ejournal of Tax Research

Transcription

1 ejournal of Tax Research Volume 5, Number July 2007 CONTENTS 5 Fscal Msperceptons Assocated wth Tax Expendture Spendng: the Case of Pronatalst Tax Incentves n Sngapore Poh Eng Hn 40 What Future for the Corporate Tax n the New Century? Rchard S. Smmons 59 Chartes for the Beneft of Employees: Why Trusts for the Beneft of Employees Fal the Publc Beneft Test Fona Martn 7 Responsve Regulaton and the Uncertanty of Tax Law Tme to Reconsder the Commssoner s Model of Cooperatve Complance? Mark Burton 05 Unravellng the Mysteres of the Oracle: Usng the Delph Methodology to Inform the Personal Tax Reform Debate n Australa Chrs Evans 35 Worawan Chandoevwt and Bev Dahlby Atax, The Unversty of New South Wales ISSN

2 (2006) vol. 5, no., pp The Margnal Cost of Publc Funds for Excse Taxes n Thaland Worawan Chandoevwt and Bev Dahlby Abstract We extend the Ahmad and Stern (984) framework for calculatng the margnal cost of publc funds (MCF) for excse taxes n Thaland by ncorporatng non-tax dstortons caused by (a) envronmental externaltes, (b) publc expendture externaltes, (c) market power n settng prces, (d) addcton, and (e) smugglng or tax evason. Our calculatons, based on our benchmark parameter values, ndcates that the MCFs are for fuel excse taxes, 2.87 for tobacco excse taxes, 2.32 for alcohol excse taxes and.080 for the VAT. Usng pro-poor dstrbutonal weghts does not change the relatve socal margnal cost of rasng revenues through the excse taxes. INTRODUCTION Excse taxes, commodty taxes, and mport dutes are the most mportant sources of tax revenue n most developng countres, represented 60.6 percent of tax revenues n developng countres n compared to 32.5 percent n OECD countres. Among Southeast Asan countres, relance on taxes on goods and servces ranges from 3.0 percent of total tax revenue n Brune to 82.3 percent n Camboda. In Thaland, commodty taxes represent 59. percent of total tax revenues, wth excse taxes contrbutng 25.6 percent of tax revenues. Gven ts heavy relance on excse taxes, the equty and effcency effects of excse taxes are mportant aspects of tax polcy n Thaland. In ths paper, we contrbute to the analyss of excse tax polcy n Thaland by computng the margnal cost of publc funds (MCF) for the excse taxes on alcohol, tobacco, and fuel and for the value-added tax. We utlze the basc analytcal framework for measurng the MCFs developed by Ahmad and Stern (984) by usng estmates of the own-and cross-prce elastctes of demand for 0 categores of goods and servces n Thaland. Ths allows us to capture the nterdependence of the varous commodty tax bases n Thaland n computng the MCFs. In addton, we extend the basc Ahmad and Stern framework by ncorporatng n the computaton of The authors would lke to thank Patanayut Santyanont and Suwmon Fakthong for ther effcent research assstance and especally to Davd Ryan for hs advce on econometrc ssues. We are grateful to an anonymous referee for hs/her comments. We are responsble for any errors or omssons. Major fundng for ths research project came from the Canadan Internatonal Development Agency (CIDA). Other fundng was provded by the Human Resources and Socal Development Program, Thaland Development Research Insttute. Senor Research Specalst at the Thaland Development Research Insttute, Bangkok, Thaland. Emal: worawan@tdr.or.th. Professor at the Department of Economcs and Fellow of the Insttute of Publc Economcs, Unversty of Alberta, Alberta, Canada. Emal: bdahlby@ualberta.ca. Calculatons based on data n Tanz and Zee (2000, Table 2, p.304). See also Burgess and Stern (993, Table 5, p.773) on developng countres relance on goods and servces taxaton. 35

3 THE THEORY OF THE MCF 2 the MCFs the non-tax dstortons created by (a) envronmental externaltes, (b) publc expendture externaltes, (c) addcton, (d) market power, and (e) smugglng. Our analyss, based on our benchmark parameter values, ndcates that the MCFs are for fuel excse taxes, 2.87 for tobacco excse taxes, 2.32 for alcohol excse taxes, and.080 for a VAT ncrease. We also use pro-poor dstrbutonal weghts and data on the spendng patterns of 90 household groups n Thaland to calculate dstrbutonally-weghted MCFs, but ths procedure does not change the rankng of the socal margnal cost of the excse taxes. Fnally, we show that a revenue-neutral margnal tax reform reducng the excse tax rates on alcohol and tobacco by one percentage pont and ncreasng the fuel excse tax would result n a net effcency gan equal to.72 Baht for every addtonal Baht of fuel tax revenue. The paper s organzed as follows. The next secton outlnes the basc theory of the MCF and how we ncorporate the varous non-tax dstortons, such as externaltes, market power, smugglng and addcton, n the formula for the MCF. Then we descrbe the parameters used n the calculatons tax rates, budget shares, elastctes of demand, and measures of the non-tax dstortons. In the subsequent secton, we present our calculatons of the MCFs, ncludng the contrbutons of the varous nontax dstortons to the overall MCFs, and the potental net gan from a revenue neutral margnal tax reform. The penultmate secton descrbes the computatons of the dstrbutonally-weghted MCFs. The fnal secton contans our conclusons. The margnal cost of publc funds measures the loss ncurred by socety n rasng addtonal tax revenues. It has emerged as one of the most mportant concepts n the feld of publc economcs, playng a key role n the evaluaton of tax reforms, publc expendture programs, and other publc polces, rangng from tax enforcement to prvatzaton. Tax Dstortons Taxes can dstort the allocaton of resources n an economy by alterng taxpayers consumpton, savngs, labor supply, and nvestment decsons. The MCF s a summary measure of the addtonal dstorton n the allocaton of resources that occurs when a government rases addtonal revenue. However, mnmzng the effcency losses s not the only crtera for evaluatng tax measures because taxes that mpose heavy burdens on low ncome ndvduals are also costly taxes. The MCF concept can be used to combne equty or dstrbutonal concerns wth effcency effects n a summary measure of the total cost to a socety of rasng tax revenue. In ths paper we use the MCF concept to evaluate the man excse taxes mposed by the government of Thaland. Our basc model follows the approach poneered by Ahmad and Stern (984). For general surveys of the methodology and ssues n evaluatng commodty tax reforms, see Ray (997), Santoro (no date) and Dahlby (forthcomng, Ch. 3). Our man methodologcal contrbuton s the ncluson of non-tax dstortons n the computaton of the MCFs for excse taxes. Intally, to smplfy the analyss, we wll gnore dstrbutonal ssues by assumng that the economy only conssts of one ndvdual whose well-beng s represented by the ndrect utlty functon, V(q, I), where q s the 2 Ths secton s based on Dahlby (forthcomng, Chapters 2 and 3). 36

