UNIVERSITY OF WAIKATO. Hamilton New Zealand. Household Energy Demand and the Equity and Efficiency Aspects of Subsidy Reform in Indonesia

Transcription

1 UNIVERSITY OF WAIKATO Hamlton New Zealand Household Energy Demand and the Equty and Effcency Aspects of Subsdy Reform n Indonesa Susan Olva Unversty of Calforna, Davs John Gbson Unversty of Wakato Department of Economcs Workng Paper n Economcs 6/06 August 2006 Correspondng Author John Gbson Department of Economcs Unversty of Wakato, Prvate Bag 305, Hamlton, New Zealand Fax: +64 (7) Emal: jkgbson@wakato.ac.nz

2 Abstract The proper desgn of prce nterventons n energy markets requres consderaton of equty and effcency effects. In ths paper, budget survey data from 29,000 Indonesan households are used to estmate a demand system for fve energy sources, whch s dentfed by the spatal varaton n unt values (expendtures dvded by quanttes). We correct for the varous qualty and measurement error bases that result when unt values are used as proxes for market prces. The prce elastctes are combned wth tax and subsdy rates to calculate the margnal socal cost of prce changes for each tem. The results suggest that even wth hgh levels of nequalty averson there s a case for reducng the large subsdes on kerosene n Indonesa, supportng the reforms that have been announced recently. Keywords demand elastctes energy subsdes unt values JEL Classfcaton D2, Q3 Acknowledgements We are grateful to Catherne Morrson Paul and semnar audences at the Unversty of Calforna, Davs and the 2006 IAAE conference on the Gold Coast for ther helpful comments. All remanng errors are our responsblty.

3 I. Introducton Energy demand s rsng rapdly n developng countres. The energy prcng polces of those countres are therefore ncreasngly mportant to the effcent use of the world s energy supply. Energy markets n some developng countres are hghly dstorted, especally from consumer subsdes (IEA, 999). For example, the Government of Indonesa spent over US$3 bllon dollars on consumer fuel subsdes n These subsdes have a major effect on the overall energy balance n Indonesa because households account for about 45 percent of total energy consumpton. There also are large fscal effects, wth about one-quarter of the government budget (and about fve percent of Indonesan GDP) gong on fuel subsdes (Sen and Steer, 2005). Dramatc reforms have been attempted by the Indonesan government n response to ths escalatng cost of fuel subsdes. In October 2005 the subsdsed prce of kerosene was rased 86 percent, from Rp 700 per ltre (US 7 cents) to Rp 2000 per ltre (US 9 cents). The prces for desel and gasolne were rased by approxmately 90 percent, followng on from ncreases of 30 percent n March 2005 (Table ). These earler prce ncreases dd not apply to kerosene. Moreover, a tmetable has been set for completely phasng out fuel subsdes, wth those on gasolne and desel tmetabled to go by the end of 2006 and those on kerosene by the end of These energy subsdes are meant to be replaced wth a set of targeted subsdes, whose benefts are to be desgned so that they are restrcted to low-ncome groups (Kompas, 2005; Jakarta Post, 2005). It s unclear whether these ambtous plans for reform wll be realsed. The frst reason s that despte the substantal prce rses enacted n 2005, there s stll a long way to go f Indonesan fuel prces are to be set at world levels. The kerosene prce n October 2005 was only 3 percent of the world prce, whle gasolne and desel prces were about two-thrds of the world level (Table ). The second reason for doubtng that these ambtous plans can be acheved s that

4 many prevous attempts at reformng energy prce polcy n Indonesa have faled because of the resultng poltcal dffcultes. Attempted reforms n 2003 were reversed after wdespread protests whle the prce rses n 998 are beleved to have precptated the downfall of the Suharto regme (BBC, 2005; Economst, 2005). Moreover, these subsdes have been long-term features of the Indonesan economy, datng back to the md-970s (Dck, 980). The subsdzaton of especally kerosene has been seen as one feasble way of meetng equty objectves, because the poor are presumed to use kerosene as ther man cookng fuel. 2 However there are debates about whether the poor are the man benefcares of kerosene subsdes (Sumarto and Saryahad, 200). Indeed, even though there was early evdence that a dsproportonate share of the subsdy was beng captured by rcher urban households, the subsdy polcy contnued to be strengthened and kerosene prces were held below one-ffth of the world level as far back as 980 (Ptt, 985). The am of ths paper s to provde emprcal evdence to help assess whether the proposed reforms of energy prce polcy n Indonesa are lkely to be welfare-enhancng. Specfcally, the equty and effcency effects of prce changes n the household energy sector are analysed. To acheve ths am, the margnal socal costs of ndrect taxes and subsdes are calculated for fve fuels and household energy sources: kerosene, gasolne, ol, LPG, and electrcty. These margnal socal costs depend on the rate at whch household welfare falls as prces ncrease, and on the rate at whch net publc revenue rses (Ahmad and Stern, 984). If a reform s optmally desgned, the costs n terms of socal welfare of the last Rupah of government expendture saved by cuttng subsdes (or rasng taxes) on each good should be equal. To obtan the two requred parameters the welfare dervatve and the revenue dervatve nformaton s needed on tax and subsdy rates, consumpton patterns, and aggregate demand responses. Thus, the analyss follows the recommendaton of Newbery (2005) to use the basc 2 Ths relance on energy subsdes reflects the lmted capacty for ncome transfers, whch s a feature of many developng countres. 2

