ESTIMATING TAX INCIDENCE, MARKET POWER AND MARKET CONDUCT: THE EUROPEAN CIGARETTE INDUSTRY. Sophia Delipalla and Owen O Donnell.

Transcription

1 ESTIMATING TAX INCIDENCE, MARKET POWER AND MARKET CONDUCT: THE EUROPEAN CIGARETTE INDUSTRY Sopha Delpalla and Owen O Donnell January 999 Abstract Recent theoretcal work has shown that the ncdence of ad valorem and specfc taxes may dffer and each may be over or under-shfted onto consumers n the presence of mperfect competton. These results are used to derve a method of estmatng market power and conduct. An applcaton s made to the European cgarette ndustry. Prevous emprcal comparson of the prce effects of ad valorem and specfc taxes s lmted. For a group of countres wth broadly smlar cgarette ndustres, there s evdence of undershftng of both taxes, wth the specfc tax havng a sgnfcantly greater mpact on prce. The extremes of both perfect competton and monopoly can be rejected. Behavour s no less compettve than the equvalent of Cournot. JEL Classfcaton: H22, L3, L66 Keywords: Tax Incdence, Commodty Taxaton, Market Power, Mark-up, Conjectural Varatons, Cgarettes. Acknowledgements: We are grateful for comments receved at semnars at the Insttute for Fscal Studes, Cty, Kent, Sussex and York, and the 998 Publc Economcs Weekend at Essex. Correspondence Address: Sopha Delpalla, Department of Economcs, Keynes College, Unversty of Kent, Canterbury CT2 7NP, UK. Tel: (++227) ; Fax: (++227) ; emal: S.Delpalla@ukc.ac.uk

2 ESTIMATING TAX INCIDENCE, MARKET POWER AND MARKET CONDUCT: THE EUROPEAN CIGARETTE INDUSTRY. Introducton Tax ncdence s a fundamental ssue n Publc Economcs. Identfcaton of market power and measurement of the degree of competton are amongst the most mportant ssues n Industral Organsaton. The taxaton of cgarettes has been used to learn about each of these separately. (See Barzel, 976; Johnson, 978; Sumner and Ward, 98 on tax ncdence. See Sumner, 98; Bulow and Pflederer, 983; Sullvan, 985; Ashenfelter and Sullvan, 987 on market power and conduct.) In ths paper, cgarette taxaton s used to examne both sets of ssues. One am s to test the predctons of recent developments n the theory of commodty taxaton under mperfect competton (Delpalla and Keen, 992). The second am s to develop the lterature on the estmaton of market power and conduct by proposng a reduced form method whch provdes pont estmates of the prce-cost mark-up and the numbers equvalent of frms. Ths extends a result of Sullvan (985), who was only able to dentfy a lower bound for the latter parameter. In ths applcaton, the method nvolves comparng the comparatve statc effects of specfc and ad valorem commodty taxes on prce and quantty. However, t wll work whenever a varable whch shfts the cost functon and another whch pvots the demand curve can be observed. Recent theoretcal work on tax ncdence has shown that commodty taxes may be overor under-shfted onto consumers n the presence of mperfect competton (Seade, 985; Stern, 987). Moreover, the ncdence of ad valorem and specfc taxes may dffer, wth the prce effect of the former never exceedng that of the latter (Delpalla and Keen, 992; Skeath and Trandel, 994). Pror to the mperfect competton model, consderaton of the relatve prce effects of ad valorem and specfc taxes focussed on ther mpact on qualty n compettve