4 vector of consumer prces and I s lump-sum ncome. Later we show how to ncorporate dstrbutonal concerns n the measurement of the socal margnal cost of publc funds (SMCF). Total tax revenues R = n = t x depend on the tax rates, t, mposed on the n commodtes, denoted by the x s, that are consumed by the ndvdual. A money measure of the harm mposed on the ndvdual n rasng an extra dollar of tax revenue by ncreasng tax rate t s defned by the expresson: λ( q, I) MCFt dr = () t dv where λ(q,i) s the ndvdual s margnal utlty of ncome. In defnng the MCF t, t s assumed that dr/ s postve,.e. that the government s operatng on the upwardslopng secton of ts Laffer curve wth respect to t. If a tax ncrease s fully reflected n the consumer prce of the product and does not affect the prces of other products dq / = and dq j / = 0 and usng Roy s theorem, the followng expresson for the MCF can be derved: 3 t MCF τ x = b = n n dx j b + b + τ j = j j ε x t j j = j dq (2) where ε j s the elastcty of demand for commodty j wth respect to the prce of commodty, b j s the budget share of commodty j, and τ j = t j /q j s the ad valorem tax rate on commodty j. Ths expresson for the MCF ndcates the mportance of the tax rates on other commodtes n evaluatng the MCF for any partcular commodty tax. If commodty j s a substtute for commodty and ε j > 0, then an ncrease n the tax on commodty wll boost the demand for commodty j. The addtonal tax revenue collected from the tax on commodty j s a measure of the welfare gan from the mprovement n the allocaton of resources n the jth commodty market arsng from the ncrease n t, and ths effect reduces the MCF t. Conversely, f the commodty j s a complement for, then rasng the tax rate on commodty wll exacerbate the tax dstorton n the allocaton of resources n the jth commodty market by reducng the consumpton of j. The resultng loss of revenue from the tax on j s a measure of the addtonal dstorton caused by the tax ncrease on commodty. The theory of optmal commodty taxaton emphaszes the nteracton between the demands for the taxed commodty and lesure. In partcular, the Corlett and Hague 3 It further assumed that the taxes do not affect relatve nput prces. Ths s an admttedly strong assumpton gven that some fuel excse taxes are pad by busnesses. 37

5 Rule for optmal commodty taxaton states that hgher taxes should be leved on the commodtes that are more complementary wth lesure n order to offset the dstorton n the labour-lesure decson caused by our nablty to tax lesure drectly. We wll consder the mplcatons of the nteracton between commodty taxes and lesurelabour supply decsons for the measurement of the MCFs for excse taxes n the secton dealng wth the estmaton of the demand elastctes. As Devarajan et al (200) have stressed, t s very mportant to consder both tax and non-tax dstortons n measurng the MCF t. In the followng sectons, we wll show how we can ncorporate the welfare effects from non-tax dstortons envronmental externaltes, publc expendture externaltes, mperfect competton, addcton, and tax evason n the measurement of the MCF. Envronmental externaltes Suppose that household s consumpton of commodty, 2 2 utlty of household 2, such that U ( x,x ) x, drectly affects the j. The margnal external beneft from household s consumpton of commodty s equal to d E = (/λ 2 )( U 2 / x ). In the case of a harmful externalty, such as second hand smokng, d E < 0. The MCF from taxng commodty (assumng a perfectly compettve market and no other tax or nontax dstortons) s: x d E x δ MCF ε τ = = (3) R + τe ε t where δ E =d El / q s the proportonal margnal external beneft generated by x. If the actvty generates a postve externalty, then δ E ε s postve and the MCF s hgher because taxng the commodty reduces the postve external beneft from the commodty. If the actvty produces a harmful externalty, then δ E ε s negatve, and the MCF s lower, reflectng a socal gan from reducng a harmful externalty when the commodty s taxed. Fnally, note that from the above equaton, the optmal tax rate on commodty s the Pgouvan tax τ = δ E f the government can levy lump-sum taxes and ts MCF s one. Publc expendture externaltes In the prevous secton, we showed how a dstorton caused by an envronmental externalty, such as second hand smokng, can be ncorporated n the formula for the MCF. There s another type of externalty whch we wll call a publc expendture externalty that operates through the government s budget constrant. For example, an ncrease n cgarette consumpton may drve up publc expendtures on health care. Even n the absence of a second-hand smoke externalty, smokng adversely affects non-smokers through the hgher taxes that they have to pay as a result of hgher publc health care expendtures. The health care costs assocated wth smokng are often used to justfy hgh taxes on tobacco products. Below, we show how these publc expendture effects can be ncorporated n the formula for the MCF. 38