5 prncples of publc fnance to ntroduce order nto dscussons of how energy taxes and subsdes mght ratonally be set. Ths emprcal analyss s needed because the prevous lterature on energy demand n Indonesa does not provde clear gudance for evaluatng the subsdy reforms. On the one hand, Ptt (985) concluded that the prce of kerosene should be ncreased on both equty and effcency grounds. In part, ths concluson rested on an estmate from household survey data that the elastcty of kerosene demand wth respect to ts own prce was -.03, suggestng that prce dstortons would create large substtuton effects. However, other estmates of the prce elastcty of demand for kerosene are not nearly so large: Koshal et al. (999) use tme seres data to estmate a long-run own-prce elastcty of demand of only If the demand for kerosene really s ths prce nelastc the effcency losses from the subsdy mght be evaluated as beng less mportant than the presumed benefts n the form of transfers to the poor. 3 Consequently, once dstrbutonal concerns were taken nto account by Yzhak and Lews (996) they reached the opposte concluson to Ptt (985); specfcally, Ytzhak and Lews concluded that there would be aggregate mprovements n welfare from ncreasng the subsdy on kerosene. In addton to provdng analyss that can help to assess the recent subsdy reforms n Indonesa, a further am of the paper s to demonstrate how a common source of data n developng countres can be used to estmate the matrces of demand elastctes that are needed for evaluatng energy prcng polces. Whle there are several tme seres studes of energy demand n developng countres, and specfcally Indonesa (Dahl and Kurtub, 200; Koshal et al. (999)), these have two major lmtatons: frst, the tme seres s often very short. For example, McRae (994) uses just 5 annual observatons to estmate gasolne demand models for Indonesa and several other Asan countres. Second, the growng 3

6 relance on targetng means that aggregate elastctes from tme seres studes may no longer be the most relevant ones for assessng prcng nterventons. Instead, dsaggregated elastctes for target groups such as rural households or the poorest quntle of households may be needed. Household surveys are one source of data that can be used to calculate dsaggregated energy demand elastctes, wth prces proxed by unt values (expendtures dvded by quanttes) and the dentfcaton of demand responses comng from spatal rather than temporal varaton n prces. Indeed ths was the source of data used by Ptt (985). However, there are at least three problems wth the use of such data. In contrast to market prces, unt values reflect household-specfc qualty and reportng error effects, and are subject to sample selecton effects because they are unavalable for non-purchasng households. Deaton (990) shows that gnorng these problems (as Ptt dd) s lkely to lead to elastctes beng too large n absolute terms, causng substtuton possbltes to be overstated and tendng to rase the calculated effcency costs of subsdes. Therefore, n ths paper we use the estmaton methods developed by Deaton (990) to correct the bases that result from usng unt values rather than actual market prces. Our example may help economsts elsewhere who are consderng the use of household survey data for estmatng demand responses, partcularly n the context of developng countres where spatal prce varaton s greatest and tme seres are typcally shortest. The rest of the paper s organsed as follows. The next secton dscusses the theoretcal framework underlyng the calculaton of the socal costs of margnal tax and subsdy reform. In Secton III, we descrbe the data and econometrc estmaton methods used n ths paper. Partcular attenton s pad to the method developed by Deaton (990) for estmatng prce elastctes from household survey data. Secton IV presents the estmaton 3 Notng agan that there s contenton about how much of the subsdy s captured by the poor. Ths ssue s 4

7 results for the dsaggregated household energy demand system. Secton V looks at the mplcatons of the elastcty estmates for the reform of household energy and fuel prces n Indonesa. Ths secton also dscusses evdence about whether the poor beneft the most from kerosene subsdes. Secton VI presents our conclusons. II. The Margnal Tax Reform Approach In an nfluental paper, Ahmad and Stern (984) proposed a methodology for evaluatng margnal tax and subsdy reforms. The crucal parameter of the Ahmad and Stern model s what they call the margnal socal cost (MSC) of a unt of publc revenue. Ths s made up of two components: a welfare dervatve and a revenue dervatve. Specfcally, a decrease n the subsdy rate, τ on good, (or equvalently, a tax ncrease) wll cause welfare to change at rate V / τ and net government revenue to change at rate R / τ. The rato of these two dervatves gves the margnal socal cost, λ of rasng one unt of net revenue (savng one unt of expendtures) by decreasng the subsdy rate on good : V / τ λ =. () R / τ Goods wth low λ ratos are those that are canddates for ether a tax ncrease or a subsdy reducton. When all the ratos are the same there s no further scope for benefcal reform. Ths approach can be mplemented by notng that the welfare dervatve (the numerator of ()) s just the rato of two average budget shares: w w ~ ε (Deaton, 997). The frst average budget share, w s weghted to reflect equty consderatons: ε w ε = M m= ( x m n ) x ε m m w m M m= x m (2) dscussed n the penultmate secton of the paper. 5