3 2 markets (c.f. Barzel, 976; Kay and Keen, 983, 99). The rankng of the relatve prce effects n ths envronment s consstent wth that generated by the olgopoly model. When qualty s measured n terms of some untaxed characterstc, a specfc tax may lead to an upgradng n qualty. Snce the ncrease n qualty per se tends to rase prce, the actual prce ncrease may exceed the (specfc) tax ncrease. An ad valorem tax bears on all commodty characterstcs whose value s reflected n consumer prce, provdng a dsncentve to mprove qualty. Note that ad valorem taxaton has a multpler effect, that s, to ncrease producer prce by, consumer prce has to ncrease by /( tv ) (tv s the ad valorem tax rate). Thus, when the ad valorem tax ncreases, t s lkely to lead to a reducton n qualty and, consequently, a prce rse lower than the amount of the tax ncrease. Emprcal comparson of the prce effects of specfc and ad valorem taxes s lmted. 2 Barzel (976) estmated prce effects by explotng state varaton n cgarette taxes n the US. No state employed both taxes smultaneously and only one state used an ad valorem tax. A dfferental effect of the two types of taxes was tested by examnng the sgnfcance of an nteracton between the level of tax and a dummy ndcatng whether the tax was ad valorem. He found overshftng of the specfc tax and could not reject full shftng of the ad valorem. He attrbuted the dfferent effects of the two types of taxes to qualty responses, the consstency of ths result wth mperfect competton not yet havng been recognsed. Johnson (978) generalses Barzel s specfcaton by allowng state specfc effects, as well as tme effects, and fnds overshftng of the specfc tax and undershftng of the ad valorem. The Cremer and Thsse (994), n a model of vertcal product dfferentaton wth two frms where each produces a varant of a dfferentated commodty, show that an ncrease n ad valorem taxaton can actually reduce the consumer prce. Ther explanaton s that ad valorem taxaton reduces the qualty of both varants, narrows the qualty gap and ntensfes prce competton. 2 In fact, emprcal work on commodty tax ncdence, n general, s sparse (rare examples are Besley and Rosen, 994, and Poterba, 996).

4 3 result s nterpreted as provdng further support for the qualty model. Sumner and Ward (98) queston ths nterpretaton on the grounds of mplausblty. They ask what s the nature of the qualty change whch s made to the product n response to a tax change and pont out that manufacturers do not produce a dfferent product for the one state levyng the ad valorem tax. An alternatve explanaton s offered for the apparent overshftng - prces may be rased at the tme of tax ncreases not only n response to the tax but also to compensate for accumulated mnor cost ncreases. Controllng for ths backlogged prce effect, Sumner and Ward fnd undershftng of both taxes and no sgnfcant dfference between the two. Ther suggested explanaton for undershftng s nterstate competton. Baltag and Levn (986) model cross-border shoppng explctly. They use the lowest prce of cgarettes n a neghbourng state to control for cross-state substtuton and fnd a small but sgnfcant effect. 3 All these studes test Barzel s hypothess ndrectly by lookng at the effect of taxes on cgarette prces. Sobel and Garrett (997) provde support for Barzel s theory through a more drect test, usng data on the relatve market shares of premum- and generc-brand cgarettes. 4 Baker and Brechlng (992) look at the effect of excse duty changes on prces n the UK. Ther fndngs suggest that for beer, sprts and petrol, changes n the specfc tax are fully reflected n changes n prces. For tobacco and wne, they fnd undershftng and overshftng respectvely. Full shftng of the ad valorem tax can never be rejected. However, as the authors acknowledge, there are only two changes n the ad valorem rate over the data perod, makng 3 Coats (995) estmates cross-border effects of state cgarette taxes by lookng at the response of state cgarette sales to state cgarette taxes. Barnett et al. (995) compare the effects of federal and state taxes. Smulaton results show that an ncrease n federal tax results n a greater ncrease n prce than does the same change n the average state and local tax. A possble explanaton s cross-border shoppng. Another explanaton s that manufacturers use federal tax ncreases as a sgnallng devce to co-ordnate a seres of prce ncreases (c.f. Harrs, 987). 4 Ther fndngs, although supportve of Barzel s theory, do not necessarly show that t was the qualty effects that Barzel captured n hs emprcal study.