6 Suppose that the government provdes a servce, G, and the cost of provdng ths servce s C(G, x) where C/ G > 0 s the margnal cost of provdng the servce and C/ x s the ncrease n the cost of provdng a gven level of servce (say health care) as a result of an ncrease n the consumpton of a prvate good x. For smplcty, we wll gnore the taxes that are leved on other goods (that mght be substtutes or complements wth x), and therefore the publc sector s budget constrant requres that tx = C(G, x). Increasng the tax rate on x can ncrease the publc sector s net revenues ether drectly by ncreasng total revenues or ndrectly by reducng net expendtures. Consequently, the MCF for taxng x wll be equal to: MCF t = x dx x + t C x dx = + ( τ δ )ε G (4) where t s assumed that the supply of the taxed commodty s perfectly elastc so that dq/ = and δg = C/ q G whch s the change n the cost of publc expendtures when ndvduals spend another dollar on x. When δ G > 0 (e.g. tobacco products), we see that the publc expendture effect reduces the MCF when a hgher tax rate reduces the demand for the commodtes that are responsble for hgher costs of provdng a gven level of publc servces. If government could mpose lump-sum taxes, and the MCF was one, then the optmal tax rate on x would be τ = δg. In other words, the commodty would be taxed at a rate that reflects ts publc expendture externalty, just as n the case of the Pgouvan tax for a drect consumpton externalty. Addcton Many ndvduals regret excessve consumpton of some commodtes, such as alcohol, tobacco and fatty foods. For example, durng 2000, 70 percent of current smokers expressed a desre to qut completely and 4 percent stopped smokng for at least one day n an attempt to qut, but only 4.7 percent successfully abstaned for more than three months. 4 Indvduals who are prone to excessve drnkng or smokng are sad to have self-control or addcton problems. In many countres, excse taxes on alcohol and tobacco are vewed favourably as sn taxes because hgher prces may reduce the degree of excessve consumpton. See Badenes-Plá and Jones (2003) for a survey of the economcs lterature on addcton and taxes, Gruber and Kőszeg (2004) and Gruber and Mullanathan (2005) for recent emprcal studes of the mplcatons of addcton for effcency and dstrbutonal effects of cgarette taxes, and the Economst (2006) for a dscusson of publc polces based on the new paternalsm. We use a smple model developed by O Donoghue and Rabn (2006) to llustrate the way n whch the self-control dstorton can be ncorporated n the evaluaton of a tax ncrease on these commodtes. Suppose an ndvdual consumes only two goods, x and x 2. The consumpton of x 2 provdes constant margnal utlty, normalzed to equal one. The consumpton of x provdes the ndvdual wth a beneft V(x ) and also a psychc cost C(x ), whch could be nterpreted as a cost that arses from a future health 4 Bernhem and Rangel (2005, p.39). 39

7 problem. The ndvdual make consumpton decsons accordng to the followng utlty functon: * U V(x) ΦC(x) + = x (5) 2 where Φ s a postve parameter. If Φ <, the ndvdual s sad to have a self-control problem because he does not take nto account the full personal cost consumng x. The ndvdual s budget constrant s q x + x 2 = I, where the prce of x 2 s set equal to one. The ndvdual consumes * x based on the frst order condton, Vx ΦC q x + However, the ndvdual s long-run happness s based on the utlty functon: ** U V(x) C(x) + 2 =. = x (6) whch fully reflects the cost that the ndvdual ncurs when he consumes x. The ndvdual wth self control problems over-consumes x because the deal consumpton s based on V = C x q x + To evaluate the effects of a tax rate change, we wll assume that the ndvdual and socety are concerned wth the mpact of the tax ncrease on the ndvdual s long-run utlty. (See Bernhem and Rangel (2005) on usng the ndvdual s long-term welfare n assessng polces.) The welfare effect of a tax ncrease s equal to: du ** dx dx ( V C q) 2 * = (Vx Cx ) + = x + x (7) x dq dq dq dx * snce, from the ndvdual s budget constrant, dx 2 /dq = - x - q dx /dq. Usng the ndvdual s frst order condton, the followng expresson measures the harm caused by a tax ncrease: du ( Φ)C ** * x * = x + = x ( δa ε) q ε (8) where ε = (dx /dq )(q /x ) < 0 s the prce elastcty of demand for x. The dstorton caused by the ndvdual s self-control problem s defned as: ( x Cx q) / q = ( Φ ) Cx / q δ (9) V A = The δa parameter reflects the dstorton that arses because there s a wedge between margnal value of an addtonal unt of x to the ndvdual and ts true margnal cost. It can be nterpreted as the neglected proporton of the addtonal cost ncurred n spendng an addtonal dollar on x. If the ndvdual has a self-control problem and Φ <, δ s negatve, and ths factor tends to reduce the socal harm from a tax A ncrease. Indeed, t s possble for a prce ncrease to make the ndvdual better off, at least as judged by hs long-run utlty functon, f δ A ε > and n ths case, the MCF would be a negatve number. The formula for the margnal cost of publc funds for the commodty tax s: 40

8 δ ε MCF = A t + τ ε (0) assumng that there are no other dstortons n the economy. If the government could rase revenue by mposng a lump-sum tax, such that ts MCF was.00, then the optmal tax rate on the commodty would be τ = δa = ( Φ) (Cx / q ). The optmal sn tax rate would reflect the neglected proporton of the addtonal cost ncurred n spendng an addtonal dollar on x. See O Donoghue, T. and M. Rabn (2006) for further dscusson of optmal sn taxes. Obvously, ncorporatng these self-control dstortons nto the calculaton of the MCF s controversal, but we thnk that lack of self-control problems, especally wth regard to tobacco products, reflects publc opnon and polcy-makers vews concernng the use of excse taxes. For ths reason, we thnk that t s mportant to ncorporate defectve decson-makng explctly n the model so that t can be compared wth the other dstortons that affect the MCF. In ths way, a better judgment can be made concernng the relatve mportance of self-control problems n the overall assessment of the approprate level of excse taxaton. Market power Suppose an excse tax s leved on a monopolst s product. To keep the model as smple as possble, we wll gnore all other tax and non-tax dstortons n dervng an expresson for the dstorton due to monopoly power. Let the after-tax proft of the monopolst be: ( τ ) [ qx C(x) tx] Π = π () where τ π s the proft tax rate, and t s the per unt tax rate. 5 proft wth respect to the per unt tax rate, we obtan: Dfferentatng after-tax dπ = = ( τ ) π x dq dx dq ( τ ) x x + ( q MC t) π dq + dx dq q dq dx MC dq dq dq x dx t dq dq (2) where MC s the margnal cost of producton. Ignorng dstrbutonal effects, the ndrect utlty functon for a representatve ndvdual s V(q, Π ). The socal welfare cost of an ncrease n the tax rate on the monopolst s product s: 5 In general, t s mportant to dstngush between ad valorem taxes and per unt taxes n evaluatng the MCF from taxng a monopolst s product because as s well known a monopolst s prce response dffers for these two types of taxes. See Dahlby (forthcomng, Chapter 3). For our purposes, because the Tha Tobacco Company s a state-owned monopoly, ths dstncton can be gnored as wll be shown below. 4