8 where w m s the budget share for good n household m, x m and n m are the total expendture and sze of household m, and ε s the coeffcent of nequalty averson. 4 A range of values of ε between zero (no nequalty averson) and two (a hgh degree of nequalty averson) are commonly used to see whether tax reform recommendatons are robust to partcular ethcal judgements (Ahmad and Stern, 984). In terms of the calculaton of equaton (2), the larger s ε the closer the average budget share wll be to budget shares of the poorest households n the sample. The second average budget share s the so-called plutocratc budget share (Pras, 957) whch s based on ratos of total expendture (rather than averages of ratos) and gves the bggest weghts to the rchest households: 5 w ~ = M m= x m w m M m= x m. (3) The denomnator of the λ-rato represents the effcency aspect of tax- and subsdynduced prce changes. A gven prce change wll produce a larger net revenue effect, the greater s the total consumpton of the good and the less the substtuton away from taxed goods: ε w ~ w λ = τ θ + ~ + + τ w k τ k + τ k θ k w~. (4) The total consumpton of the good s shown by w%, whle the substtuton effects are shown by θ k, the dervatve of the budget share for good k wth respect to the (log) prce of good. The tax factor gves the share of tax n the fnal prce. For example, n Indonesa household purchased of ol face a VAT rate of 0 percent, so the tax factor s 0.0 ( + 0.0) = 0.09, whle subsdes 4 Consder judgements about the effect of takng Rp000 from someone to gve some of t to a person wth half the ncome and destroyng the rest (e.g., due to effcency losses). When ε=0 the judge would approve of ths transfer only f the poorer person receved all Rp000. But when ε takes the values of (or 2) the amount the poorer person receves has to be only Rp500 (or Rp250 f ε=2) n order for the resultng dstrbuton to be judged as gvng the same level of socal welfare as before the transfer (Creedy, 996). 5 The plutocratc budget share s wdely used n Consumer Prce Index calculatons. Accordng to calculatons by Deaton (998a) the average consumer accordng to plutocratc budget shares n the Unted States was located at about the 75 th percentle of the dstrbuton of household expendtures. 6

9 meant that kerosene prces snce October 2005 are only 3 percent of the prces n other countres, so the tax factor s 0.69 ( 0.69) = The frst term of the denomnator n equaton (4) measures the own-prce dstortonary effect of the tax or subsdy. If t s large and postve, as would be the case for a heavly subsdsed and prce elastc good, the term wll contrbute to a small λ and would ndcate the low socal cost of rasng net revenue by decreasng the subsdy on ths good. The last term s the sum of the tax factors multpled by the cross prce elastctes, and captures the effects on other goods (and the resultng net revenue changes) from the change n the tax on good. Ths framework for evaluatng margnal reforms can be contrasted wth the requrements for calculatng optmal tax and subsdy rates. More nformaton s needed to calculate optmal tax and subsdy rates because explct utlty functons are needed for agents snce demand responses have to be evaluated at the optmum, whch s a pont that may be far away from the current poston of the economy. Furthermore, estmates have to be made of how behavoural responses themselves change n response to taxes and subsdes (Madden, 996). III. Data and Estmaton Methods Data from the consumpton module of the 999 SUSENAS survey are used for 28,964 households located on Java. 6 Respondents n ths survey are asked to recall ther expendtures over the past month for approxmately 300 dfferent products. For food, fuels and electrcty they are also asked about the quantty purchased so that unt values can be derved. These 6 Ths household budget survey s only carred out every three years and the 999 results were the most recent when the analyss was begun. Java contans approxmately 60 percent of the Indonesan populaton and, economc actvty s even more heavly concentrated there, so the results should stll be relevant to natonwde prce reform. 7

10 unt values are needed because the survey does not collect market prces. The survey s samplng procedure nvolves selectng clusters of up to sxteen households wthn census enumeraton areas. Ths spatal clusterng encourages the assumpton that households wthn each cluster face the same prces and ths feature s exploted by the estmaton method used below. The fve energy products consdered electrcty, LPG, kerosene, gasolne and ol contrbute almost 4.4 percent of the average household budget, rangng from LPG at 0.4 percent to electrcty at.82 percent (Table 2). 7 The frst three columns of Table 2 descrbe some of the characterstcs of the unt values. Items lke electrcty and kerosene have unt values avalable for almost every household whle nformaton for tems lke LPG and gasolne s less readly avalable. Means of unt values are also shown n the table. These are computed from those households who make market purchases of the commodty under consderaton. On average, consumers n 999 pad approxmately Rp. 460 per ltre for kerosene, whle gasolne and ol cost more than Rp.,000 per ltre. The mean prces of these energy products vary somewhat by locaton, wth rural households payng a slghtly hgher prce than urban households for all of the commodtes except ol, perhaps because of hgh transport costs n rural areas (Appendx Tables a and b). The coeffcent of varaton ndcates the degree of heterogenety wthn each group, whch s least for commodtes lke kerosene and gasolne where there s lttle qualty varaton. The procedure used to get the prce responses that are needed for the margnal socal cost calculatons starts wth a two-equaton system of budget shares (w Gc ) and unt values (v Gc ) that are both functons of the unobserved prces, (p Hc ): w G c N = α G = β ln x c + γ z c + θ G H ln p +( f +ug c ) G G H= H c G c (5) 7 Whle gasolne and ol can also be consdered as transport fuels they are wdely used n the household sector to power domestc generators, and n fact are more wdely consumed than LPG, especally n the rural sector. 8

