The Magnitude and Distribution of Fuel Subsidies: Evidence from Bolivia, Ghana, Jordan, Mali, and Sri Lanka

Transcription

1 WP/06/247 The Magntude and Dstrbuton of Fuel Subsdes: Evdence from Bolva, Ghana, Jordan, Mal, and Sr Lanka Davd Coady, Moataz El-Sad, Robert Gllngham, Kangn Kpodar, Paulo Medas, and Davd Newhouse

2

3 2006 Internatonal Monetary Fund WP/06/247 IMF Workng Paper Fscal Affars Department The Magntude and Dstrbuton of Fuel Subsdes: Evdence from Bolva, Ghana, Jordan, Mal, and Sr Lanka Prepared by Davd Coady, Moataz El-Sad, Robert Gllngham, Kangn Kpodar, Paulo Medas, and Davd Newhouse 1 Authorzed for dstrbuton by Gerd Schwartz November 2006 Abstract Ths Workng Paper should not be reported as representng the vews of the IMF. The vews expressed n ths Workng Paper are those of the author(s) and do not necessarly represent those of the IMF or IMF polcy. Workng Papers descrbe research n progress by the author(s) and are publshed to elct comments and to further debate. Wth the recent jump n world ol prces, the ssue of petroleum product prcng has become ncreasngly mportant n developng countres. Reflectng a reluctance of many governments to pass these prce ncreases onto energy users, energy prce subsdes are absorbng an ncreasng share of scarce publc resources. Ths paper dentfes the ssues that need to be dscussed when analyzng the fscal and socal costs of fuel subsdes. Usng examples from analyses recently undertaken for fve countres, t also dentfes the magntude of consumer subsdes and ther fscal mplcatons. The results of the analyss show that n all of these countres energy subsdes have sgnfcant socal and fscal costs and are badly targeted. JEL Classfcaton Numbers: C40, C67, D31, H20, I38 Keywords: Energy prces, subsdes, welfare dstrbuton, household survey data Authors E-Mal Addresses: dcoady@mf.org; melsad@mf.org; rgllngham@mf.org; kkpodar@mf.org; pmedas@mf.org; dnewhouse@mf.org 1 All the authors are wth the Internatonal Monetary Fund. We would lke to thank staff n the Afrca, Asa and Pacfc, Fscal Affars, Mddle East and Central Asa, and Western Hemsphere Departments for ther helpful comments.

4 2 Contents Page I. Introducton...3 II. Magntude, Fnancng, and Dstrbuton of Consumer Subsdes...6 A. Consumer Prce Subsdes and Ther Fnancng...6 B. Welfare Impact of Hgher Fuel Prces...7 C. Alternatve Approaches to Mtgaton...9 III. Country Case Studes...10 A. Market Structure, Prcng Regme, and Fuel Subsdes...11 B. Real Income Effect of Increasng Petroleum Product Prces...15 C. Identfyng Alternatve Approaches to Mtgaton...18 IV. Concludng Remarks...22 References...36 Appendxes I. Identfyng Magntude and Fnancng of Fuel Subsdes...24 A. Defnng Consumer Prce Subsdes...24 B. Quantfyng Aggregate Consumer and Producer Subsdes...30 C. Fnancng Subsdes...30 II. Evaluatng Sze and Dstrbuton of Real Income Effects...32 A. Prce-Shftng Model...32 B. Applyng Model...35 Tables 1. Summary of Emprcal Evdence on Real Income Impact of Fuel Subsdes Fuel Budget Shares by Income Quntles...16 Fgures 1. Major Events and Real Prce of U. S. Ol Imports, Magntude and Fnancng of Consumer Subsdes Demand and Supply n Presence of Exports Prce Subsdy n Pure Export Case Prce Subsdy n Mxed Export/Nontraded Case Regulated Prcng n Nontraded Case Introducton of Subsdzed Prce: Nontraded Case Domestc Producton and Imports Pure Import Case...29

5 3 I. INTRODUCTION Government control of the domestc prces of petroleum products s a common feature n developng countres. In some cases, governments drectly control mport levels, domestc dstrbuton, and domestc prces. In other cases, the prvate sector can freely mport and dstrbute petroleum products, but governments set domestc prce celngs and compensate prvate sector dstrbutors to cover ensung losses. It s also common for prces to be set by a formula that anchors domestc prces to mport prces, wth adjustments for dstrbuton margns and domestc taxes. These prcng formulas may be mplemented by ether government-controlled or ndependent prcng boards. 2 A recent revew carred out by the IMF found that, out of the 48 developng and emergng economes consdered, only 16 could be classfed as lberalzed (.e., wth the prvate sector determnng prces wthout havng to explctly seek government permsson), 9 countres fxed prces accordng to an automatc formula, and 16 drectly controlled prces and adjusted them on an ad hoc bass. Governments that drectly control prces often mpose prce subsdes that keep domestc prces below border prces. Ths s partcularly the case when nternatonal fuel prces ncrease sharply and governments are reluctant to pass these ncreases fully on to the domestc prces of petroleum products. 3 Where prcng formulas are n use, ther applcaton s often temporarly suspended or permanently abandoned sx of the revewed countres had recently abandoned formula prcng. In a number of countres, fuel subsdes were projected to exceed 2 percent of GDP n 2005 even after recent substantal ncreases n domestc prces (e.g., Azerbajan, 12.7 percent; Bolva, 3.1 percent; Ecuador, 3.6 percent; Egypt, 4.1 percent; Indonesa, 3.2 percent; Jordan, 5.8 percent; and Yemen, 9.2 percent). In some countres, these expendtures were as least as large as publc educaton and/or health budgets. For example, n Indonesa and Yemen, total subsdes were hgher than the health and educaton budgets combned. The recent sharp ncreases n world ol prces are, however, not unusual. 4 Snce reachng a 28-year low n late 2001, ol prces have clmbed by over 200 percent n constant U. S. dollars through November Whle such an ncrease s substantal, t s mportant to place t n hstorcal context. Over the past 35 years, ol prces have fluctuated wdely. Usng real mport prces for the Unted States as an llustraton (Fgure 1), prces ntally jumped n 2 In order to avod sharp and frequent changes n domestc prces, automatc formulas typcally use an average of past world prces and trgger changes n domestc prces once the average change n world prces exceeds a certan range. These formulas often also nclude an element of taxaton. See the Regonal Economc Outlook for Sub-Saharan Afrca (May 2006) for a dscusson of how countres n ths regon dealt wth hgher energy prces. For a dscusson of the welfare and fscal mplcatons of alternatve prce smoothng rules, see Federco, Danel and Bngham (2001). See Gupta and Mahler (1995) and Gupta and others (2000, 2003) for revews of experences wth petroleum prcng n developng and developed countres. 3 We wll use the terms fuel subsdes and petroleum product subsdes nterchangeably throughout the paper. 4 Between July 2003 and August 2005, the nternatonal prce of crude ol ncreased by more than 200 percent, from US$26 to US$60 per barrel. Ths ncrease s perceved to reflect structural changes generatng hgher demand (e.g., from relatvely hgh growth n Inda and Chna), low stocks, and short-term supply constrants.