9 λ dv = λ = dq x = x dv dq ( τ ) dv dπ + dπ dq dq π x + δ M ε (3) dq dq π π M ( τ ) + τ ( τ ) δ ε π dq where δ M = (q MC t) / q s a measure of the dstorton n the market caused by mperfectly compettve behavour, ε s the elastcty of demand for the monopolst s product, λ = dv / dπ s the margnal utlty of ncome, and dv / dq = λ x by Roy s Theorem. The frst term n square brackets represents the net ncrease n taxes pad by the prvate sector, gven that excse taxes are deductble n computng the frm s proft tax lablty. The second term represents the addtonal proft tax that s pad as a result of the ncrease n the prce of the monopolst s good. The thrd term s the reducton n after-tax profts sustaned by the monopolst as a result of the declne n output caused by the tax. Total tax revenue s equal to R = tx + τππ/( τπ) Dfferentatng wth respect to t we obtan: dr dq τ π = x + τε + τπ dq = x π π ( τ ) + τ + ( τ + τ δ ) ε ( τ ) x x + ( q MC t) π π M dq dq dx dq dq (4) The frst term n square brackets represents the net ncrease n tax revenues, for a gven level of output by the monopolst, the second term s the ncrease n proft tax revenues from the nduced ncrease n the prce of the monopolst s product, and the thrd term s reducton n total tax revenues from the reducton n the output produced by the monopolst. Note that the government sustans a reducton n proft taxes, as a result of the reducton n the monopolst s proft. τ π δ M From the above equatons, we can obtan the followng expresson for the margnal cost of publc funds from an excse tax leved on a monopolst s product: MCF t = dq dq ( τ ) + τ ( τ ) π π dq ε dq ( τπ ) + τπ + ( τ + τπδm ) ε π δ M (5) An nterestng specal case s where τ π =, whch corresponds to a stuaton where the monopoly s owned by the government and all of the profts and taxes on the product are receved by the publc treasury. Ths case s partcularly relevant for Thaland because of the Tha Tobacco Company, a state-owned enterprse, has a 42

10 monopoly on the sale and dstrbuton of domestcally produced cgarettes n Thaland. In ths stuaton, the MCF s equal to: τ MCF t ( = ) = π + ( τ + δ ) (6) ε M whch s ndependent of the degree of tax shftng. In ths case, the total tax rate on the product s effectvely τ + δ. M Smugglng Norton (988) has developed an economc model of smugglng and Usher (986) and Ray (997, ) have ncorporated tax evason nto the calculaton of the MCF. Below, we outlne a smple model that ncorporates smugglng nto the MCF for an excse tax. Suppose the elastcty of the supply of the smuggled commodty s η s > 0. The prce of the smuggled commodty wll reflect ts producton cost plus the smugglng costs that are ncurred by the smugglers, q s = p + c s. It wll be assumed that these smugglng costs are less than the per unt excse tax mposed on the legtmate goods. Consumers are wllng to buy smuggled goods as long as the prce of a smuggled good plus the search costs, f, are less than the prce of a legtmate good cgarette, q s = q f. Assumng the excse tax ncreases are fully reflected n the prce of the legtmate good, ths mples that dq/ = dq s / = f search costs are relatvely constant. The demand for the legtmate goods that are fully taxed s the dfference between the total demand and the demand for smuggled goods or x l = x T (q) x s (q s ) where x T s the total number of cgarettes consumed. The government s tax revenue (gnorng all other taxes) s R = tx l. The margnal cost of publc funds from taxng cgarettes can then be expressed as: MCF s l dq s dq x + x = = (7) l l l dx dq ( v)( + τε ) x + t dq where v = x s /x T l s the share of the smuggled goods n total consumpton and ε s the elastcty of demand for legtmate goods. Smugglng ncreases the MCF because the tax base s smaller and the tax base s more tax senstve because smugglng gves ndvduals the opportunty to swtch to a non-taxed alternatve. The elastcty of demand for legtmate goods s related to the elastcty of demand for total consumpton and the smugglng supply elastcty as follows: l T s s ( v) ε = ε v(q / q ) η where (q/q s ) = (p+c s +f)/(p+c s ) < ( ) τ (8) When the tax rate s rased, the volume of taxed goods decreases because total consumpton falls and the volume of smuggled goods ncreases. For example, f 20 percent of the cgarettes are smuggled and f the elastcty of total demand for cgarettes s -0.40, the elastcty of demand for legtmate cgarettes could be as hgh as f the elastcty of the supply of smuggled cgarettes s 0.50 and as much as -.44 f the elastcty of the supply of smuggled cgarettes s.50. Therefore, gnorng the mpact of smugglng by usng the elastcty of total demand for cgarettes n the 43

11 calculaton of the MCF, rather then the elastcty of demand for legtmate cgarettes, may sgnfcantly under-estmate the MCF for cgarette taxes. Dstrbutonal consderatons To ths pont, we have focused on the effcency aspects of the margnal cost of publc funds. However, all socetes are concerned about the dstrbutonal mpact of ther tax system, and a tax ncrease that s borne manly by the poor can be vewed as havng a hgh socal cost. Indeed, governments use dstortonary taxes because of ther concern for dstrbutonal equty,.e. n the absence of these concerns, governments could smply rely on lump-sum taxes. Consequently, we need to ncorporate dstrbutonal concerns n the measurement of the socal margnal cost of publc funds to fully evaluate tax and expendture reforms. To ncorporate dstrbutonal consderatons, we follow the procedure developed by Feldsten (972) and mplemented by Ahmad and Stern (984) n the analyss of commodty tax reform n Inda. 6 Suppose there are H households n the economy. Household h purchases x unts of commodty at the prce q. The household s h budget constrant s q x = n = h I h where I h s the household s lump-sum ncome. The level of utlty or well-beng that household h can obtan, gven consumer and producer prces, ts lump-sum ncome, ts ownershp of nputs, and ts preferences, s ndcated by ts ndrect utlty functon, V h = V h (q, I h, G) where q s the vector of consumer prces, p s the vector of producer prces, and G s a vector of publclyprovded goods and servces. By Roy s theorem, V h / q = -λ h x h < 0 where λ h (q, I h, h G) s the household s margnal utlty of ncome and x (q, I h, G) s the household s ordnary demand functon for commodty. The total demand for commodty s H h x = x. h= Suppose that tax and expendture decsons are based on the socal welfare functon, S = S(V, V 2,, V H ), whch reflects the trade off that a socety s wllng to make when a polcy makes some households better off and other household worse off. The dstrbutonal weght, β h = ( S/ V h )λ h, represents the value that the socety places on an extra dollar of lump-sum ncome receved by household h. It wll be assumed that the socal welfare functon reflects a pro-poor preference such that β h s hgher when V h s lower. The socal valuaton of the households welfare loss from an ncrease n the prce of commodty s: S q = H h = S h V V q h = H h = h β x h = H h = β h s x = ωx h (9) 6 For a recent applcaton of ths methodology to commodty tax reform n Italy, see Lberat (200). 44