11 ln v G c = α G = β ln G x c + γ G z c + N H = ψ G H ln p H c + u G c (6) the G ndcates goods, ndcates households and the c ndexes clusters. Amongst the explanatory varables, x s total expendture of household, p H are the unobserved prces, z s a vector of other household characterstcs, f Gc s a cluster fxed-effect n the budget share for good G and u 0 G c and u G c are dosyncratc errors. The estmaton proceeds n three stages, whch are dscussed fully n Deaton (997). In the frst stage, the procedure removes the household-specfc effects of ncome and other demographc characterstcs from the budget shares and unt values. To do so, equatons (5) and (6) are estmated usng OLS, where n addton to x and z, the specfcaton also controls for all cluster fxed effects, ncludng those of unobserved prces, so that the 0 0 G, γ G, β G, and γ G β parameters can be estmated consstently, even n the absence of market prce data. These four parameters are used to create adjusted budget shares and unt values that have the qualty effects due to ncome and other factors removed. For example, f rcher households buy hgher grade gasolne (whch wll have a hgher unt value) the frst stage regresson can account for ths and the adjusted unt value s more lke a prce because prces should not vary across households n the same communty. The frst stage regressons also produce the resduals that are needed n the second stage for estmatng the covarances that are used to correct for the effect of any measurement error n unt values and budget shares. The error terms, e e 0 Gc and Gc, from equatons (5) and (6) contan all the varablty n w Gc and v Gc that are not explaned by x, z, or the cluster fxed effects. Assumng a sngle prce per cluster, the unexplaned varaton around the cluster mean can ndcate measurement error. In the second stage of the Deaton procedure, a between-clusters errors-n-varables regresson s appled to the (adjusted) average budget shares and unt values, whch have 9

12 been purged of household characterstcs at the frst stage. If t were not for the effect of prces on cluster-wde qualty varaton, the parameters estmated at the second stage would be suffcent for calculatng the prce responses. Instead, a separablty theory of qualty (Deaton 988) has to be used to dentfy the prce effects at the thrd and fnal stage. Full detals of the estmaton method can be found n Deaton (997). One feature of the procedure s that the budget share equaton (5) s unlke Tobt-style models n that t pays no specal attenton to non-purchasng households, who have budget shares of zero. Snce the revenue effect of a tax ncrease does not depend on whether demand changes take place at the extensve or ntensve margns, when studyng tax and revenue reform one needs to nclude all households, whether they purchase or not (Deaton, 990). Therefore, equaton (5) s smply vewed as a lnear approxmaton to the regresson functon of the budget share condtonal on the rght-hand-sde varables, averagng over both zeros and non-zeros n much the same way that an aggregate demand functon does (Deaton, 997). 8 IV. Econometrc Results Table 3 contans results from the frst stage (wthn-cluster) estmaton of the budget share and unt value equatons. The coeffcents reported are for the effects of log total expendtures on budget shares, β 0 and on log unt values, β plus summary statstcs for each equaton. The tables also nclude the elastcty of quantty demanded wth respect to total expendtures, whch depends on coeffcents from both the budget share and unt value equatons. 9 The other (unreported) varables used at the frst stage nclude (log) household sze, a set of demographc varables (the number of household members n each of thrteen 8 Other studes applyng Deaton s method to household survey data also follow the same specfcaton and nclude households wth budget shares of zero. See, for example, Ncta (2004). 9 See equaton 5.36 of Deaton (997) for a dervaton. 0

13 age and sex categores as a rato of household sze), and nne educatonal dummes. These varables are based on those used by Deaton (990) n hs study of food demand from the same survey n an earler year. As can be seen from Table 3, the frst stage estmaton of the budget share equatons explans from 27 percent of the varaton for ol to 47 percent for kerosene. More of the varaton n unt values s explaned, rangng from 3 percent for gasolne to 66 percent for ol. 0 LPG, gasolne and ol attract postve β coeffcents ndcatng luxury goods whose expendture elastctes are greater than one, and whose budget shares wll rse more than proportonally as household expendture rses. The total expendture elastcty for kerosene s 0.44, whch s consstent wth the estmates made by Ptt (985) and Ytzhak and Lews (996), ndcatng that kerosene s a necessty. 0 Appendx Tables 2a and 2b show that rural households tend to have hgher expendture elastctes than urban households. For nstance, rural households have expendture elastctes of 3.08 and 4.08 for gasolne and LPG compared wth 2.38 and 2.36 for urban households. These results mply that rural households are lkely to have larger proportonate ncreases n demand for these products as ther ncome rses. The qualty elastctes, β are estmated from the effect of changes n the logarthm of total expendture on the log of the unt value. These show how unt values dffer between rch and poor households n the same communtes, where ths varaton s presumed to reflect the purchase of hgher qualty commodtes by the rch. Yet accordng to the estmates shown n column 5 of Table 3, wth the excepton of kerosene and ol, all of the qualty elastctes are negatve although none are very large. Thus, rather than rcher households payng more per unt they pay less, presumably because of bulk purchases or more favourable 0 In developed countres, kerosene has been found to be an nferor good (Kennedy, 974). A plausble explanaton for ths s that n developed countres, other energy sources are used for cookng and lghtng.

