Price Asymmetries in International Gasoline Markets

Transcription

1 Price Asymmeries in Inernaional Gasoline Markes Marzio Galeoi Universiy of Bergamo and Fondazione Eni Enrico Maei Alessandro Lanza Eni S.p.A., Fondazione Eni Enrico Maei and CRENos Maeo Manera Bocconi Universiy and Universiy of Milano-Bicocca Absrac: This paper re-examines he issue of asymmeries in he ransmission of shocks o crude oil prices ono he reail price of gasoline. Relaive o he previous lieraure, he disinguishing feaures of he presen paper are: i) use of updaed and comparable daa o carry ou an inernaional comparison of gasoline markes; ii) wo-sage modeling of he ransmission of oil price shocks o gasoline prices (firs refinery sage and second disribuion sage), in order o assess possible asymmeries a eiher one or boh sages; iii) use of asymmeric error correcion models o disinguish beween asymmeries ha arise from shor-run deviaions in inpu prices and from he speed a which he gasoline price revers o is long-run level; iv) explici, possibly asymmeric, role of he exchange rae, as crude oil is paid for in dollars whereas gasoline sells for differen sums of naional currencies; v) boosrapping of F ess of asymmeries, in order o overcome he low-power problem of convenional esing procedures. In conras o several previous findings, he resuls generally poin o widespread differences in boh adjusmen speeds and shor-run responses when inpu prices rise or fall. 1. Inroducion This paper deals wih he ransmission of posiive and negaive changes in he price of oil o he price of gasoline. There is a sizeable lieraure looking for empirical evidence in suppor of asymmeries in his ransmission mechanism: allegedly, gasoline prices go up faser han hey go down whenever here is urbulence in inernaional crude oil markes. Indeed, in every recurring period of ension in he price of oil here has been renewed ineres and even heaed debae abou he level of gasoline prices, he magniude of is cos componens, including reailers margin, and he axes ha conribue o keep hose prices high and sluggish. The debae invariably ceners on he fac ha gas prices do no decrease so rapidly as oil prices do. Asymmeries in he ransmission of changes in he price of oil are in general possible if gasoline markes are non-compeiive. Indeed, unlike he case of perfec compeiion, varying degrees of price rigidiy are a feaure of oligopolisic and monopolisically compeiive markes. Economic heory offers a few explanaions of such price rigidiy, and hese ofen conemplae he possibiliy of an asymmeric ransmission of cos changes ono produc prices. However, perhaps because srucural models of asymmeric price effecs are no easily ranslaed ino empirical models, he lieraure on gasoline prices has ypically employed reduced-form dynamic equaions relaing he price of gasoline o he price of oil. 472 Findings vary across counries and periods, bu on he whole hey do no provide firm evidence ha prices rise faser han hey fall. To dae, herefore, he empirical evidence does no seem o jusify he blame ha he press, public opinion, some poliical groups and environmenalis movemens pu on he oil indusry paricularly in periods of highly volaile prices. This paper provides fresh new evidence o bear on he issue of price asymmeries in gasoline markes: are hey real or are hey only imaginary as people end o pay aenion o gasoline prices only when hey are rising rapidly? In his paper we sudy poenial price asymmeries in he markes of leaded gasoline of five European counries, namely Germany, France, U.K., Ialy, and Spain. The daa are monhly and in general range from he 1985 o Relaive o he previous lieraure, he paper is novel in several respecs. Firs of all, he paper presens an inernaional comparison employing he same empirical model and esimaion echnique and using a very recen, comparable daa se. The fac ha no unanimous conclusions could be drawn by previous individual counry sudies may, among oher hings, depend upon he fac ha no similar daa across counries for a uniform ime period were used. A second consideraion concerns he economeric mehodology and he dynamic model used o assess price asymmeries. In his paper an error correcion