12 h h where s = x / x s household h s share of the total consumpton of commodty. The ω parameter s known as the dstrbutonal characterstc commodty, and t measures the socal harm caused by ncreasng total household expendture on x by a dollar. Note that ω wll tend to be larger when β h and s are postvely correlated. Ths means that ω wll be hgh for commodtes that are consumed manly by the poor. The socal margnal cost of publc funds from taxng commodty can be defned as: SMCF ds = = ω MCF (20) dr t t To compute the ω s, we need the β h s whch reflect a socety s, or perhaps more accurately ts polcy-makers, wllngness to trade-off gans and losses sustaned by dfferent segments of socety. The dstrbutonal weghts are based on value judgments, and economsts have no specal nsghts nto what consttutes the approprate set of dstrbutonal weghts. Economsts, however, have tred to help polcy-makers apply a consstent set of dstrbutonal weghts. One approach s to use an explct functonal form for the relatve dstrbutonal weghts such as: h h β r β Y Y h = r ξ (2) where Y r s the ncome of a reference household (such as a household wth the average ncome) and ξ 0 s a parameter that measures the socety s averson to nequalty. A standard normalzaton s to set β r =. If ξ = 0, β h = for all h, and no consderaton s gven to dstrbutonal concerns. On the other hand, f Y h = 0.5 Y r, then β h =.44 f ξ = 0.5 and β h = 2 f ξ =. We use ths approach to compute the SMCFs of the alcohol, tobacco, and fuel excse taxes. The SMCF In the absence of a general equlbrum model to trace the effects of the excse taxes on the prces of all commodtes, we have assumed that the excse taxes on alcoholc beverages, tobacco, and fuel are fully reflected n ther product prces and that the prces of other commodtes are not affected. Therefore, dq / = and dq j / =0 for j. Combnng the tax and non-tax dstortons dscussed n the prevous secton wth the dstrbutonal characterstc of the taxed good, we have the followng formula for the socal margnal cost of funds for an excse tax: SMCF t = ω b b ( δ + α δ + ( τ ) δ ) l T ( v ) + b ( v )( τ + τ δ ) ε b δ ε n b j j = n j j = E j j j j A j π j M j π j j M j n ε T j j j = G j j (22) 45

13 Note that the components of the MCF that reflect the dstortons are multpled by the ε s, whch reflect the change n total demands for goods, whle the tax revenue T j changes (the second set of terms n the denomnator) are multpled by the elastctes of demand for legtmate commodtes, the ε l j s. PARAMETER VALUES In ths secton, we descrbe how the varous parameters used n the calculatons were chosen. Tax rates and budget shares The tax rates and budget shares for the 0 commodty groups that were ncluded n the analyss are shown n Table. The data used for calculatng average tax rates are from the Mnstry of Fnance and Natonal Economc and Socal Development Board (NESDB). The statutory value-added tax was 7.0 percent n 2002, but the average tax rates are around 3.5 percent for most commodtes except for food and clothng because some tems and small frms are exempt from VAT. The average tax rates for alcoholc beverages and tobacco are 39.3 and 58.7 percent. Note that the tax rate for tobacco does not nclude the proft earned by the Tha Tobacco Monopoly (TTM), the state-owned company that has a monopoly n the producton of domestc cgarettes. The average tax rate for fuel was 53.6 percent. An appendx descrbng the computatons of the tax rates s avalable from the authors upon request. The budget shares were calculated from the aggregate consumpton data from the NESDB. The budget share of alcohol was 4.2 percent, tobacco was.7 percent, and electrcty and fuels was 2.4 percent of aggregate consumpton spendng n TABLE : TAX RATES AND BUDGET SHARES FOR COMMODITIES IN THAILAND Commodty Tax Rate, τ Budget Share, b Food Alcohol Tobacco Clothng Health Electrcty and Fuels Telecommuncatons Housng and Water Entertanment Other Goods and Servces

14 Demand elastctes The estmated demand elastctes are shown n the matrx below. (The own-prce elastctes are along the dagonal.) The prce elastctes of demand for the ten commodtes were estmated, usng the Almost Ideal Demand System (AIDS) developed by Deaton and Muellbauer (980), based on data on consumpton expendtures from 983 to 2002 n the Thaland Natonal Income Account. The observatons for were omtted because of the non-normal consumpton shares n that year due to the economc crss that began n the fall of 997. (An appendx descrbng the demand estmaton s avalable from the authors upon request.) Our estmated own-prce elastcty for alcoholc beverages s qute hgh, , compared to the estmate obtaned by Sarntsart (2003). However, t s less elastc than the values n the TDRI (2005) study where the prce elastctes for color lquor, whte lquor, mported lquor, beer and wne were -.56, -2.73, -0.6, and Part of the reason for the dfferences n these estmates may be the fact that Sarntsart used household consumpton data that ncluded both tax and untaxed consumpton whle TDRI used the data from taxed consumpton provded by the Excse Department. See Leung and Phelps (993) and Badenes-Plá and Jones (2003, Table 3, p.40) for a summary of emprcal estmates of the prce elastcty of alcohol consumpton n the US and other countres. These studes generally ndcate that the demand for beer s relatvely prce nsenstve (around -0.3) and the demand for sprts s prce elastc (around -.5) wth the demand for wne havng an ntermedate prce elastcty (around -.00). Our estmate of the elastcty of the total demand for taxed alcohol falls wthn the usual range of estmates from other countres. Our elastcty estmates ndcate that alcohol s a gross complement for tobacco ( ) and for electrcty and fuel ( ) whle tobacco s a very weak complement for alcohol (-0.025), but a substtute for electrcty and fuel (0.28). 7 Therefore an alcohol tax rate ncrease wll reduce the demand for both tobacco and fuel, and therefore some of the ncrease n the alcohol tax revenues from an alcohol 7 A recent study by Decker and Schwartz (2000) usng household level data from the U.S. ndcated that the demand for cgarettes declnes when alcohol prces ncrease whle the demand for alcohol ncreases when the prce of cgarettes ncrease. An early study usng aggregate US data by Goel and Morey (995) had ndcated that alcohol and tobacco were substtutes, whle a study usng UK data by Jones (989) ndcated that tobacco was a complement for alcohol. Gruber, Sen and Stable (2002) based on Canadan data found that hgher cgarette prces reduced alcohol consumpton. 47