14 electrcty tarffs for these households than are avalable to the poor. However, the magntude of these effects s small, and t s only for the unt value of ol, where the qualty elastcty s 0.07, that there would appear to be any sgnfcant wthn-cluster varaton n unt values. The small sze of these qualty effects s consstent wth other studes usng the Deaton method and s also ntutvely sensble because the qualty varaton wthn a household energy source s lkely to be less than that wthn a food group (e.g. steak versus hamburger). Therefore t s the ablty of ths method to also mtgate the effect of measurement error n unt values that becomes ts crucal feature (Deaton, 997). Table 4 contans the estmated own- and cross-prce elastctes for Java. The symmetry restrctons from demand theory have been mposed on these estmates n order to mprove precson of the estmates. In addton to the fve energy sources, there s an extra row and column for all other goods, the estmates for whch are obtaned from the homogenety and addng-up restrctons. The elastctes are condtonal not only on household sze and the dummy varables for household characterstcs mentoned above, but also on a set of provnce and urban dummy varables. In addton to the prce elastctes, the table also nclude bootstrapped estmates of standard errors. To calculate these standard errors, 000 random draws are taken from the second stage data (.e., the cluster average budget shares and unt values, after the effect of household total expendtures and other characterstcs have been controlled for). For each of these random draws, all of the elastctes are recalculated, n effect creatng 000 versons of Table 4. The length of the nterval around the mean of each bootstrapped elastcty that contans 68.3 percent of the bootstrap replcatons s calculated and one-half of ths nterval s used as the estmate of the standard error. The ratonale s that f the dstrbuton of the It s not possble to add them at the frst (wthn-cluster) stage because the cluster fxed effects would oblterate them. 2

15 elastcty estmates was normal, s the fracton of a normal random varable wthn two standard devatons of the mean (Deaton, 997). The own-prce effects are well determned as are several cross-prce effects. All of the estmated own-prce elastctes are negatve, as they should be. The estmated own-prce elastcty of demand for kerosene n Java s found to be (wth a standard error of 0.). Dsaggregatng ths result, the demand for kerosene appears consderably more prce elastc n rural areas than n urban areas, possbly reflectng the lower ncomes of the rural populaton. Specfcally, a 0 percent ncrease n the prce of kerosene wll brng about a 3.3 percent declne n the quantty of kerosene bought by the rural households, but only 6.4 percent declne n the urban sector (Appendx Tables 3a and 3b). These are close to the estmates obtaned by Ptt (985) who also used household survey data, but dd not correct for the possble bases caused by unt values. The smlarty of elastcty estmates, and the contrasts wth the much more nelastc tme seres estmates, suggests that Ptt s concluson that t would be both equtable and effcent to reduce the subsdy on kerosene may n fact have been correct. The own-prce elastctes are also large (n absolute terms) for electrcty and LPG, although only the one for electrcty s precsely estmated. Ths suggests that subsdes wll have caused a consderable amount of substtuton nto these products. Amongst the crossprce elastctes, the demand for electrcty, LPG, gasolne and ol wth respect to the prce of kerosene s negatve, and n some cases, large n absolute value. Ths apparent complementarty may reflect the lmted but specalzed use of these energy sources by Indonesan households, who use electrcty manly for lghtng, whle they use LPG manly for cookng. Even wth the spatal prce varaton provded by 900 clusters, and wth the mposton of theoretcal symmetry restrctons, a majorty of the cross-prce effects are 3