2 mechanism (ECM henceforh) is esimaed hroughou afer applying modern uni roo and coinegraion echniques. Relaive o parial adjusmen, oldfashioned ECM and general dynamic models, we are here able o disinguish beween wo ypes of asymmeries: he firs refers o he fac ha adjusmen of curren price levels o desired arges may differ depending on he sign of he adjusmen, he second one insead has o do wih asymmeries in ransiory price movemens. A hird aspec ha disinguishes he presen conribuion is a more saisfacory consideraion of he organizaional srucure of he indusry under scruiny. The majoriy of previous sudies invesigaed he relaionship beween crude oil and reail gasoline prices as a single sage process. However, he indusry has a more complex srucure, ofen varying counry by counry. We may make a small sep forward and roughly suppose he exisence of wo sages in he producion and disribuion process: he firs one concerns he ransformaion of crude oil ino he refined produc, while he second one has o do wih he disribuion of gasoline o reailers. The relevan prices involved in he firs sage, herefore, are crude oil price and ex-refinery price. In addiion, since crude oil is paid for in dollars whereas gasoline sells for differen sums of naional currencies, he exchange rae plays a relevan, possibly asymmeric, role a his sage. The subsequen sage insead deals wih he relaionship beween ex-refinery and reail gasoline prices. We hink ha his sraegy provides a more saisfacory represenaion of he complex chain linking crude oil o pump prices. Moreover, we believe ha i is of ineres o find ou wheher price asymmeries originae upsream or downsream in he ransmission process: in view of he suspicions of collusive behavior ha ofen accompany increases in reail gasoline prices, he wo sage naure of our invesigaion is clearly useful. A final poin made in he paper is mehodological. A few applied sudies employing he asymmeric ECM model have recenly documened ha he commonly used F-ess of equaliy among he coefficiens accouning for he asymmeries are biased oward acceping he null of symmery in small samples. This fac could explain why he daa fail o urn up he asymmeric price adjusmens ha many commonly suspec. A way around his echnical difficuly is o boosrap he asymmery ess. We carry ou his ask in he paper and presen resuls from boh sandard and boosrapped es oucomes. The resuls of he esimaed parameers generally poin o widespread differences boh in adjusmen speeds and shor-run elasiciies when inpu prices rise or fall. This appears o confirm he common percepion ha price increases are larger han price reducions. This finding characerizes, albei o a differen exen, nearly all counries and boh sages of he ransmission chain. Convenional ess, however, fail o rejec he 473 symmery hypohesis. Ye, drawing from a few conribuions which have addressed his specific aspec in relaed conexs, we have reasons o believe ha hose usual ess have low power when samples are of limied size. We herefore boosrap he F saisics and repor rejecion frequencies of our ess. The resuls srongly confirm he emergence of widespread price asymmeries in he daa we have examined. 2. Daa and Economeric Mehodology We represen he complex chain of ransformaion of crude oil ino he refined gasoline produc as a wosage process: firs we consider he ransformaion of oil ino he refined produc, and hen is disribuion o gas saions. Of each sage we invesigae he poenial asymmeries in he ransmission of inpu price changes ono oupu prices. For he sake of comparison wih he bulk of he lieraure and in order o appreciae he advanages of a wo-sage represenaion, we also consider he ransmission of changes in crude oil prices direcly on he price of gasoline. We conduc an inernaional comparison of he issue a hand among five European counries, namely Germany, France, U.K., Ialy, and Spain. The sample period ranges from January 1985 o June 2000 (he German sample sops a February 1997) and leaded gasoline has been considered. The variables used in his paper are he price of crude oil (C), he gasoline spo price (S), he before-ax gasoline reail price (R), and he exchange rae beween U.S. dollar and individual naional currencies (ER). We denoe he naural logarihm of hese variables by lowercase leers. In paricular, crude oil price is he Crude Oil Impor Coss in U.S. dollars/bbl (average uni value, c.i.f.) published by he Inernaional Energy Agency. As a proxy for he ex-refinery gasoline price we use he gasoline spo price f.o.b. Roerdam for he European counries. The gasoline reail prices are also from he Inernaional Energy Agency. Since any company purchasing in he spo marke mus pay in dollars, he exchange rae beween he naional currency and U.S. dollar is used. These series are aken from he Inernaional Moneary Fund. As already emphasized by Bacon (1991), i is clear ha he exchange rae may be a relevan source of asymmery in non-u.s. counries. Hence he imporance of allowing separaely for posiive and negaive changes in exchange raes. Summarizing, he basic relaionships we ake o he daa are he following: s r = a0 + a1c + a2er + ε (1) = a0 + a1s + ε (2) r = a0 + a1c + a2er + ε (3)