15 tax rate ncrease wll be offset by declnes n tobacco and fuel excse tax revenues. (The net effect on other commodty tax revenues s ndetermnate, but lkely to be relatvely small.) Ths negatve effect on tobacco and fuel excse tax revenues wll tend to rase the MCF for alcohol excse taxes. However, the reductons n the consumpton of tobacco and fuel would also reduce the MCF for alcohol excse taxes f the net dstorton for these commodtes, captured by the δ ( ) E + δ j A + τ δ j π j M j terms n the MCF formula, are negatve.e. margnal socal cost exceeds margnal socal beneft. The prce elastcty for tobacco products s , whch s close to the value obtaned n a study by Pattamasrwat (989), but substantally hgher than the prce elastcty found by Sarntsart (2003) based on household tobacco consumpton data. 8 The dfferences may be due to smuggled or non-taxed cgarettes whch the study by Sarntsart ndcated are farly prevalent n Thaland. (He found that about 46 percent of mported cgarette package ltterng n fve provnces across Thaland were untaxed cgarette.) In other words, the prce elastcty usng data from the Natonal Income Account s hgher than for total household cgarette consumpton, where taxed and untaxed cgarettes are ncluded. Galbrath and Kaserman (997) found the same relatonshp n Canada where the prce elastcty for taxed cgarettes was hgher (-.0) than that for total (taxed and untaxed) cgarette consumpton (-0.4). Another study from Canada by Gruber, Sen and Stable (2002) also found that the demand for taxed cgarettes was hgher than the total demand (-0.70 versus -0.45). Our crossprce elastctes of demand mply that an ncrease n tobacco taxes wll ncrease excse tax revenues from fuel, but ncrease dstorton n the allocaton of resources f there s a negatve dstorton n the market for fuel. The demand for fuel and electrcty consumpton s qute prce nelastc (-0.833). Econometrc studes of prce elastcty of gasolne n the U.S. revewed by Parry and Smart fall n the -0.3 to range 9, and therefore our estmate of the own-prce elastcty s consderably lower than that found n other countres. However, Wade (2003) showed that the short-run prce elastctes of dstllate fuel for resdental and commercal uses were -0.5 and In hs revew, he showed that short-run prce elastcty of fuel ol for resdental use n the U.S. was -0.0 to and for commercal use was to Our econometrc estmates ndcate that electrcty and fuel s a substtute for alcohol (0.5244) and a weak complement for cgarettes ( ). Consequently, a fuel tax ncrease would tend to ncrease alcohol excse tax revenues and mprove the allocaton of resources f the net non-tax dstorton n the alcohol market s postve. Our demand estmaton s based on the assumpton that total consumer expendture s exogenously determned. In partcular, t assumes that varatons n the prces of commodtes do not affect labour supply decsons. Most of the prevous studes of commodty tax reform such as Ahmad and Stern (984) and Decoster and Schokkaert (990) have ether adopted ths assumpton or assumed separablty between lesure and all other goods n consumers utlty functons. These assumptons mply that n the absence of non-tax dstortons the optmal commodty 8 See N. Badenes-Plá and Jones (2003, Table 2, p.39) for a summary of emprcal estmates of the elastcty of cgarette consumpton whch are n the to range. 9 See Parry and Smart (2005, p. 283). 48

16 tax rate s a unform tax rate because all good are equally substtutable wth lesure, the non-taxed good. Gven the mportance that the theoretcal lterature on optmal taxaton has attached to the cross-prce elastctes between lesure and commodtes, t s mportant to brefly revew the few papers have examned the emprcal sgnfcance of the separablty assumpton for computng MCFs for commodty taxes. Madden (995, p. 497), notng that several econometrc studes of consumer demands and labour supples reject the separablty assumpton, estmated models wth and wthout the separablty assumpton, based on data for Ireland , and concluded that the MCF rankngs do not appear to be very senstve to assumptons regardng separablty between goods and lesure. In partcular, he found that the MCFs for alcohol, tobacco, and fuels were.664,.397, and.93, respectvely, wthout mposng separablty and 2.304,.504, and.48 when separablty was mposed. 0 Although Madden s estmates of the MCFs were hgher when separablty between lesure and commodtes was mposed n estmatng the demand elastctes, the rankngs of the MCFs for the three commodtes subject to hgh levels of excse taxaton dd not change. In hs computatons of the effcency effects of excse taxes n the U.K., Parry (2003) assumed that petrol and alcoholc beverages were substtutes for lesure and that cgarettes were a complement. However, the mpled cross-prce elastctes between lesure and the prce of these commodtes were very low and dd not have a materal effect on Parry s measures of the margnal excess burdens mposed by the excse taxes. In marked contrast wth the above studes, West and Wllams (2006) found that ncludng the cross-prce effect between labour supply and the prce of gasolne had a sgnfcant effect on the magntude of the MCF for the excse tax on gasolne n the Unted States. They estmated a model based on ndvdual household s expendtures gasolne and all other goods and ther labour ncome, and found that hgher gasolne prces ncreased labour ncome (reduced the demand for lesure). Ths reduced the MCF from taxng gasolne and ncreased the optmal gasolne tax rate. However, only one of the three cross-prce elastcty between labour ncome and the prce of gasolne that they estmated was sgnfcantly dfferent from zero (males n households wth two adults) and that pont elastcty was very low The West and Wllams results are somewhat surprsng, and the mportance of the cross-prce effects between excse taxes and labour supples need to be nvestgated more completely. Gven our current and very lmted knowledge about the mportance of these effects, we have proceeded by adoptng the conventonal assumpton that these effects do not have a materal effect on the rankngs of the MCFs for excse taxes. Envronmental externaltes In spte of a sgnfcant body of research, there s a great deal of uncertanty regardng the approprate values to use for the δe parameters for developed countres, such as the Unted States or the Unted Kngdom. There s even greater uncertanty for a developng country, such as Thaland, where much less emprcal research has been 0 Madden calculated the margnal revenue cost of ncreasng welfare, whch s the nverse of the MCF. See Dahlby (forthcomng, Chapter 3). 49