6 4 1973, wth the Arab ol embargo. They jumped n the late 1970s and early 1980s, after the Iranan revoluton, reachng a peak of almost $76 a barrel (2005 prces) n February of Prces fell steadly through the md-1980s and, untl recently, fluctuated prmarly between $20 and $30 a barrel. Despte the huge recent ncreases, real ol prces have not yet reached the hghs of the early 1980s. Fgure 1. Major Events and Real Prce of U. S. Ol Imports, $2005Q2 per barrel $80 $70 $60 $50 $40 $30 Iran-Iraq War begns; ol prces peak Saud Araba abandons "swng producer" role; ol prces collapse Gulf War Prces spke on Iraq war, rapd demand ncreases, constraned OPEC capacty, etc. Prces rse sharply on OPEC cutbacks, ncreased demand $20 Iranan revoluton; Shah deposed $ Ol Embargo Iraq nvades Kuwat Asan economc crss; ol oversupply; prces fall sharply Prces fall sharply on 9/11 attacks $ Although they are poltcally popular, fuel prce subsdes have adverse consequences for both government fnances and the effcent use of energy and often result n shortages. 5 Large subsdes redrect publc expendtures away from more productve spendng or contrbute to unsustanable budget defcts. Low fuel prces fal to provde the approprate ncentves to households to be more effcent n ther use of energy, whch would mtgate the 5 Increasng domestc fuel prces s often an extremely poltcally senstve ssue, especally n crude olproducng countres. For example, ncreases n Yemen n July 2005 led to wdespread socal dsrupton, whch resulted n 22 deaths and hundreds njured. Smlar publc reactons have occurred n the past n Ecuador (1998), Indonesa (1998), Ngera (2000), and Venezuela (1990). However, Gupta and others (2000, p ) conclude that such volent reactons to subsdy reform are the excepton rather than the norm, and often are not trggered by the reform alone.

7 5 overall adverse effect of hgher world prces on the economy. 6 In fact, gven that energy demand s nelastc and that there are negatve consumpton externaltes assocated wth ts use, taxaton of petroleum products s generally regarded as an effcent way to rase government revenue. 7 Moreover, ths paper wll argue that unversal energy subsdes are not a cost-effectve way to protect the real ncomes of poor households, snce they nvolve substantal leakage of benefts to hgher-ncome groups. Although the removal of fuel subsdes can often be regressve, especally where kerosene prces are ncreased substantally, the hgh underlyng nequalty n consumpton results n hgher-ncome households bearng a dsproportonately hgh share of the total burden. Therefore, large cost savngs (or, equvalently, greater protecton) can be provded through the use of bettertargeted subsdes, transfers, or other socal expendtures. The focus n ths paper s prmarly on evaluatng the magntude and dstrbuton of fuel subsdes, wth specal emphass on the lkely mpacts of ther removal on the poorest households as well as alternatve approaches to mtgatng these adverse effects. In ths sense, t sets out n detal how to mplement the methodologcal approach dentfed n Gupta and others (2000b, Appendx 3). To ths end, we draw on recent evaluatons for Bolva, Ghana, Jordan, Mal, and Sr Lanka. The format of the paper s as follows. In Secton II, we summarze the methodology used to calculate the magntude of consumer subsdes and ther dstrbuton across households at dfferent parts of the ncome dstrbuton. 8 In Secton III, we summarze the fndngs from some background studes on specfc country cases. It also descrbes some of the possble mtgatng measures that mght be adopted to protect lowncome households from hgher fuel prces. Fnally, Secton IV dentfes some general polcy lessons regardng the reform of fuel prcng. 6 For example, Iran, whch has some of the lowest prces n the world, s one of the most energy-ntensve countres n the world. 7 It s possble that other, second-best consderatons may dlute the argument for relatvely hgh taxaton of commercal energy. For example, hgh energy prces may encourage rural households to swtch to the use of already overexploted natural resources such as fuelwood. However, a more effcent polcy response to such overexplotaton may be to drectly manage these resources more effcently and to mprove access by these households to commercal energy sources. Note also that large cross-prce effects between alternatve commercal energy sources suggest that dstortng ther relatve prces s a very neffcent approach to achevng dstrbutonal or envronmental objectves. It s qute common for low kerosene prces to lead to neffcent substtuton toward kerosene and llegal adulteraton of desel or gasolne. Such substtuton toward low-taxed commodtes results n revenue losses so that hgher average tax rates are requred to rase a gven amount of revenue. 8 We use the term ncome dstrbuton to represent the dstrbuton of consumer welfare. We typcally use total household consumpton per household member to proxy for welfare. For smplcty, we refer to the dstrbuton of ths varable as the ncome dstrbuton.