5 4 t dffcult to have confdence n the robustness of the estmated ncdence of ths tax and to compare t wth that of the specfc tax. Such lack of varaton n the data, partcularly wth respect to the ad valorem rate, s a falure from whch all prevous attempts to estmate the relatve ncdence of specfc and ad valorem taxes have suffered. We avod ths lmtaton by usng data from the EU, where all member states levy both a VAT and an excse duty on tobacco, wth the excse duty consstng of both a specfc and an ad valorem element. The new emprcal ndustral organsaton (N.E.I.O.) lterature s concerned wth testng for market power and estmatng the degree of both market power and competton wthout usng accountng data on cost and/or proft (for surveys, see Bresnahan, 989; Carlton and Perloff, 994; Gerosk, 988). A dstncton can be made between structural and reduced form approaches to the problem (Hyde and Perloff, 995). The former s based manly on the conjectural varatons model and nvolves estmatng a structural market demand functon smultaneously wth supply relaton(s). Identfcaton of the degree of market power and conduct s acheved through comparatve statcs wth respect to some varable whch pvots the demand curve (Bresnahan, 982; Lau, 982). The advantage of the structural approach s ts power. Not only can market power be tested but estmates can be made of the prce-cost mark-up and the degree of competton wthn the ndustry. There are four man dsadvantages. Frst, data must be avalable on prce, output, nput prces and demand and cost shfters. Second, msspecfcaton of the structural demand and/or cost functon wll bas the estmates, and so the tests, of market power/conduct. Thrd, when usng ndustry level data, the supply relaton estmated does not correspond to the frst order condtons unless frms are homogeneous. Otherwse, the supply relaton s, n part, ad hoc and the parameters must be nterpreted as ndustry averages (Bresnahan, 989, p.030). Fnally, the conjectural varatons approach s vulnerable to theoretcal crtcsm. Gven ths, Bresnahan (989) clams an as-f nterpretaton of the estmated conduct parameter - t ndcates behavour s as compettve as-

6 5 f frms held certan conjectures. Corts (998) demonstrates ths argument s vald only under certan condtons. 5 Tests of market power whch do not nvolve estmaton of structural demand and supply relatons avod the above mentoned problems at the cost of losng power wth respect to the hypotheses whch can be tested and the parameters whch can be estmated. Hall (988) provdes a jont test of the hypotheses of perfect competton and constant returns to scale. The jont nature of the test mpedes nterpretaton somewhat. Estmates of the degree of market power and market conduct can be obtaned only by mposng further restrctons and wth addtonal nformaton avalable (Shapro, 987). Panzar and Rosse (987) are able to test the extreme cases of market conduct - perfect competton and monopoly - but do not obtan an estmate of the degree of market power or competton. Sumner (98) clamed the mpact of a unt tax n a reduced form prce equaton dentfed the ndustry average mark-up and the frm level elastcty. Bulow and Pflederer (983) demonstrated ths clam was vald only for specal cases of the demand functon. 6 Sullvan (985) lnked the method of Panzar and Rosse (op ct) wth that of Sumner (op ct) and Bulow and Pflederer (op ct) and showed that the effects of a unt tax n reduced forms for prce and quantty can be used to dentfy a lower bound on the numbers equvalent of frms. Ths allows testng of the hypothess of monopoly, but not competton. 7 We take ths lterature one step further by proposng a reduced form method whch allows dentfcaton of the prce-cost mark-up and the numbers equvalent of frms. The 5 Inference of market power from the estmated conduct parameter s vald only f the behavour underlyng the observed equlbrum s dentcal at the margn, and not just on the average, to a conjectural varatons game. 6 Genesove and Mulln (998) also note that dentfyng the conduct parameter through the responsveness of prce to cost alone s essentally dependent upon the demand specfcaton (p.37). 7 Ashenfelter and Sullvan (987) gve a non-parametrc mplementaton of the Sullvan (985) method.

7 6 hypothess of market power can therefore be tested and the degree of market power and competton wthn the ndustry estmated. The method works through comparng the prce effects of specfc and ad valorem commodty taxes. Non-equvalence s an ndcaton of market power. The ndustry average prce-cost mark-up s dentfed through takng the rato of the prce effects of the two taxes. Havng estmated the mark-up, a parameter reflectng the conduct of the ndustry (.e. the numbers equvalent of frms) s dentfed f the prce elastcty of market demand s known or can be estmated. The method proposed combnes the best of the structural and reduced form approaches descrbed above - t s powerful, yet parsmonous wth respect to data requrements and assumptons mposed. The methodology s applcable to other ndustres wth both specfc and ad valorem taxes or, more generally, where there s an observable varable whch shfts unt costs and another whch pvots the demand curve. The European cgarette ndustry, beng hghly concentrated, provdes an approprate context for tryng out our method of estmatng market power and to test the recent developments n the theory of tax ncdence allowng for mperfect competton. The estmates are not only of nterest n relaton to the academc economcs lterature but also as a source of nformaton for a varety of polcy dscussons. There has been a long runnng debate n Europe over the harmonsaton of cgarette taxes. Although all EU countres tax cgarettes heavly, there s a splt between those favourng ad valorem taxaton - roughly, the south - and those wth a more balanced tax structure - roughly, the north. These dfferences have mpeded fscal harmonsaton. Evdence on the relatve effects of the two taxes mght help to resolve the debate. The taxaton of cgarettes s motvated, n part, by publc health concerns. Evaluatons of the effectveness of taxaton as an ant-smokng nstrument have concentrated on the estmaton of prce elastctes of demand, adoptng the assumpton that taxes are fully shfted onto consumers (c.f. Chaloupka and Warner, 998). Tests of the valdty of ths assumpton are mportant n assessng the health polcy role, and the dstrbutonal effects, of