3 Equaions (1)-(3) represen he long-run or equilibrium relaionships relaing oupu prices o inpu prices and exchange raes. Equaion (1) refers o he firs sage in which he price of crude oil, along wih he exchange rae, deermines he spo gasoline price; according o (2) his price affecs he reail price of gasoline. I is apparen ha he exchange rae eners only he firs sage of he chain as boh reail and spo prices are denominaed and posed in he same naional currency. Finally, (3) relaes in a single sage he reail price of perol o ha of crude oil. The asymmery in he ransmission of changes in inpu prices o oupu prices can be accomodaed wihin a dynamic model. However, i is imporan o disinguish beween wo ypes of asymmeries. This disincion can be bes enerained by a dynamic ECM specificaion esimaed in wo seps. Tha is, le εˆ denoe he residual of (1) hrough (3) and define ( + ) ( ) ECM = ˆ ε > 0 and ECM = ˆ ε < 0. These wo erms accoun for asymmery in he adjusmen o equilibrium, whereas shor-run asymmery is capured by similarly decomposing price (and exchange rae) ( + ) changes ino x = x x 1 > 0 and ( ) x = x x 1 < 0 for x=c,s,er. Thus asymmery of he firs ype is relaed o he speed a which a gap beween he curren and he equilibrium level (as described by he long-run relaionships (1)-(3)) of spo or reail prices is filled wihin he period. This speed can differ according o wheher he curren price is below or above is equilibrium level. The asymmeric ECMs can herefore be formulaed as follows: s = α + β ECM + β ECM γ c + γ c + δ er + δ er + u r = α + β ECM + β ECM + γ 1 1 s + γ s + u r = α + β ECM + β ECM γ c + γ c + δ er + δ er + u (4) (5) (6) where is he firs difference operaor. Augmened Dickey-Fuller (ADF) ess for uni roos have been used and all variables have been found o be inegraed of order one, or I(1), mos of hem wih inercep and rend. Though non-saionary, hese series may form a linear combinaion which is saionary, or I(0). In his case here are long-run or equilibrium relaionships 474 beween relevan price series as represened in (1)-(3). The relevan series are said o be coinegraed and his implies ha he residual of he equaions εˆ is I(0). Again ADF ess are used o es for coinegraion. All he above consideraions apply o he case of symmeric effecs, bu he ECM has been exended o he case of asymmeric adjusmen originally by Granger and Lee (1989). In his case firs differences and coinegraion residual can be decomposed ino posiive and negaive changes as shown above. 3. Empirical Resuls Table 1 repors he magniude and saisical significance of he adjusmen speed coefficiens β and β, which allow us o evaluae long-run or persisen asymmery, and magniude and significance of he coefficiens of price changes γ and γ, which insead accoun for shor-run or ransiory asymmery. The resuls sugges several remarks. Firsly, he able shows ha, wih jus five excepions, all coefficiens are saisically significan, in he vas majoriy of cases a he 1% confidence level. According o Granger and Lee (1989), he significance of individual coefficiens is a necessary condiion for esing for asymmeric effecs. Secondly, he comparison beween posiive and negaive coefficiens shows ha he former in general exceed, in absolue value, he laer. This holds for each sage of he producion and disribuion chain and boh for longrun and shor-run. However, o esablish significan divergences beween he wo groups of parameers requires rigorous saisical esing, an aspec we ackle below. Thirdly, if we consider he wo-sage approach we have adoped o describe he ransmission chain of price shocks o final gasoline prices relaive o he single-sage pracice, we see ha numerical esimaes are very differen. In paricular, firs-sage adjusmen speeds are generally smaller in absolue value han second-sage