the real prce of ol. Ol supply for non-opec producers s derved from a compettve behavour takng nto account the effect of geologcal and economc varables. Ol prces are defned by a prce rule based on changes n market condtons and OPEC behavour. In partcular, OPEC acts accordng to a co-operatve behavour and ensures the global equlbrum at the prce determned by the prce rule. Econometrc estmatons show that the demand for ol s explaned by economc actvty, the real prce of ol and a tme trend representng technologcal developments related to energy effcency. To estmate non-opec producton, a complex nterrelaton between resource depleton, techncal change, economc ncentves and poltcal consderatons has been taken nto account through a methodology that combnes the curve fttng technque developed by Hubbert wth econometrc modelng methods. Ths so-called hybrd method has proved to be satsfactory to explan non-opec supply. The econometrc estmaton of the prce rule ndcates that the real ol prces are affected by OPEC capacty utlsaton, OPEC producton quotas, the degree to whch OPEC members cheat on those quotas, and crude ol stocks n OECD natons. Underlyng analyss ndcates also that these varables Granger cause real ol prces but real ol prces do not Granger case OPEC capacty utlsaton, OPEC producton quotas, the degree to whch they cheat on those quotas, and ol stocks n OECD natons. Ths result shows that OPEC can exert a powerful nfluence on real ol prces over the medum and long term and undermnes clams about the declne n OPEC market power. Smulaton results show that the model satsfactorly reproduces past developments n ol markets and t performs accordng to expectatons when exogenous shocks are ntroduced. The response of prces s sgnfcant and consstent n a hstorcal perspectve. As expected, the responses of demand and non- OPEC supply are rather nelastc to changes n prce. Although OPEC s assumed to close the model by absorbng any excess n supply or demand, the model shows that OPEC decsons about quota and capacty utlsaton have a sgnfcant, mmedate mpact on ol prce. Page 19 of 24

20 References Al Fars A. (1991), The determnants of crude ol prce adjustment n the world petroleum market, O.P.E.C. revew. Bacon R. (1991), Modellng the prce of ol, Oxford revew of economc polcy. Balabanoff, S. (1995), Ol futures prces and stock management: a contegraton analyss, Energy Economcs, 17(3), p Belton, B. (1998), Opec can no longer make 'people quake', USA Today, March, 27. Cleveland, C.J. and Kaufmann, R.K. (1991), Forecastng ultmate ol recovery and ts rate of producton: ncorporatng economc forces nto the models of M Kng Huggert, The Energy Journal, 12(2), p Fsher, A.C. (1981), Resources and Envronmental Economcs, Cambrdge Unversty Press, Cambrdge. Frankel, P.H. (1946), Essentals of Petroleum, Franck Cass, London. Gately, D. (1995), Strateges for Opec's prcng and output decsons, The Energy Journal, 16(3), p Gately, D. and J.F. Kyle. (1977), Strateges for OPEC's prcng decsons, European Economc Revew, 10, p Grffn, J.M. (1985), Opec Behavor: a test of alternatve hypotheses, Amercan Economc Revew, 75(5), p Hubbert, M.K. (1962), Energy Resources: A Report to the Commttee on Natural Resources, Natonal Academy of Scences. Kaufmann, R.K. (2003), Dees, S, Karadeloglou, P and Sanchez, M (2003), Does OPEC matter? An Econometrc Analyss of Ol prces, mmeo. Kaufmann, R.K. (1991), Ol producton n the lower 48 states: reconclng curve fttng and econometrc models, Resources and Energy, 13, p Kaufmann, R.K. (1995), A model of the world ol market for Project LINK: Integratng economcs, geology, and poltcs, Economc Modelng, 12(2), p Page 20 of 24

21 Stock, J.H. and M.W. Watson (1993), A smple estmator of contegratng vectors n hgher order ntegrated systems, Econometrca, 61, p Page 21 of 24

22 Appendx : Ol demand equaton results Ths appendx presents detaled econometrc results of the demand equatons presented n the man text. Table 5 presents unt root tests of the varables and Tables 6a and 6b gve the complete estmaton results. Table 5: Unt root tests ADF - constant and trend ADF - constant ADF - wthout constant DEM DEM DEM DEM DEM DEM US * *** *** Japan *** ** *** UK -4.83*** -6.77*** *** *** Euro area *** *** *** ODE * Transton ** ** NJA *** -3.96*** -3.26** * Lat. Amerca ** RoW -3.71** -9.50*** -4.32*** -3.16*** Y Y Y Y Y Y US ** *** ** Japan UK * Euro area *** *** ODE -3.21* -4.98*** *** Transton ** ** NJA * ** Lat. Amerca *** ** *** RoW ROIL ROIL ROIL ROIL ROIL ROIL US -3.70** -4.87*** -3.34** -4.76*** *** Japan *** -3.37** -4.71*** *** UK -3.61** -4.86*** -3.72*** -4.49*** *** Euro area -3.79** -7.23*** -2.99** -7.28*** *** ODE *** -3.15** -4.77*** *** Transton -3.74** ** -2.96** ** NJA -3.35* -3.26** -3.12** -3.17** *** Lat. Amerca -3.36* -5.13*** *** *** Note: We use data-drven lag selecton procedures n the ADF tests, takng as the crtcal value used for sgnfcance of lagged terms and 4 as the maxmum number of lags allowed n these procedures nto account. We denote wth */**/** the rejecton of the null hypothess at a 10%/5%/1% crtcal levels. Crtcal levels used for ADF are the followng: - In the model wth constant and trend: (1%), (5%) and 3.15 (10%). - In the model wth constant: (1%), (5%) and 2.58 (10%). - In the model wthout constant: (1%), (5%) and 1.62 (10%). Page 22 of 24