17 done on the envronmental mpacts of alcohol, tobacco, and fuels and where economc, socal, and envronmental condtons may be substantally dfferent than n the developed countres. Nonetheless, we have had to make some choces regardng these parameters, whch are shown n Table 2. A detaled descrpton of the benchmark parameter values s gven n the followng sectons of ths paper. 2 2 Parry (2003) has provded an extensve revew of the emprcal lterature on the externaltes generated by the consumpton of gasolne, alcohol, and cgarettes n the Unted States and the Unted Kngdom. Whle there s stll a great deal of uncertanty concernng the magntudes of these parameters, Parry s choces for hs base case estmates seem reasonable, but ther applcablty to Thaland s unknown. Based on hs revew of the lterature, Parry concluded that tobacco products mpose the largest harmful externaltes, representng 28.3 percent of the consumer prce of the product, followed by petrol at 7.8 percent, and alcohol at percent of the product prce. It should be noted that Parry treated all externaltes as drect consumpton externaltes even though hs dscusson and the lterature ndcate that these externaltes, especally for smokng and alcohol consumpton, take the form of hgher publc expendtures on health care, and n our framework would be ncluded n the δ G parameters. 50

18 TABLE 2: PARAMETER VALUES FOR NON-TAX DISTORTIONS Low Case Benchmark Case Hgh Case Envronmental Externalty, Alcohol Cgarettes Fuel Our estmates for the envronmental externaltes from alcohol are based on Smth (2005) s recent survey of alcohol excse taxes because he decomposed these externaltes n a way that s consstent wth our framework. 3 Smth estmated that the total externalty cost of alcohol n the U.K. s 7 percent of the pre-tax prce. Based on hs breakdown of the socal costs of alcohol, we have decomposed hs total externalty nto an 8.2 percent prvate sector envronmental externalty (losses sustaned by employers etc.), a.3 percent publc expendture externalty (health δ E Publc Expendture Externalty, Alcohol Cgarettes Fuel Addcton, δ Alcohol -0.03, , , 0.07 Cgarettes -0.8, , , 0.8 Fuel 0, 0 0, 0 0, 0 A Market Power, Alcohol , 0.26 Cgarettes Fuel Telecom Net Non-Tax Dstorton: δ E - δ G + αδ A + ( - τ π )δ M Alcohol Cgarettes Fuel Telecom Smugglng, Alcohol -0.54, , , Cgarettes -0.40, , , Fuel Na na na T ε and δ M and α v δ G 3 For further dscusson of the externaltes caused by alcohol consumpton and tax polces to deal wth these ssues, see Pogue and Sgontz (989), Grossman et al. (993), Irvne and Sms (993), Kenkel (996), Cook and Moore (2002), and Chaloupka, Grossman, and Saffer (2002) and Grossman (2004). 5

19 costs, crme, and socal responses) and 7.3 percent nternalty from unemployment and pre-mature death. (The latter s ncluded n the δ A parameter for alcohol to be dscussed n Secton 3.6.) The δ E parameter for the benchmark case was calculated as *( 0.393)*0.27 = The s the tax rate on alcohol n Thaland. We multply by ( 0.393) to express the externalty as a percentage of the tax nclusve prce. We then multply by the 0.27 whch s the rato of the purchasng power party Tha GDP per capta to the U.K GDP per capta. 4 The Hgh Case s the benchmark case wthout the adjustment for the relatve GDPs n Thaland and the U.K. The Low Case s 50 percent of the benchmark case. The envronmental externalty from tobacco s manly second-hand smoke, and we do not know of any estmates for ths type of externalty. As noted n the lterature, much of the second-hand smoke problem occurs wthn the famly, and therefore t s debatable whether ths s an externalty. The ncdence of second-hand smoke n Thaland has also been reduced wth non-smokng n publc transt, schools and publc offces, but smokng s stll permtted n bars and non ar-condtoned restaurants n Thaland. Overall, we thnk that the second-hand smoke externalty s lkely to be small (not many people offer to pay smokers to butt out ther cgarettes), but obvously ths s controversal and based on a value judgment that we admt s dffcult to defend. Newbery s (2005) estmate of the envronmental cost s 4 pence per ltre for gasolne n UK, excludng road costs whch we treat as a publc expendture externalty, and ncludng 3.2 pence per ltre for accdents. Our benchmark value for fuel envronmental externalty s -(0.4 /ltre)(67.8b/ )(0.27)(25B/ltre) = -0.0 usng the relatve Tha to UK GDP per capta to s 27 percent of the U.K GDP per capta. For the Hgh Case, we do not adjust for dfferences n Tha to UK real GDP per capta - (0.4 /ltre)(67.8b/ )/(25B/ltre) = The Low Case s 50 percent of the Benchmark case. Publc expendture externaltes As s wdely recognzed, alcohol, tobacco and fuel consumpton may drectly or ndrectly drve up publc expendtures, forcng taxpayers to pay hgher taxes to fnance them or crowdng out other valuable publc servces. Ths dstorton operates through the government s budget constrant, and therefore t has a dstnct effect on the margnal cost of publc funds, even though most studes do not dstngush between envronmental externaltes and publc expendture externaltes. The publc expendture externalty for alcohol s based on an estmate of.3 percent of the pre-tax prce n the U.K by Smth (2005). The dstorton parameter was calculated as 0.03*( 0.393)*0.27 = where, as before, we multply by ( 0.393) to express the externalty as a percentage of the tax nclusve prce. We then multply by the 0.27, whch s the rato of the purchasng power party Tha GDP per capta to the U.K GDP per capta. The Hgh Case s the Benchmark Case wthout the adjustment for the relatve GDPs n Thaland and the U.K. The Low Case s 50 percent of the benchmark case. 4 Ths beneft transfer technque (the value transfer method) s used to convert the study ste values (U.K n ths case) to polcy ste values (Thaland n ths case). The converson usng real per capta GDP s wdely appled n envronmental assessment n Thaland. See Rosenberger and Looms (2003) for more detals of the technque. 52