speeds, whereas he conrary appears o be rue for shor-run price elasiciies. These differences do no surface in he single-sage approach. If we resric our aenion o poin esimaes, he picure ha emerges appears o confirm he general percepion of a more rapid adjusmen in he case of price rises relaive o price falls. This can be seen, as far as he shor-run is concerned, by direc comparison of he esimaed parameers in ha hey represen conemporaneous price adjusmens: in he majoriy of cases he coefficien γ associaed wih price increases is larger han ha corresponding o price reducions. There are excepions hough, bu he differences among coefficiens do no appear o be large. I is clear ha in hese cases a es of equaliy beween coefficiens is called for. As for he long-run, noe ha he posiive and negaive β coefficiens are respecively associaed wih adjusmen o he long-run equilibrium level of price from above and from below. Also hese

4 parameers differ in general. I is however perhaps useful and more informaive o use an alernaive saisics based on hose adjusmen speed coefficiens. The general formulaion of he asymmeric ECM, originally inroduced by Granger and Lee (1989), has been ofen used as he appropriae saisical framework for convenional F ess of he null hypohesis of symmeric adjusmen, boh in erms of adjusmen speeds owards a coinegraing equilibrium and of shor-run responses. In erms of equaions (4)-(6) a L ( ) ( ) rejecion of he null hypohesis : + H 0 β = β implies asymmeric adjusmen o a long-run equilibrium, whereas shor-run asymmeries arise when S ( ) ( ) he null hypohesis : + H 0 γ = γ is rejeced. Table 2 repors he calculaed convenional F ess of L S H 0 and H 0. I immediaely appears ha he symmery hypohesis is rejeced only in 5 cases ou of 30 and, of hese 5 cases, only 1 is a rejecion a 1% significance level (3 cases a 5 % and 1 a 10%). In addiion, shor-run symmery is rejeced in 2 cases ou of 5, while symmeric adjusmen o he long-run is rejeced in 3 cases, including he one a 10% significance level. The counries which do no experience any asymmery are Ialy, U.K. and Germany. France and Spain show boh ypes of asymmeries (a single and second sage, respecively). The overall picure which emerges from esing he symmery hypohesis herefore runs conrary o boh he common percepion and o he visual inspecion of he magniude of he esimaed coefficiens made in he previous ables. A few recen papers (Cook, Holly and Turner, 1998, 1999; Cook, 1999) have quesioned he reliabiliy of convenional ess of symmery in he above and similar conexs. In paricular, hese conribuions noe ha here is a endency o overrejec he null hypohesis of symmery which can be raced o he low power of sandard F saisics. In order o overcome he documened unreliabiliy of sandard ess of symmery we have boosrapped he F L S saisics for boh H 0 and H 0 and have calculaed he corresponding rejecion frequencies a he 5% significance level on he basis of 1,000 replicaions. The resuls are presened in Table 3. Rejecion frequencies greaer han 15% are found in 17 cases ou of 30, whereas 3 are he cases wih high (i.e. greaer han 58%) rejecion frequencies. If we reinerpre he resuls of Table 2 in he ligh of hese findings, he picure changes dramaically. In paricular, each counry is now more likely o be characerized by boh long-run and shor-run asymmeries. Moreover, in Ialy, Spain and U.K. asymmeries arise in he second sage, whereas in France and Germany hey appear mainly a firs and single sage. Generally speaking, we find ha in (almos) all counries asymmeries arise in he second sage. The sraighforward inerpreaion of his resul lies in he more compeiive environmen of he refining secor wih respec o he disribuion secor where several differen operaors could ac o increase upward rigidiy. 