16 mprecsely estmated. Possbly more success would come from usng a sample from all regons of Indonesa, because of the greater mpact of transport costs n causng relatve prce dfferences, but expandng the sample would also brng the rsk of greater uncontrolled heterogenety. V. Margnal Socal Cost Calculatons Table 5 shows the effcency effects of cuttng subsdes (or rasng taxes) on each of the goods, dstngushng between the terms n the denomnator of the margnal socal cost formula (equaton (4)). The frst column shows the tax factors (calculated from comparson of the world and domestc prces), whle the second column shows the own prce elastctes of qualty and quantty together. The product of the frst and second columns, whch s shown as the thrd column, gves the own-good contrbutons to the tax dstorton that would be caused by a margnal ncrease n prce. As can be seen from the table, the own effects for these energy products are reasonably small, wth the excepton of kerosene and electrcty, for whom rasng prces would save the largest amount from the subsdy budget (gnorng any cross-prce effects). Wth regard these cross-prce effects, they are largest for gasolne and LPG. Accordng to the combned results n the last column of Table 5, and notng that nothng yet has been sad about dstrbutonal ssues, kerosene and then LPG are the most attractve canddates for prce rses. In Table 6, the results of brngng n the equty effects are reported for a range of the dstrbutonal parameter, ε. The frst two columns are for ε = 0, where there are no dstrbutonal concerns; the cost-beneft ratos are smply the recprocals of the last column n Table 5 and gve the same rankng n terms of relatve margnal socal costs as was gven by the revenue dervatves. The margnal cost of rasng kerosene prces s lowest, and for rasng ol prces (whch are already taxed) s hghest amongst the energy sources. However, 4

17 all of the λ for the energy sources are much lower than for other consumpton ndcatng the general desrablty of removng all energy subsdes. As we move across to the rght-hand sde of Table 6, and the ε parameter ncreases, the equty columns gve larger values to the goods most heavly consumed by the poor and relatvely smaller values to those consumed by households that are better off. For ε = 0.5, kerosene receves the hghest socal weght, wth w ε w ~ greater than unty. Consequently, as ε ncreases further, kerosene loses ts place as the most attractve canddate for a prce rse, becomng the second most attractve when ε = and the thrd most attractve when ε = 2. Indeed, wth an nequalty averson parameter of ε = 2 the lowest socal cost of reduced government expendture (or equvalently addtonal revenue) would come from rasng LPG prces, followed by rasng gasolne prces. The best canddate for a subsdy at these hgher nequalty averson levels s electrcty. When the results are dsaggregated nto the rural and urban sector the recommendatons are largely the same. In the rural sector kerosene s the best canddate for prce ncreases when nequalty averson levels are low, wth LPG becomng the best canddate at hgher nequalty averson levels (Appendx Table 5b). Electrcty and gasolne appear to be best canddates for any prce subsdy n rural areas, as these commodtes attracts the hghest λ ratos. In the urban sector LPG s consstently ranked as the commodty wth the lowest socal cost of prce ncreases and the attractveness of rasng kerosene prces dmnshes wth the degree of nequalty averson that s assumed. The results suggest that at low levels of nequalty averson kerosene s the best canddate for reduced subsdes but that there may be some reluctance by polcy makers who are more nequalty averse to cut kerosene subsdes. Gven ths possble ambguty over the recommended polcy reform, addtonal evdence on household kerosene use may be useful. Table 7 reports for each sector (rural and urban) and each quntle of the household 5

18 expendture dstrbuton, the average quantty of kerosene consumed, the kerosene budget share and the proporton of households who use kerosene stoves. Whle kerosene does have a larger share of the budgets of poor households, the dfference s only marked n comparson to the rchest quntle of households. Indeed, n the rural sector, the budget share s almost constant (at.4 percent) across the frst four quntles. The results n the table also show that t s not true that kerosene s predomnantly the fuel of the poor. In rural areas there s a monotonc ncrease n the proporton of households usng kerosene stoves and n the consumpton level of kerosene when movng from poorer to rcher quntles. In urban areas the three mddle quntles all have hgher rates of kerosene stove usage than the poorest quntle and the consumpton level of kerosene s roughly constant across quntles. Thus, a majorty of the kerosene subsdes wll not be captured by the poor, especally n rural areas, and the replacement of energy subsdes wth targeted ncome subsdes s lkely to be both more effcent and more equtable. VI. Conclusons In ths paper we have used household budget survey data from Indonesa to estmate the margnal socal cost of ndrect taxes and subsdes on fve household energy sources. Regardless of assumptons about nequalty averson, all of these energy sources are attractve canddates for prce ncreases, when compared wth the socal cost of revenue rased from taxes on other goods and servces. The calculatons also show that reducng expendture on subsdes by allowng further ncreases n the consumer prce of kerosene would be desrable, takng nto account both effcency and equty. Thus the suggested drectons of reform are n lne wth the actual prce reforms carred out n 2005, and they ndcate the desrablty of furtherng these reforms. 6

19 The motvaton for these reforms n Indonesa has been the heavy cost of mantanng consumer energy subsdes as world energy prces soar. There are lkely to be other developng countres facng a smlar set of ssues. From a methodologcal pont of vew, the avalablty of household survey data n other developng countres suggests that smlar analyses could be carred out n settngs where prce elastcty estmates are needed to evaluate the effects of prce dstortons n consumer energy markets. 7