8 7 cgarette taxaton. The Tobacco Resoluton proposed n the U.S. and the hgh profle legal actons taken aganst cgarette manufacturers there have focussed attenton on the ndustry. Gven the scale of the tax ncreases contemplated n the Resoluton, knowledge of the degree of shftng of taxes onto prces and the conduct of the ndustry would be crucal n predctng the consequences of such legslaton (Bulow and Klemperer, 998). The next secton presents the theoretcal framework for analysng how ad valorem and specfc taxes affect prces. The comparatve statcs are used to develop the new reduced form method of dentfyng the prce-cost mark-up and the numbers equvalent of frms. In secton 3, we descrbe the European cgarette ndustry. The data are dscussed n secton 4 and the results presented n secton 5. Secton 6 concludes. 2. Market power and the relatve ncdence of ad valorem and specfc taxaton We consder the conjectural varatons model, as n Delpalla and Keen (op. ct.), only we look at the non-symmetrc equlbrum. In an ndustry wth n frms, the after-tax proft earned by frm s [( tv) P( X ) ts] x c( x ), () where P s the consumer prce, X s the ndustry output, x s the frm s output, c x ) s the ( frm s total cost of producng the gven level of output and ts and tv are the specfc and ad valorem tax rates respectvely. The strategc nteracton between frms s captured by dx dx [ 0, n]. Wth 0, conjectures are compettve ; corresponds to Cournot conjectures and n to tact colluson. The frst-order condton for proft maxmsaton s gven by ( tv)[ P( X ) P x ] c x ts 0, (2) x

9 wth subscrpts ndcatng dervatves. Dvdng (2) by 8 and summng over, yelds P ( tv) P X X c ts 0. (3) P x Solvng (3) for P and usng e X P P / X, P, (4) e c where x ts. tv (5) Comparatve statcs show that taxes affect prce as dp dts (6) ( tv) [ A E] dp dp and, (7) dtv dts where c x x A and E PXX X / PX denotes the elastcty of the slope of the nverse ( tv) np X demand functon. Equaton (6) s mmedate on applyng the mplct functon theorem to (3); (7) follows smlarly on notng from (3) that P PX X. (8) Comparng (6) and (7), usng (4) and denotng /( ), we get e

10 9 dp dts dp Pdtv. (9) That s, the rato of the margnal effects of the specfc and ad valorem tax s equal to, the mark-up parameter. Under perfect competton, ths parameter s equal to one and the two taxes have equvalent effects on prce. However, wth mperfect competton (.e. ), the prce effect of the specfc tax exceeds that of the ad valorem by a proporton gven by the value of the mark-up. Snce prces are set above margnal cost, an ncrease n cost due to a change n taxaton need not be reflected n an dentcal ncrease n prce. There s full shftng of a tax onto the consumer f the producer prce, p ( tv) P ts, s nvarant to the level of the tax. Then, the degree of tax shftng s gven by dp dts dp ( tv) dts (0) dp dp and ( tv) Pdtv Pdtv () Expressons (0) and () are less than, equal to and greater than zero wth undershftng, full shftng and overshftng respectvely. The outcome whch emerges depends upon the value of parameters related to the market structure, cost structure and demand elastcty of the ndustry. 8 Overshftng of the specfc tax s necessary but not suffcent for overshftng of the ad valorem tax. The realsm of an assumpton of homogeneous products can obvously be questoned. However, Anderson et al. (997) show the results of Delpalla and Keen (op. ct.) on the 8 Perfect competton s suffcent but not necessary for full shftng to occur. For example, n the presence of mperfect competton combned wth constant margnal costs and constant elastcty of demand, there wll be full shftng of the ad valorem tax (and overshftng of the specfc tax).