4. Conclusions In a recen paper, Pelzman (2000) saes ha economic heory suggess no pervasive endency for prices o respond faser o one kind of cos change han o anoher (p.467). However, afer an examinaion of lierally hundreds of markes (p.469), he concludes ha he person in he sree is righ and we [economiss] are wrong (p.493). This paper has re-examined he issue of presumed asymmeries in he ransmission of shocks o crude oil prices ono he reail price of gasoline. Especially in periods of volaile inernaional markes, his is an issue o which boh public opinion (nearly all are driving cars and rucks) and policy makers are quie sensiive. I has herefore araced he ineres of energy economiss, who have carried ou several empirical invesigaions on gasoline markes wih a special eye o he hypohesis ha prices rise faser han hey fall. The resuls of he esimaed parameers generally poin o widespread differences boh in adjusmen speeds and shor-run elasiciies when inpu prices rise or fall. This appears o confirm he common percepion amply echoed by newspapers in periods of increasing inernaional oil prices of more rapid price increases relaive o price reducions. This finding characerizes, albei o a differen exen, nearly all counries and boh sages of he ransmission chain. When however we urn o convenional esing, we find ha he usual F ess overwhelmingly fail o rejec he symmery hypohesis. Ye, drawing from a few conribuions which have addressed his specific aspec in relaed conexs, we have reasons o believe ha hose usual ess have low power when samples are of limied size. We herefore boosrap he F saisics and repor rejecion frequencies of our ess. The resuls srongly confirm he emergence of widespread price asymmeries in he daa we have examined. In summary, and in conras o several previous findings, we do find ha rockes and feahers appear o dominae he price adjusmen mechanism of gasoline markes in many European counries. Acknowledgemens We would like o hank Luigi Buzzacchi, Seve Cook, Michael McAleer, Colin McKenzie, Les Oxley and he paricipans o he IEP-IGIER seminar a Bocconi Universiy for helpful commens. This sudy does no necessarily reflec he views of Eni S.p.A.. 475

5 References Bacon, R.W. (1991), Rockes and Feahers: The Asymmeric Speed of Adjusmen of U.K. Reail Gasoline Prices To Cos Changes, Energy Economics, July, Cook, S. (1999), Nonsymmeric Error Correcion Revisied, Applied Economic Leers, 6, Cook, S., S. Holly, and P. Turner (1998), DHSY Revisied: he Role of Asymmeries, Applied Economics, 31, Cook, S., S. Holly, and P. Turner (1999), The Power of Tess for Non-lineariy: he Case of Granger- Lee Asymmery, Economics Leers, 62, Granger, C.W.J. and T.H. Lee (1989), Invesigaion of Producion, Sales and Invenory Relaionships Using Mulicoinegraion and Non-symmeric Error Correcion Models, Journal of Applied Economerics, 4, S145-S159. Pelzman, S. (2000), "Prices Rise Faser Than They Fall", Journal of Poliical Economy, 108, TABLE 1: Asymmeric Adjusmen Speeds and Shor-run Price Asymmeries Ialy France Spain Germany U.K. Firs Sage: spo = f(crude, exchange rae) Asym. adj. Speed β * ** * ** ** Asym. adj. Speed β * * ** Shor-run asymmeryγ 0.877** 0.801** 0.839** 0.867** 0.766** Shor-run asymmeryγ 0.856** 0.923** 0.581** 1.156** 0.718** Second Sage: reail = g(spo) Asym. adj. Speed β ** ** ** ** ** Asym. adj. Speed β ** ** ** ** ** Shor-run asymmeryγ 0.180** 0.362** 0.216** 0.478** 0.306** Shor-run asymmeryγ 0.205** 0.281** 0.229** 0.474** 0.194** Single Sage: reail = h(crude, exchange rae) Asym. adj. Speed β ** ** ** ** Asym. adj. Speed β ** ** ** ** * Shor-run asymmeryγ 0.196** 0.562** 0.236** 0.788** 0.435** Shor-run asymmeryγ 0.240** ** 0.552** 0.237** Noes o he able: a single (double) aserisk denoes significance a 5% (1%) level. 476

6 TABLE 2: Compued F Tess of Asymmeric Adjusmen Speeds and Shor-run Price Effecs Ialy France Spain Germany U.K. Null Hypohesis Firs Sage: spo = f(crude, exchange rae) Sym. adj. Speeds Shor-run symmery Second Sage: reail = g(spo) Sym. adj. Speeds ** Shor-run symmery ** Single Sage: reail = h(crude, exchange rae) Sym. adj. Speeds ** * Shor-run symmery *** Noes o he able: (i) enries are calculaed F ess of he equaliy beween esimaed coefficiens associaed wih error correcion erms (sym. adj. speeds) and price changes (shor-run symmery); (ii) a single (double) [riple] aserisk denoes significance a 10% (5%) [1%] level. TABLE 3: Simulaed F Tess of Asymmeric Adjusmen Speeds and Shor-run Price Effecs Ialy France Spain Germany U.K. Null Hypohesis Firs Sage: spo = f(crude, exchange rae) Sym. adj. Speeds Shor-run symmery Second Sage: reail = g(spo) Sym. adj. Speeds Shor-run symmery Single Sage: reail = h(crude, exchange rae) Sym. adj. Speeds Shor-run symmery Noes o he able: enries are simulaed rejecion frequencies, i.e. he percenage number of imes (ou of 1,000 replicaions) he null hypohesis of symmeric adjusmen speeds (resp. shor-run symmery) is rejeced by an F ess a 5% level. 477