20 References Ahmad, E. and Stern, N. (984) The theory of reform and Indan ndrect taxes Journal of Publc Economcs 25 (3): BBC News (2005) Indonesa clashes over fuel hke [On-lne] Avalable Creedy, J. (996) Measurng ncome nequalty Australan Economc Revew 29(2): Dahl, C., and Kurtub. (200) Estmatng ol product demand n Indonesa usng a contegratng error correcton model OPEC Revew 25(): -25. Deaton, A. (988) Qualty, quantty, and spatal varaton of prce Amercan Economc Revew 78(3): Deaton, A. (990) Prce elastctes from survey data: Extensons and Indonesan results Journal of Econometrcs 44(3): Deaton, A. (997) The Analyss of Household Surveys: A Mcroeconometrc Approach to Development Polcy. Baltmore: The John Hopkns Unversty Press. Deaton, A. (998) Gettng prces rght: what should be done? Journal of Economc Perspectves 2():

21 Dck, H. (980) The ol prce subsdy, deforestaton and equty Bulletn of Indonesan Economc Studes 6(3): Economst (2005) Indonesa s ropey Rupah September st. [On-lne] Avalable Internatonal Energy Agency (999) World Energy Outlook: Lookng at Energy Subsdes, Gettng the Prces Rght. Internatonal Energy Agency: Pars. Jakarta Post (2005). Indonesa may cut fuel subsdes by as much as 60 percent by November [On-lne] Avalable Kompas (2005) Dana kompensas BBM: 40 juta orang dapat subsd secara langsung [Onlne] Avalable Kennedy, M. (974) An economst model of the world ol market Bell Journal 5(4): Koshal, R., Koshal, M., Boyd, R., and Rachmany, H. (999) Demand for kerosene n developng countres: a case of Indonesa Journal of Asan Economcs 0(3): Madden, D. (996) Margnal tax reform and the specfcaton of consumer demand systems Oxford Economc Papers 48(4):

22 Newbery, D. (2005) Why tax energy? Towards a more ratonal polcy The Energy Journal 26(3): -40. Ncta, A. (2004) Effcency and equty of a margnal tax reform: ncome, qualty and prce elastctes for Mexco World Bank Polcy Workng Paper No. 3266, World Bank, Washngton DC. Ptt, M.M. (985). Equty, externaltes and energy subsdes: The case of kerosene n Indonesa Journal of Development Economcs 7 (3): Sadoulet, E., and A. de Janvry. (995). Quanttatve Development Polcy Analyss Johns Hopkns, Baltmore. Sen, K. and Steer, L. (2005) Survey of recent developments Bulletn of Indonesan Economc Studes 4(3): Sumarto, S. and Suryahad, A. (200) Kerosene: Government subsdes and household consumpton SMERU Newsletter No. 2, March-Aprl, 200, pp -4 Ytzhak, S. and Lews, J. (996) Gudelnes on searchng for a Dalton-mprovng tax reform: An llustraton wth data from Indonesa World Bank Economc Revew 0(3):

23 Table. Recent Changes n Regulated Fuel Products Prces, Rupah per ltre and percentage change (percent of world prce) Jan 2003-Feb 2005 Mar-Sep 2005 October 2005 onwards % change % change % change Prce Gasolne,80 (36%) Prce 2,400 (56%) from Feb 05 Prce 33% 4,500 (68%) from Mar 05 from Feb 05 88% 49% Kerosene (household) 700 (%) 700 (%) 0% 2,000 (3%) 86% 86% Automotve Desel,650 (33%) 2,00 (45%) 27% 4,300 (68%) 05% 6% Source: World Bank (2005) Table 2. Commodtes, Sample Szes and Budget Shares for Java, 999 Commodtes Number of households wth unt values (a) Mean unt value (b) Coeffcent of varaton of unt value (c) Number of clusters wth unt values (d) Number of households n clusters n (d) (e) Percentage shares of total expendture (f) Electrcty 26, ,87 26, LPG 2,32, , Kerosene 23, ,846 23, Gasolne 5,735, ,430 5, Ol 3,698, ,430 5, Note: (a) s the number of households wth a well-defned unt value, whch equals the number of purchasng households mnus those who report n rregular unts. (b) n Rupah. Table 3. Frst Stage Estmates: Effect of Total Expendtures on Quantty and Qualty for Java Budget Share Equaton Unt Value Equaton Commodtes o o β t( β ) R 2 β t ( β ) R 2 ε Electrcty LPG Kerosene Gasolne Ol Note: o β s the dervatve of the budget share wth respect to log total expendtures, β s the dervatve of the (log) unt value wth respect to log total expendtures (a.k.a. the qualty elastcty ), R 2 s for the budget share and unt value regressons, and ε s the expendture elastcty of quantty. 2