23 Table 6a: Econometrc results Unted States Japan U. Kngdom Euro area Swtzerland Long-term equaton Constant 5.49 (11.27) 5.78 (15.87) 6.67 (28.07) 6.59 (41.1) 4.74 (16.17) GDP 0.98 (8.61) 0.61 (7.61) 0.17 (3.36) 0.57 (16.32) 0.18 (2.89) Tme trend (-4.99) Short-term equaton Adj. Coef (-5.37) (-2.72) (-2.17) (-6.89) (-7.84) Constant (-1.73) 0.07 (8.37) 0.00 (0.52) 0.02 (2.79) (-2.48) GDP 0.77 (2.29) 0.45 (0.85) 1.08 (0.70) GDP(-1) 0.89 (2.63) 0.65 (1.21) ROIL (-1.45) (-2.01) ROIL(-1) (-1.78) (-1.63) (-2.73) Q (-2.20) (-24.2) (-5.40) (-7.50) (-0.13) Q (1.99) (-3.46) 0.00 (0.14) (-2.63) 0.08 (4.15) Q (2.65) 0.02 (1.64) 0.02 (2.18) 0.01 (1.06) 0.05 (2.91) Sample 1984:1-2002:1 1984:1-2002:1 1985:2-2002:1 1984:1-2002:1 1984:1-2002:1 R Table 6a: Econometrc results (contnued) NJ Asa Transton Latn Amerca ODE RoW Long-term equaton Constant 5.77 (35.27) 6.67 (0.88) 4.81 (22.89) 6.41 (27.91) 6.25 (123.2) GDP 0.77 (22.33) 0.51 (2.49) 0.85 (18.80) 0.39 (7.90) 0.55 (50.66) Tme trend (-5.27) Short-term equaton Adj. Coef (-1.88) (-4.75) (-0.64) (-1.98) (-3.65) Constant (-1.69) 0.01 (0.83) 0.10 (0.57) (-1.66) (-0.41) GDP 1.73 (2.75) 0.82 (3.03) (0.49) 0.58 (1.56) GDP(-1) (0.48) ROIL (-1.55) (-0.51) ROIL(-1) (-0.84) Q (-1.13) (-5.27) 0.03 (4.14) (-2.41) 0.01 (1.22) Q (-1.04) (-2.83) 0.03 (3.56) 0.03 (3.39) 0.01 (2.21) Q (3.76) 0.02 (1.80) (-0.87) 0.03 (5.10) (-1.52) Sample 1993:1-2002:1 1995:1-2002:1 1993:1-2002:1 1993:1-2002:1 1991:1-2002:1 R Page 23 of 24

24 Table 6b: Econometrc results wth tax-ncludng ol prces Unted States Japan U. Kngdom Euro area Swtzerland Long-term equaton Constant 5.49 (11.27) 5.78 (15.87) 6.67 (28.07) 6.59 (41.1) 4.74 (16.17) GDP 0.98 (8.61) 0.61 (7.61) 0.17 (3.36) 0.57 (16.32) 0.18 (2.89) Tme trend (-4.99) Short-term equaton Adj. Coef (-5.79) (-2.88) (-2.17) (-7.01) (-8.28) Constant (-2.01) 0.06 (8.02) 0.00 (0.52) 0.02 (2.80) (-2.67) GDP 0.86 (2.57) (0.37) 0.82 (0.55) GDP(-1) 0.78 (2.34) 0.65 (1.21) ROIL (-2.71) (-2.65) ROIL(-1) (-1.03) (-1.84) (-2.73) Q (-2.94) (-23.9) (-5.40) (-7.13) 0.00 (0.16) Q (2.24) (-3.53) 0.00 (0.14) (-2.56) 0.08 (4.21) Q (2.56) 0.02 (1.52) 0.02 (2.18) 0.01 (1.37) 0.06 (3.28) Sample 1984:1-2002:1 1984:1-2002:1 1985:2-2002:1 1984:1-2002:1 1984:1-2002:1 R Page 24 of 24