20 The benchmark value for the mpact of smokng on health care costs uses the estmates from Mannng et al. (989) of $US 0.25 per package (fgures updated to 2003) See Cnossen (2005, p.37). Ths value was multpled by 0.20 to reflect the relatve GDP n Thaland and dvded by.08, the prce of a package of cgarettes n Thaland. The resultng estmate of the δ G parameter s (0.25)(0.20)/(.08)= 0.046, rounded to The Hgh Case was obtaned usng the poston expressed by the Drector-General for WHO, Dr. Lee Jong-wook, that 5 percent of all health care costs n hgh ncome countres are due to smokng. Publc health care costs are two-thrds of total health care costs n Thaland. Total health care costs n 2002 were 333,798 mllon Baht and total value of cgarette consumpton was 55,832 mllon Baht. Therefore the Hgh Case parameter value was calculated as (0.32)(0.5)(333,3798)/(45,29) = 0.29, rounded to The Low Case parameter value was based on the Sarntsart (2003, p. 43) estmate that the drect health care costs of tobacco were 249 mllon Baht n Ths would mply that the δ parameter would be (249)/(55,832)= G Newbery s (2005) estmate of road costs are 25.2 pence/ltre n the U.K. The benchmark value for fuel publc expendture externalty s (0.252 /ltre)(67.8baht/ ) (0.27)(25Baht/ltre) = 0.8. The Hgh Case s 50 percent hgher and the Low Case s 50 percent lower that the Benchmark Case. Addcton As noted n the ntroducton, excse taxes are often vewed as sn taxes, leved n order to dscourage the consumpton of products that are bad for people. In Secton 2.3, we used the O Donoghue and Rabn (2006) model to formalze the vew that some ndvduals engage n excessve consumpton of alcohol and tobacco because of defectve decson-makng. Obvously, the choce of the parameters s dffcult n the absence of emprcal research that mght shed lght on the degree of excessve consumpton. Some progress n ths drecton has made wth the study by Gruber and Mullanathan (2005) whch suggested that cgarette taxes n the U.S. and Canada mght make some ndvduals better off by nducng them to qut smokng, or at least reduce ther consumpton of cgarettes. More research on ths topc s obvously needed before anyone can feel fully comfortable n ncorporatng addcton n the MCF calculatons. However, strong vews about addcton domnate publc vews about the mportance of excse taxes on alcohol and tobacco. We hope that our formalzaton of these vews wll help to assess ther mportance relatve to the other factors, such as externaltes, market power, and smugglng, whch also nfluence publc polcy regardng excse taxes. The calculaton of the addcton parameter was based on Smth s estmate that the ncome loss from unemployment and premature death n the U.K. was 7.3 percent of the pre-tax prce of alcohol. The value of value C x /q x was calculated as (0.073*(-0.373)*0.27)/(0.05) = (The dvson by 0.05 represents the calculaton of the present value of the annual stream of lost ncome at a fve percent dscount rate.) Gruber and Kőszeg (2004, Table 2, page 977) used values of Φ = 0.60 to Φ = 0.9 to reflect hyperbolc dscountng of future costs and benefts by ndvduals wth addcton problems. We use the md-range value of Ths mples that our benchmark parameter value for δ A for alcohol s (0.75-)0.24 = The Low Case s 50% of the benchmark case and the Hgh Case s twce the benchmark case. The proporton of the populaton addcted to alcohol, α, s the 3.34 percent of the 53

21 populaton who reportedly drnk every day plus 50 percent of the 3.79 percent who drnk 3 to 4 tmes per week. 5 Thus the Benchmark fgure for α s 3.34+(0.5)3.79 = 5.2 percent. The Hgh Case fgure s = 7. percent. The Low Case fgure s half the percentage that drnks every day. The Benchmark value for the addcton dstorton for cgarettes was obtaned usng Gruber and Kőszeg s (2004, p.979) estmate that the cost n terms of lfe years lost per pack of cgarettes n the Unted States s $ The purchasng power equvalent per capta GDP n Thaland s 20 percent of the U.S. and prce of cgarettes n Thaland n U.S. s.08. See Gundon, Tobn and Yach (2002). We also used a value of 0.75 for Φ as n the alcohol addcton calculatons. Taken together, our benchmark value for δ A for cgarettes s (0.75-)(35.64/.08)(0.20) = -.65, mplyng that the neglected cost per package of cgarettes n Thaland s 65 percent of the actual prce. The Low Case s 50% of the Benchmark Case and the Hgh Case s twce the Benchmark Case. The estmate for the proporton of addcted smokers s the 8 percent of Thas who are reported to be regular smokers. 6 Market power Imperfect competton s a market dstorton, but t has played lttle role n the dscusson of excse tax polcy, even though n beer and tobacco markets are hghly concentrated n many countres. 7 For example, Cnossen and Smart (2005) do not dscuss the mplcatons of frms market power for settng cgarette taxes. In our calculaton, we ncorporate a measure of the dstorton caused by market power n the beer and whte whskey market, the tobacco market, and the moble phone market n Thaland. The latter s ncluded, even though an excse tax was not leved on telecommuncaton servces n 2002 because excse tax ncreases on alcohol, tobacco, and fuels mght ncrease (decrease) the demand for telecommuncaton servces leadng to an mprovement (deteroraton) n resource allocaton. The domestc beer market n Thaland s domnated by two large frms Boon Rawd Brewery Co. and Tha Beverage PLC. In 2002, Tha Beverage PLC had 65 percent of the beer market and Boon Rawd Brewery Co. had 26 percent. Tha Beverage PLC also has a monopoly power over the whte lquor market. 8 The market power parameter for alcohol was based on the assumpton that the sale of beer and whte lquor, whch represent approxmately 70 percent of total alcohol sales, s a Cournot duopoly. Therefore, δ M = 0.5(/-2.7)0.70 = 0.3 where the 0.5 s one dvded by the number of frms and -2.7 s an estmate of the elastcty of demand for beer and whte lquor from the study by TDRI (2005). 9 Ths calculaton mples that the frms earn a pure proft margn of 3 percent. The Hgh Case s twce the Benchmark Case and the Low Case s 50 percent of the Benchmark Case. It s 5 These fgures are from the Health and Welfare Survey (HWS) n The survey s conducted by Natonal Statstcal Offce of Thaland. It ntervews health status, healthcare nsurance coverage, healthcare expendture, and health related behavour. 6 From Health and Welfare Survey n See Delpalla and O Donnell (200) on the concentraton of EU cgarette markets. 8 (For a summary of market competton, see n an appendx avalable from the authors upon request.) 9 In the TDRI study the prce elastctes for color lquor, whte lquor, mported lquor, beer and wne were -.56, -2.73, -0.6, and respectvely. 54