8 6 II. MAGNITUDE, FINANCING, AND DISTRIBUTION OF CONSUMER SUBSIDIES 9 A. Consumer Prce Subsdes and Ther Fnancng To calculate the magntude of consumer subsdes, one needs to compare an actual consumer prce wth a reference prce that captures the true opportunty cost of domestc consumpton. Snce most countres are ether net exporters or net mporters of petroleum products the approprate reference prce s the relevant border prce,.e., world prces adjusted for trade and transport costs to the country s border. In the case of an exportng country, the border FOB prce mnus trade and transport margns represents the forgone revenue from consumng domestcally nstead of exportng. In the case of an mportng country, the border CIF prce plus trade and transport margns represents the cost of domestc consumpton. These reference prces can be vewed as effcent prces n that they maxmze the sum of consumer and producer surpluses. The dfference between the actual consumer prce and the reference prce for each petroleum product represents the unt subsdy (actual less than reference) or tax (actual greater than reference) for that product. It s not uncommon for some products to be subsdzed, whle others are taxed. Multplyng ths dfference by annual product consumpton and summng across products gves the total fuel subsdy (or the total cost of the subsdy). Comparng the actual to the reference prce also provdes the bass for dentfyng how much actual prces need to ncrease to elmnate the subsdy. Whether or not the full consumer subsdy s reflected n the government budget (.e., as an on-budget fscal cost) wll depend on the market structure of the petroleum sector and the government fnancng strategy. For example, consder the smplfed case presented n Fgure 2 where the country has a publcly owned refnery that produces Q c at an average (equal to margnal) producton cost of P c. Assume also that a publc sector frm s responsble for mportng fuel when demand exceeds domestc supply. The reference border prce s P m, the subsdzed consumer prce s P s, at whch consumers (ncludng households and frms) consume Q s, and mports are (Q s - Q c ). 10 The total consumer subsdy s thus represented by the area (A + B + C). If the government does not remburse the refnery or mporter for sellng at P s, then refnery profts are reduced by area A and the mporter ncurs a loss equal to areas (B + C). In other words, the consumer subsdy s fnanced by changes n the net profts of the two publc sector frms. Notce also that the refnery stll makes profts at the subsdzed prce, but that the mporter makes a loss. Such losses are often fnanced by borrowng from the bankng system and would be pcked up by a comprehensve measure of publc debt. Only f the government makes an explct transfer to the publc sector frms wll the consumer subsdy be made explct n the budget. If only the mporter s compensated for losses that s, the reducton n 9 More detals on these ssues are provded n Appendxes I and II. 10 To smplfy exposton we are mplctly assumng that domestc trade and dstrbuton margns are zero.

9 7 profts of the refnery are not rembursed the subsdy expendture wll fall short of the total consumer subsdy by area A. Ths often happens n practce. If petroleum products are taxed, the reference mport prce should nclude any exstng taxes. For example, n Fgure 2, f P m ncludes a tax and publc enterprses are stll requred to sell at P s, then the net prce to producers and mporters s reduced to P p ( = P s tax). In ths case, the net profts of publc enterprses are decreased further by area (D + E), and the government receves revenue (D + E). The tax smply shfts revenue from the publc enterprses to the budget, wth no change n publc sector fnances. Often governments wll respond to hgher border prces by decreasng fuel taxes, whle mantanng consumer prces at P s,.e., rasng P p, so as to avod any affect on proftablty of publc sector frms or consumer welfare. In ths case, government revenue absorbs the fscal cost of the ncrease n the subsdy. Improvng the fscal poston by the full extent of the consumer subsdy then nvolves ncreasng prces to the reference prce, P m, wth publc sector frms recevng (P m - tax). Fgure 2. Magntude and Fnancng of Consumer Subsdes P Demand P m P s P p P c A B C D E Q c Q s Q B. Welfare Impact of Hgher Fuel Prces Typcally the bulk of petroleum s consumed ndrectly through household consumpton of other goods and servces that use petroleum products as nputs. Therefore, the welfare effect of hgher fuel prces or, equvalently, lower fuel subsdes on household real ncomes wll depend both on the drect effect of hgher prces for petroleum products consumed by households and on the ndrect effect arsng from hgher prces for other goods and servces consumed by households to the extent that hgher petroleum costs are passed on to consumer prces.

10 8 Drect effects Calculatng the drect effect, and how t s dstrbuted across ncome groups, essentally requres nformaton on the level of drect consumpton of varous petroleum products (e.g., gasolne, kerosene, desel, and lquefed petroleum gas (LPG)) by ndvdual households across the natonal ncome dstrbuton. The man source of nformaton s typcally a household survey contanng expendtures by each household on ndvdual fuel products. A frst-order estmate of the drect real ncome effect of fuel prce ncreases can be calculated as follows. For each household one calculates the budget share of fuel expendture tems,.e., fuel expendtures dvded by total household consumpton. Multplyng budget shares by the percentage ncrease n prce due to the ncrease n fuel prces gves a frst-order estmate of the real ncome effect of the prce rse, whch assumes that fuel consumpton stays fxed. Ths overestmates the real ncome effect snce, n practce, households can reduce ths mpact by substtutng away from fuel. 11 The ncdence of the real-ncome effect can be analyzed by examnng how the magntude of the effect vares across the ncome dstrbuton. Typcally household per capta consumpton, possbly adjusted for famly composton, s taken as the best proxy of household welfare. Based on ths measure, households can be allocated to quntles or decles of the natonal dstrbuton. One can then analyze the ncdence of the real ncome effect by calculatng the average percentage real ncome loss for each decle or quntle. If the percentage real ncome loss s hgher (lower) for low ncome households, then the ncdence s sad to be regressve (progressve). Indrect effects Identfyng the magntude of the ndrect effect requres an estmate of the effect of hgher fuel costs on the prces of other goods and servces consumed by households. These prce effects can be estmated usng an nput-output table of the economy showng the energy ntensty of each sector and a smple model of the effect of hgher fuel costs on prces (such a model s presented n Appendx II). 12 As wth the drect effect, the ndrect real-ncome effect can be calculated by multplyng the budget shares of the varous goods and servces by the estmated percentage prce ncreases n these sectors. The ncdence of the ndrect effect can be determned by estmatng the effect separately for households across the natonal ncome dstrbuton. The total real ncome effect s calculated as the sum of the drect and ndrect real ncome effects, and the ncdence can also be determned by calculatng the average effect for households n dfferent parts of the ncome dstrbuton. In practce, reflectng the hgh 11 For a dscusson of the theoretcal foundatons of ths approach n the context of prce and tax reforms, see Ahmad and Stern (1984, 1991), Newbery and Stern (1987), and Deaton (1997). 12 Usng the dstrbutonal characterstc wdely used n tax reform analyss, Hughes (1987) summarzes the results of the applcaton of a smlar model to that used here to analyzng fuel prce ncreases n Indonesa, Thaland, and Tunsa. Hope and Sngh (1995) derve estmates for the drect mpact of prce ncreases for kerosene and electrcty durng the 1980s n Colomba, Ghana, Indonesa, Malaysa, Turkey, and Zmbabwe.