24 Table 4. Symmetry Constraned Estmates of Own and Cross Prce Elastctes for Java, 999 Electrcty LPG Kerosene Gasolne Ol Other Cons Electrcty (0.02) (0.02) (0.03) (0.05) (0.02) (0.06) LPG (0.2) (0.89) (0.97) (4.42) (0.30) (3.5) Kerosene (0.03) (0.08) (0.) (0.26) (0.04) (0.22) Gasolne (0.3) (0.85) (0.65) (.28) (0.5) (0.87) Ol (0.5) (0.2) (0.32) (0.54) (0.4) (0.67) Other Cons (0.00) (0.00) (0.00) (0.00) (0.00) (0.37) Note: Standard error n ( ); Results for All Other Goods derved from homogenety and addng up restrcton. Table 5. Effcency Aspects of Prce Reform n Java wth Symmetry Restrcton Commodtes τ θ Own effect Cross effects Total + τ ~ w Electrcty LPG Kerosene Gasolne Ol Other consumpton Table 6. Equty Effects and Cost-Beneft Ratos for Prce Reform n Java wth Symmetry Restrctons ε = 0 ε = 0. 5 ε = ε = 2 Commodtes w w λ w w~ λ w w λ w w λ Electrcty LPG Kerosene Gasolne Ol Other consumpton Table 7. Kerosene Consumpton, Budget Shares and Kerosene Stove Use Budget share (%) Quantty (ltre/month) Kerosene Stove Use Quntle Rural Urban Rural Urban Rural Urban (= poorest) (=rchest)

25 Appendx Table a. Commodtes, Sample Szes and Budget Shares for Urban Java, 999 Commodtes Number of households wth unt values (a) Mean unt value (b) Coeffcent of varaton of unt value (c) Number of clusters wth unt values (d) Number of households n clusters n (d) (e) Percentage shares of total expendture (f) Electrcty 2, , LPG 2,088, , Kerosene 0, , Gasolne 3,53, ,03.0 Ol 2,206, , Note: See Table 2. Appendx Table b. Commodtes, Sample Szes and Budget Shares for Rural Java, 999 Commodtes Number of households wth unt values (a) Mean unt value (b) Coeffcent of varaton of unt value (c) Number of clusters wth unt values (d) Number of households n clusters n (d) (e) Percentage shares of total expendture (f) Electrcty 4, , LPG 233, Kerosene 3, ,023 3,66.35 Gasolne 2,204, , Ol,492, , Note: See Table 2. 23

26 Appendx Table 2a. Frst Stage Estmates: Effect of Total Expendtures on Quantty and Qualty for Urban Java Budget Share Equaton Unt Value Equaton Commodtes o o β t( β ) R 2 β t ( β ) R 2 ε Electrcty LPG Kerosene Gasolne Ol Note: See Table 3. Appendx Table 2b. Frst Stage Estmates: Effect of Total Expendtures on Quantty and Qualty for Rural Java Budget Share Equaton Unt Value Equaton Commodtes o o β t( β ) R 2 β t ( β ) R 2 ε Electrcty LPG Kerosene Gasolne Ol Note: See Table 3. 24

27 Appendx Table 3a. Symmetry Constraned Estmates of Own and Cross Prce Elastctes for Urban Java, 999 Electrcty LPG Kerosene Gasolne Ol Other Cons Electrcty (0.03) (0.02) (0.03) (0.06) (0.03) (0.08) LPG (0.4) (0.62) (0.67) (.69) (0.9) (.87) Kerosene (0.04) (0.) (0.8) (0.6) (0.04) (0.29) Gasolne (0.) (0.47) (0.26) (0.23) (0.4) (0.62) Ol (0.2) (0.2) (0.28) (0.55) (0.6) (0.7) Other Cons (0.00) (0.00) (0.00) (0.00) (0.00) (0.35) Note: See Table 4 Appendx Table 3b. Symmetry Constraned Estmates of Own and Cross Prce Elastctes for Rural Java, 999 Electrcty LPG Kerosene Gasolne Ol Other Cons Electrcty (0.03) (0.04) (0.06) (0.00) (0.02) (0.08) LPG (2.3) (2.85) (8.9) (0.8) (.92) (0.5) Kerosene (0.07) (0.2) (0.25) (0.00) (0.09) (0.36) Gasolne (0.02) (0.0) (0.00) (.09) (0.00) (.02) Ol * (0.2) (0.39) (0.77) (0.00) (0.20) (0.92) Other Cons (0.00) (0.00) (0.00) (0.00) (0.00) (0.26) Note: See Table 4 25

28 Appendx Table 4a. Effcency Aspects of Prce Reform n Urban Java wth Symmetry Restrcton Commodtes τ θ Own effect Cross effects Total + τ ~ w Electrcty LPG Kerosene Gasolne Ol Other consumpton Appendx Table 4b. Effcency Aspects of Prce Reform n Rural Java wth Symmetry Restrcton Commodtes τ + τ θ Own effect Cross effects Total ~ w Electrcty LPG Kerosene Gasolne Ol Other consumpton Appendx Table 5a. Equty Effects and Cost-Beneft Ratos for Prce Reform n Urban Java wth Symmetry Restrctons ε = 0 ε = 0. 5 ε = ε = 2 Commodtes w w λ w w~ λ w w λ w w λ Electrcty LPG Kerosene Gasolne Ol Other consumpton Appendx Table 5b. Equty Effects and Cost-Beneft Ratos for Prce Reform n Rural Java wth Symmetry Restrctons ε = 0 ε = 0. 5 ε = ε = 2 Commodtes w w λ w w~ λ w w λ w w λ Electrcty LPG Kerosene Gasolne Lubrcant Ol Other consumpton