11 9 proporton of fuel consumed n the producton and dstrbuton of goods and servces, the ndrect effect accounts for over half of the total effect. Approxmaton errors How much, these frst-order mpacts overestmate the welfare effect of hgher fuel and other prces wll depend on how easly households and frms can n practce swtch ther consumpton away from fuels and goods and servces wth relatvely hgh-prce ncreases towards those wth relatvely low ncreases. For example, consder the case where fuel prces ncrease by 50 percent, the ntal budget shares for fuel and other goods and servces are 0.05 and 0.95 respectvely, and the ntal cost share of fuel n the producton of other goods and servces s 20 percent. The prce ncrease for other goods and servces wll depend on the ablty of producers to substtute between fuel and other nputs. Wth the assumpton of a zero elastcty of substtuton mantaned n the nput-output framework, the prce of other goods and servces would ncrease by 10 percent (0.2 * 0.5). Our frst-order estmate of the real-ncome loss would be 12 percent (1.5 * * ). Alternatvely, f we assume that producton s characterzed by a untary elastcty of substtuton, the prce ncrease for other goods and servces would be 8.4 percent ( * ). If we also assumed that consumer preferences exhbted untary (margnal) elastcty of substtuton, the total loss of real ncome would be 10.5 percent ( * ). The frst-order estmate s an upper bound on the loss n real ncome, and the bas ncreases wth the elastctes of substtuton n producton and consumpton. Reasonable estmates of the elastctes of substtuton n producton and consumpton are lkely to be less than one, however, so the alternatve estmate s lkely to be based downward. The frstorder estmate s much easer to calculate, provdes a bound on the real-ncome effect, and s lkely to closely approxmate a more sophstcated estmate. Fnally, snce one expects that short-run substtuton elastctes are smaller than long-run elastctes, the frst-order estmate wll be a better approxmaton of the short-run welfare mpact. C. Alternatve Approaches to Mtgaton The adverse mpact of fuel prce ncreases on already poor households s often hghlghted as a key constrant on the removal of fuel subsdes. It s, therefore, mportant that the removal of subsdes be accompaned by measures to mtgate the adverse effects on the poorest households. In addton, t s mportant to emphasze that the budgetary savngs from reducng fuel subsdes can be used to ncrease expendtures n areas that are typcally seen as havng hgher prorty, e.g., ncreasng access to or the qualty of educaton and health servces or physcal nfrastructure, or used to reduce taxes. In the context of reducng budget defcts, the counterfactual to fuel subsdy removal can be seen as a reducton of these socal expendtures, an ncrease n taxes, or hgher nflaton, all whch can have more adverse effects on the poor.

12 10 Ideally, governments would already have n place a socal protecton system that can be used to safeguard the real ncomes of the poorest households. 13 If such a system s n place and s well desgned and mplemented, then t provdes the most cost-effectve approach to socal protecton. In a sense, access to such a system promotes the generaton of effcency gans assocated wth structural adjustment more generally by addressng drectly any concerns regardng possble adverse affects on low-ncome households. In the context of prce ncreases resultng from subsdy reforms, the desred transfer to low-ncome households can be mantaned n real terms by nflaton ndexng the transfer. If such a system s not avalable, ether because no such program currently exsts or that whch exsts s not effectve, then a government s ablty to protect the poor from prce ncreases n the short term s restrcted. Introducng an effectve program from scratch obvously takes tme, but so too does reformng an exstng program. In ths stuaton, the gradual wthdrawal of subsdes may be warranted, whle a more effectve socal protecton mechansm s developed. Ths can be combned wth some shorter-term measures that ncrease the resources avalable to any exstng nformal socal assstance programs delvered through exstng networks of communty, relgous, or other nongovernmental organzatons. In addton, access costs to other publc servces, e.g., fees for educaton or health servces, can be reduced n the poorest rural localtes and urban dstrcts. The partcular approach used wll obvously depend on the specfc characterstcs of each country, especally the nature of ts socal nsttutons and the extent of exstng access to publc servces. 14 In order to sgnal to the publc ts ntenton to use the budgetary savngs more effectvely, government can specfy the expendtures to be fnanced by these savngs. For example, t can announce the allocaton of savngs to the expanson of access to qualty educaton and health servces, electrcty, or roads n rural areas. Or budgetary savngs can be used to promote low-cost urban transport networks or nvestments n electrcty, roads, and transport. In these cases, however, t s mportant to avod mmutable earmarkng, whch wll only reduce the government s ablty to respond to future challenges. III. COUNTRY CASE STUDIES In ths secton, we summarze the results from a set of country studes that appled the approach descrbed above to evaluate the mpact of proposed ncreases n the prces of petroleum products on household real ncomes. The country case studes analyze (actual or hypothetcal) prce reforms n Bolva, Ghana, Jordan, Mal, and Sr Lanka. The structure of ths secton follows that of the prevous secton. Subsecton A summarzes market structure, 13 Ths pont s made by Gupta et al (2000, p. 4) n the context of subsdes n general. They argue that the speed of subsdy reform can be faster when an effectve socal protecton system exsts n a country. Many of the countres n ther sample adopted a gradual approach to subsdy reform, whle smultaneously adaptng exstng socal protecton nstruments or establshng a new safety net. 14 A more detaled dscusson of alternatve approaches to protectng poor households s avalable n Gupta et al (2000). Coady, Grosh, and Hoddnott (2004) provde a detaled dscusson of alternatve methods of targetng transfers n developng and transton economes.

13 11 the approach to fuel prcng, and the magntude of fuel subsdes. Subsecton B descrbes the magntude of the real ncome effect from the wthdrawal of subsdes, and how t s dstrbuted across the drect and ndrect effects and across dfferent household groups. Fnally, Subsecton C dscusses alternatve approaches to protectng the real ncomes of poor households n the face of fuel prce ncreases. We conclude by descrbng the polcy responses of the varous governments. Table 1 provdes a summary of the reforms and ther mpacts dscussed n more detal below. A. Market Structure, Prcng Regme, and Fuel Subsdes Of the fve countres, only Bolva s an exporter of crude ol, and all of the countres except Mal have ol refneres. The Tema Ol Refnery n Ghana produces about 70 percent of Ghana s consumpton requrements and buys crude from Ngera at a dscount on world prces. The refnery has a monopoly on the producton and mportng of refned products, but recent reforms have allowed for prvate tenderng for mportng. The dstrbuton of petroleum products s prvatzed, although wth a sgnfcant publc sector presence. Untl early 2005, Jordan had a tradton of buyng crude ol at concessonal prces from neghborng countres (orgnally Iraq, more recently Saud Araba). Ths ol s refned by the Jordan Petroleum Refnery Company nto petroleum products, whch are sold on the domestc market at controlled prces. The government remburses the refnery for any losses relatve to a cost-plus bass. The Ceylon Petroleum Corporaton n Sr Lanka mports crude ol to produce refned petroleum products. Domestc producton meets only around 50 percent of domestc consumpton requrements wth the gap met from mports. The producton, mport, and dstrbuton of petroleum products were the exclusve doman of the refnery up to 2002 when prvate sector partcpaton n mport and dstrbuton was allowed. Mal mports all of ts consumpton needs. All the countres, except Jordan, prevously used and have recently abandoned automatc prcng formulas for settng the domestc prces of petroleum products. Bolva ntroduced a prcng formula n 1996 as part of a major restructurng of the petroleum sector, but abandoned t n the late 1990s. Ghana ntroduced a formula n January 2003, whle smultaneously ncreasng prces by an average of 90 percent. The formula was effectvely abandoned n early 2003 when contnued ncreases n world prces were not passed onto consumers. Sr Lanka ntroduced a prcng formula n 2002, but ths was suspended n early Mal ntroduced a formula n 1994, but abandoned t n Untl md-2005 domestc prces tracked world prces and ncluded a sgnfcant element of taxaton. The concessonal prces on crude ol receved by Jordan helped avod the need for ether hgh domestc prces or automatc formulas untl recently. The decson to drectly regulate domestc prces below border prces has resulted n substantal fscal costs n the form of explct budgetary subsdes and forgone revenue. Bolva kept domestc prces low through a combnaton of explct subsdes, lower tax rates, low refnery prces, and subsdzed crude ol for the refnery. Prces were fxed from md-2000 to early 2004, wth only small ncreases snce then. In 2004, the total consumer subsdy reached 4.3 percent of GDP, of whch only 2.6 percentage ponts showed up on the budget n the form of explct subsdes and forgone revenue. Low prces also led to

14 12 Table 1. Summary of Emprcal Evdence on Real Income Impact of Fuel Subsdes Bolva Ghana Jordan Mal Sr Lanka Background Automatc formula ntroduced n 1996 as part of sectoral reforms - abandoned n late 1990s. Low prces mantaned by explct subsdes, lower producer prces and decreasng taxaton. Resulted n ncreasng subsdes, smugglng abroad and demand shortages. Automatc formula ntroduced n January 2003 together wth an average prce ncrease of 90 percent. Formula subsequently abandoned when world ol prces ncreased, but new one (whch ncludes taxes) has recently been ntroduced. By 2004, recept of crude ol on concessonal terms was expected to end, and fuel subsdes were rsng. In early 2004, the government announced ntentons to gradually elmnate subsdy over four-year perod and n lne wth a prcng formula that would nclude taxes. Prcng mechansm s ad hoc. Untl early 2004, domestc prces ncreased to cover most of world prce rse (aded by substantal apprecaton of currency aganst the US dollar). Monthly prcng formula was ntroduced n 2002 wth Rs. 2 per month cap on ncrease allowed for each product. Ths was suspended n early VAT on desel was suspended. Now effectvely an ad hoc prcng regme. Sze of fuel subsdy (percent of 2004 GDP) Total subsdy equal to 4.3 percent, of whch only 2.6 percent shows up n budget n form of explct subsdy and forgone revenues Explct subsdy of 2.2 percent (In 2003 also reached 2 percent of GDP) 3.2 percent (In 2005, ths was projected to ncrease to 8.5 percent f no acton taken. Subsequent prce ncreases meant that the actual subsdy bll was 5.8 percent of GDP n 2005) 2 percent loss n tax revenue due to decreasng tax rates and exemptons for some sectors (especally mnng). 2.1 percent (but also energy taxes VAT on desel and excses on gasolne - rased 1.2 percent) Prce ncreases (smulated) Rasng prces of gasolne and desel to world prces and LPG by 67 percent. Restorng a formula prce that ncludes taxes Elmnatng subsdes and mposng a standard general sales tax at 16 percent. Rasng ol prces by 34 percent to mmc ncreases n world prce between 2001 and Rase average ol prces by 36.7 percent to reach formula prces. Kerosene n.a. 49 percent 80 percent 34 percent 94 percent LPG 67 percent 108 percent 48 percent 34 percent n.a. Desel 40 percent 67 percent 77 percent 34 percent 40 percent Gasolne 40 percent 17 percent No change snce tax>16 percent 34 percent 14 percent Other n.a. 50 percent (Fuel Ol) percent 34 percent n.a Total 50 percent 50 percent 68 percent (excl. gasolne) 34 percent 37 percent

15 13 Table 1. Summary of Emprcal Evdence on Real Income Impact of Fuel Subsdes (concluded) Bolva Ghana Jordan Mal Sr Lanka Aggregate real ncome mpact (range from bottom to top ncome quntles) Drect 1.6 percent ( ) 1.9 percent ( ) 2.0 percent (3-1.7) 0.9 percent ( ) 1.2 percent ( ) Indrect 3.3 percent ( ) 6.7 percent ( ) 2.4 percent ( ) 0.8 percent ( ) (1.1 f electrcty ncluded: ) 1.2 percent ( ) Total 5.0 percent ( ) 8.5 percent ( ) 4.4 percent ( ) 1.7 percent (regressve, U) (1.9 f electrcty ncl.) 2.4 percent ( ) Share of subsdy receved by poorest 40 percent Drect (kerosene) n.a percent (39.7 percent) 22.9 percent (33.3 percent) 22.6 percent (35.4 percent) 27.2 percent (40.0 percent) Indrect n.a percent 19.8 percent 24.4 percent 23.0 percent Total 15.3 percent 23.0 percent 21.2 percent 23.9 percent 25.1 percent Polcy response Increased prce by 10 percent for gasolne and 23 percent for desel the latter was reduced to 15 percent after socal unrest. No plans for further ncreases. No specfc mtgatng expendtures mplemented. Prces ncreased on average by 50 percent n February 2005 together wth package of expendtures to mtgate mpact on poor. Set up ndependent authorty to mplement prcng system. In July 2005, government ncreased domestc prces by around 27 percent to reduce subsdes to 3 percent of GDP n Also ntroduced mtgatng measures. Because of further ncreases n world prces, domestc prces were ncreased by 14 percent n September Government ntends to fully lberalze the prcng of fuel products by Consderng rentroducng a prcng formula. Prces of all petroleum products were ncreased n the frst half of 2006 to lmt ncreases n fuel subsdes.

16 14 substantal smugglng to neghborng countres facng hgher domestc prces. 15 In Ghana, explct subsdes to the refnery and dstrbutors to compensate for below-formula prces reached 2.2 percent of GDP n 2004, equvalent to around 3.2 percent on an annualzed bass. Tax revenue from fuel products accounted for nearly 4 percent of GDP n Consequently, the ncdence of the subsdes fell entrely on the refnery and dstrbutors. In Jordan, explct net subsdes on petroleum products were 3.2 percent of GDP n 2004 and, n the absence of prce ncreases, would have ncreased to over 8 percent n Gasolne has tradtonally been taxed, generatng revenues equvalent to 1.2 percent of GDP n 2004, whch were used to cross-subsdze products. Despte several domestc prce ncreases, net fuel subsdes are expected to reach about 2 percent of GDP n In Mal, aded by a substantal currency apprecaton aganst the dollar, domestc prces were above nternatonal prces untl md Snce then, prce ncreases have been restraned reducng excse tax rates. Petroleum products have tradtonally been taxed, wth the tax component n pump prces rangng from over 20 percent for kerosene to nearly 50 percent for gasolne. The revenue losses have been estmated at 2 percent of GDP for 2004, largely on reducng excse tax rates. In Sr Lanka, formula prces ncluded value-added tax (VAT) on desel and gasolne as well as excses on all products. The VAT on desel was elmnated n August 2005 and the domestc prces of desel and kerosene were below mport party. The total fscal cost of subsdes was estmated at 2.1 percent of GDP on an annualzed bass, but subsequent prce ncreases meant that the outturn for 2005 was approxmately a subsdy level equvalent to 1.1 percent of GDP. Domestc prce controls have resulted n sgnfcant dstortons on petroleum products that vary across products. 16 In Bolva, the ex-refnery domestc prces for regular gasolne and desel fell to about 70 percent of nternatonal prces by early 2005, whle for LPG the domestc prce fell to one half of the reference export prce. To brng the domestc consumer prces n lne wth the nternatonal average and restore excse tax rates to ther 2000 level, retal prces for desel and regular gasolne would have to ncrease by around 40 percent, and prces for LPG would have to double. In Ghana, rasng prces to formula prces would have requred a 50 percent average ncrease n prces n early 2005, rangng from 17 percent for gasolne to 108 percent for LPG. In Jordan, domestc prces for gasolne were substantally above equvalent mport prces n early 2005, wth an mplct tax rate on regular gasolne of 42 percent and on premum of 60 percent. The prce ncreases necessary to elmnate the subsdes on other products were substantal, rangng from 59 percent for desel to 65 percent for fuel ol used n the generaton of electrcty. Movng to a stuaton where petroleum products faced the same 16 percent general sales tax as other fnal goods, the requred prce ncreases for these products would be even more substantal, rangng from 77 percent for desel to nearly 84 percent for fuel ol used n electrcty generaton. 15 Smlar problems wth smugglng subsdzed fuel to neghborng countres wth hgher prces have been encountered n Iran, Iraq, Ngera, and Yemen. 16 Typcally subsdy rates are hgher for kerosene than for desel, although absolute desel subsdes are typcally substantally hgher due to hgher consumpton levels. Gasolne usually carres lower subsdes and s often taxed. The mport cost of gasolne, desel, kerosene, and LPG are approxmately equal.

17 15 B. Real Income Effect of Increasng Petroleum Product Prces It s clear from above that restorng the prcng formula n each country would requre substantal ncreases n domestc petroleum prces. Usng the approach outlned earler, ths secton presents estmates of the lkely mpact of these prce changes on household real ncomes, and how t s dstrbuted across households wth dfferent ncome levels. These estmates were obtaned from ndvdual country-specfc analyses that evaluated the effects of ncreasng fuel prces to nternatonal levels (see Table 1 for the requred ncreases n each country). The drect effect on households depends on the total budget share for petroleum products as well as the dstrbuton of expendture across ndvdual products. 17 The drect effect on household real ncomes of the prce ncreases consdered ranges from 0.9 percent for Mal (where smulated prce ncreases were lowest) to 2 percent for Jordan (where smulated prce ncreases were hghest). In all cases, the drect effect s ether approxmately dstrbutonally neutral (Bolva and Mal), or regressve (Ghana, Jordan and Sr Lanka). Where t s regressve, t reflects the combnaton of the hgh mportance of kerosene for the poorest households and relatvely hgh prce ncreases for ths product. In Bolva, kerosene was not an mportant fuel, whle n Mal low-ncome households are hardly affected by gasolne prce ncreases. Table 2 presents estmated budget shares by product and ncome group for each country. The budget shares for petroleum products range from percent. The 6.6 percent share for Jordan ncludes spendng on electrcty, whch accounts for 2.3 percent of total spendng. In general, desel s manly consumed n producton, wth households consumng relatvely small quanttes. In all countres except Ghana and Jordan, low-ncome households allocate a lower proporton of ther budget to energy. In Ghana, the relatvely hgh budget share for low-ncome households reflects the relatvely hgh usage of kerosene. In Jordan, where the budget shares are smlar across all ncome groups, electrcty accounts for a relatvely hgh share of energy consumpton the budget share for petroleum products s hghest n the top quntles. In all countres, gasolne s consumed prmarly by hgher ncome households, whereas kerosene s relatvely more mportant n the budgets of lower ncome households. In Ghana, Mal, and Sr Lanka, where kerosene s extensvely used, t s the domnant component of the energy budget for lower-ncome households, accountng for over 67 percent n all cases. In Jordan, where electrcty access s almost unversal, kerosene expendtures are relatvely less 17 Typcally kerosene s used for lghtng and heatng, especally where households do not have access to electrcty. Desel s typcally used n goods and passenger transport, agrculture (e.g., pumps and engnes) and ndustry the latter two channels are ncorporated through the ndrect effects. Gasolne s typcally used for transport. Desel and kerosene are near perfect substtutes snce large quanttes of kerosene can be added to desel fuel wthout much mpact on vehcle performance low kerosene prces relatve to desel thus usually result n the dverson of kerosene to the automotve desel sector. Adulteraton of gasolne wth kerosene n other than small quanttes can cause damage to vehcles. In the long run, gasolne and desel are close substtutes, e.g., through the swtchng from gasolne- to desel-powered vehcles wth an assocated worsenng of ar polluton.

18 16 mportant for lower ncome households. LPG consumpton s low for all households n Ghana and Mal and relatvely hgh for lower ncome groups n Bolva and Jordan, but more mportant for hgher ncome groups n Sr Lanka. The drect effect on household real ncomes of the prce ncreases consdered ranges from 0.9 percent for Mal (where smulated prce ncreases were lowest) to 2 percent for Jordan (where smulated prce ncreases were hghest). In all cases, the drect effect s ether approxmately dstrbutonally neutral (Bolva and Mal), or regressve (Ghana, Jordan and Sr Lanka). Where t s regressve, t reflects the combnaton of the hgh mportance of kerosene for the poorest households and relatvely hgh prce ncreases for ths product. In Bolva, kerosene was not an mportant fuel, whle n Mal low-ncome households are hardly affected by gasolne prce ncreases. Table 2. Fuel Budget Shares, by Income Quntles (In percent of total consumpton) Quntle 1 Quntle 2 Quntle 3 Quntle 4 Quntle 5 All Bolva LPG Gasolne and desel Total Ghana Kerosene LPG Gasolne Total Jordan Kerosene LPG Gas regular Gas premum Desel Electrcty Total Mal Kerosene Gasolne Desel Charcoal Electrcty Total Sr Lanka Kerosene LPG Gasolne and desel Electrcty Total

19 17 The ndrect effect on household real ncomes ranges from 1.1 percent for Mal to 6.7 percent for Ghana. In all cases, the magntude of the ndrect effect s at least equal to that of the drect effect. Wth the excepton of Sr Lanka, the ndrect effect s domnant when the electrcty prce effect s ncluded n ths effect. In all cases except for Bolva, the dstrbuton of the ndrect effect s ether very slghtly progressve or neutral. In Bolva, where the analyss focused only on the ndrect effect of hgher fuel prces on the prces of food and publc transportaton, the ndrect effect was strongly regressve. The total effect,.e., combned drect and ndrect effects, ranges from nearly 2 percent for Mal to 8.5 percent for Ghana. A smple averagng across country studes suggests that a 50 percent average ncrease n fuel prces results on average n a 4.6 percent decrease n real ncomes. In all cases except Mal, the dstrbuton of the total effect s regressve. For Ghana, Jordan, and Sr Lanka, ths reflects the dstrbuton of the drect effect, gven that the ndrect effect s approxmately neutral n these countres. In Bolva, the ndrect effect s more regressve than the drect effect. In Mal, the total effect s approxmately neutral n fact, the percentage decrease n real ncome s somewhat smaller for the mddle quntle compared to the top and bottom quntle. Although the removal of fuel subsdes s regressve (or, at best, neutral), t s stll the case that a substantally hgher proporton of the aggregate burden s borne by hgher ncome groups. Ths, of course, reflects the very unequal dstrbuton of ncome n these countres: a regressve dstrbuton of the burden smply means that the share borne by low-ncome groups s greater than ther share of total ncome. Conversely, the share of low-ncome groups n the benefts from fuel subsdes s also relatvely low. For example, the share of the poorest 40 percent of households n the total benefts from fuel subsdes ranges from 15.3 percent for Bolva to 25.1 percent for Sr Lanka. Therefore, between percent of subsdy benefts accrue to the rchest 60 percent of households. 18 Even an equal unform transfer to all households would be better targeted than exstng subsdes, snce 40 percent of benefts would accrue to the poorest 40 percent of households. The very poor targetng of fuel subsdes s not surprsng; almost any unversal consumpton subsdy wll dsproportonately beneft the rch snce they, by defnton, account for a relatvely hgh proporton of total ncome and consumpton. Even the drect subsdy, whch reflects subsdes to kerosene, nvolves substantal leakage to the non-poor, wth percent accrung to the top 60 percent of households. One would expect a reasonably effectve drect transfer program to target substantally better than ths, and a recent revew of transfer programs n developng countres by Coady et al (2004) provdes ample emprcal evdence of such programs n developng countres. The followng secton expands further on ths ssue. 18 Smlar fndngs have been reported for other countres: n the early 1990s, the top 20 percent of households n Venezuela receved sx tmes the subsdy receved by the bottom 30 percent. A study of fuel subsdes n Indonesa n 1999 found that only 20 percent of subsdy benefts went to the poorest 30 percent of households.

20 18 C. Identfyng Alternatve Approaches to Mtgaton Although a relatvely low proporton of fuel subsdes reaches low ncome households, t s stll the case that these households can suffer a substantal decrease n real ncome as a result of ther wthdrawal. For example, the percentage decrease n real ncome for the poorest ncome quntle ranged from 1.8 percent for Mal to 9.1 percent for Ghana. Even the relatvely low declnes for Mal (1.8 percent) and Sr Lanka (2.9 percent) should be assessed n the context of the extreme poverty of these households, and some households wthn ths ncome group wll suffer more substantal losses than others. These mpacts are also relatvely large when compared to growth n real ncome. For example, the 1.8 percent declne n real ncomes for Mal s relatvely large when compared to recent per capta realncome growth rates, whch have averaged 2 3 percent per annum. It s mportant for any reform strategy to dentfy alternatve and more effectve mechansms for provdng and approprately targetng protecton. More generally, beyond an exclusve emphass on the poor, t s mportant to dentfy more desrable uses for the budgetary savngs from the wthdrawal of fuel subsdes. The very poor targetng performance of unversal fuel subsdes means that t should be possble to dentfy more effectve socal protecton mechansms that protect the poorest households from ncreases n fuel prces and stll have substantal savngs left over to allocate to hgher prorty expendtures or tax cuts that beneft the populaton more broadly. Persuadng the populaton that budgetary savngs wll ndeed be used n ths fashon s a crucal component of the poltcal economy of the reform process. 19 Moreover, f t s lkely that any savngs wll be squandered on programs that beneft an elte and poltcally connected mnorty, then (unfortunately) fuel subsdes n realty may be an attractve second-best polcy. In such a stuaton, mprovng governance should be the hghest prorty reform. 20 Use of exstng safety nets In practce, how to compensate poor households wll depend on whether an effectve socal safety net already exsts. If t does, then some of the budgetary savngs can be used to expand the program, e.g., expandng elgblty for cash or raton card transfers as well as ncreasng ther value. A well-desgned transfer program can avod dstortng economc decsons, whle both ensurng extensve coverage of poor households and mnmzng leakage to hgher ncome groups. Fuel subsdes, on the other hand, encourage an neffcent level and composton of fuel consumpton and, at the same tme, transfer substantal publc resources to hgher ncome groups. In the case of Sr Lanka, a natonal safety net does exst, and the study demonstrated that ts core component, the Samurdh food stamps program, was 19 In Indonesa, for example, publc percepton was that any expanson of exstng safety net expendtures would lkely be captured by hgher ncome households. To counteract ths, the government ntroduced an unprecedented cash (as opposed to n-knd) transfer, delvered drectly to most elgble households va an extensve network of post offces. Ths program helped avod a repetton of the rots, deaths and wdespread socal and poltcal dsrupton that accompaned prevous fuel prce ncreases. It s stll unclear how effectvely ths program has been mplemented. 20 See Esfahan (2002) for a poltcal economy nterpretaton of polcy reform n the context of fuel subsdes.

21 19 substantally better targeted than kerosene subsdes. Under ths program, 52 percent of transfers accrued to the poorest 40 percent of households compared to 41 percent of kerosene subsdes. The program also had extensve coverage of lower ncome households 63 percent of the poorest 20 percent of households receved transfers compared to the 78 percent who benefted from fuel subsdes. Although exstng safety net programs may provde an attractve alternatve to kerosene subsdes, often ther current desgn and mplementaton effectveness could be substantally mproved. In the case of Sr Lanka, the World Bank has been dentfyng better targetng mechansms. Under the exstng program, elgblty s determned by local admnstrators through Samurdh organzatons and s, n prncple, based on the assessment of household ncome and economc status. The proposed targetng mechansm was to be based on a proxy-means targetng approach that dentfed poor households by attachng a numercal weght to varous household soco-economc characterstcs to calculate a household score. Households would then be chosen f ther score was below a certan threshold score. Smulatons for Sr Lanka ndcated that ths scheme could sgnfcantly mprove targetng, wth the proporton of transfers accrung to the bottom 40 percent ncreasng to 67 percent (manly at the expense of the hghest ncome group) and coverage of the poorest 20 percent of households ncreasng to 82 percent. A smlar smulaton was undertaken for Ghana estmated that 65 percent of transfers would accrue to the poorest 40 percent of households compared to a correspondng 40 percent for kerosene subsdes. Of course, such performance assumes effectve mplementaton, so suffcent resources also need to be devoted to mprovng mplementaton as well as desgn of programs. Snce ths may take tme, the potental for usng exstng safety net programs may be lmted n the short run. For example, n Jordan, the prmary safety net program, The Natonal Ad Fund, had been establshed n 1986 after the elmnaton of food subsdes. Although t s the most comprehensve transfer program n the country and manages to target transfers to the poorest households, ts coverage of the poorest households was extremely low. Over 50 percent of ts transfers accrued to the poorest quntle and over 75 percent to the bottom 40 percent of households. Ths compares very favorably to the dstrbuton of kerosene subsdes n Jordan, where less than 15 percent of benefts reach the poorest quntle. However, the program covered only 14 percent of the lowest ncome decle and less than 12 percent of the lowest quntle. The challenge s, therefore, to mantan the program s ablty to channel funds to low-ncome households wthout ncurrng hgher leakage as the program expands to cover a substantally greater proporton of low ncome households. Snce ths could only be acheved by nvestng resources n mprovng the desgn and mplementaton of the program, ths program could not be used to effectvely mtgate the adverse mpact of subsdy wthdrawal n the short run, so that alternatve approaches would be requred. Short-term mtgaton measures In the absence of an effectve socal safety net, the partcular approach used to provde short-term protecton wll depend on the specfc characterstcs of each country, especally the nature of socal nsttutons. Governments can often undertake a number of mmedate expendture measures to sgnal the government s ntenton to protect the poorest households from the adverse effects of prce hkes. For example, user charges for educaton and health