11 0 relatve ncdence of the two taxes carry over to a model wth horzontally dfferentated products n Bertrand-Nash olgopoly. 9 Note that from (4), e P e (2) The term on the left-hand-sde s the numbers equvalent of frms, whch can be estmated provded one has estmates of the two parameters on the rght-hand-sde - the mark-up () and the prce elastcty of demand (e). For the latter, one could use an extraneous estmate. Alternatvely, (2) can be rewrtten as. dx / dts dp / Pdtv dp / dts dp / dts (3) Provded one has data on market quantty, as well as prce, the margnal effects of the taxes n reduced form prce and quantty functons gve an estmate of the degree of competton wthn the ndustry, n addton to the mark-up. Wth both types of taxes, we are able to obtan pont estmates of both parameters, whereas Sullvan (985), wth only a specfc tax, could only get a lower bound for one of the parameters - the numbers equvalent of frms. 3. The European cgarette ndustry The emprcal analyss s based on data from the twelve members of the EU pror to ts expanson n 995. The European cgarette ndustry s charactersed by a hgh degree of concentraton. In 992/93, the top fve frms n each country held n excess of 90% of the 9 Whle Anderson et al. (997) relax the homogeneous product assumpton, ther model s more restrctve than Delpalla and Keen (992) n respect of the market structures and consumer preferences admtted.

12 market n every case (see Table ). 0 A major dfference n the nature of the markets across Europe arses from drect state nvolvement n France, Italy, Span and Portugal. In these countres, the state has an effectve monopoly on the manufacture and dstrbuton of domestc cgarettes and the market s even more concentrated than t s elsewhere. The remander of the market n these countres, and the vast majorty of the market n the other countres, s domnated by a group of Amercan and Brtsh multnatonals. The excepton s Denmark, where a prvate domestc company (Skandnavsk Tobak) enjoys an almost monopoly poston. There are also dfferences n the nature of the product. In the four countres where the state s nvolved n producton, the market s led by domestc brands made from European tobacco. In most of the other countres, Greece beng a notable excepton for most of the perod of our analyss, the Amercan tobacco brands of the multnatonals lead the market. Cgarette prces and taxes for the perod of analyss, , are summarsed n Table 2. The prces refer to the hghest sellng category, defned by prce (.e. the most popular prce category (MPPC)), n each country. There are large dfferences n gross prces, wth the lowest prces beng n southern Europe. Wth two exceptons, the real gross prce of cgarettes ncreased over the perod. 2 In a number of cases, the ncrease was substantal. The heavy burden of taxaton s ndcated by the large dfferences between gross and net prces. The smaller varance across countres n net prces ndcates tax dfferences are, to an extent, responsble for the dfferences n gross prces. The tax burden has ncreased n a number of countres and there s now a degree of consstency wth respect to the level of taxaton. Ths s 0 The fgure shown n the table for Greece s less than 90% but, as noted, ths refers to the share held by domestc producers only. It s lkely that ncludng MNCs would push the fgure above 90%. Even n ths case, an Anglo-Amercan multnatonal has a one thrd share n the company. 2 The fall n prce ndcated for Greece s apparent, rather than real, reflectng deprecaton of the currency.

13 2 the result of EU legslaton, whch now requres that the total tax burden be at least 70 percent of the gross prce. The lack of progress n harmonsng the structure of cgarette taxaton s apparent from the fnal two columns of Table 2. Accordng to the theory dscussed n secton 2, specfc taxaton leads to hgher prces, for a gven tax revenue. It s understandable that there s greater tolerance of ths type of taxaton n some of the countres of northern Europe, where smokng preventon movements are more frmly establshed. 3 In southern countres, smokng preventon - although growng - s a poltcally senstve ssue because of the cultural and economc mportance of tobacco. These countres prefer ad valorem taxaton snce, through the multpler effect, t ncreases the prce advantage to the local brands, often made from domestcally grown tobacco, relatve to those of the multnatonals. Theory also predcts specfc taxaton s more advantageous for proft relatve to ad valorem (Delpalla and Keen, 992). The multnatonal companes predomnant n the north of Europe would therefore be expected to lobby for ths type of taxaton. On the other hand, state producers mght be more nterested n tax revenue than proft and would be expected to favour ad valorem taxaton. It s strkng that n Portugal, where there s effectvely a state monopoly, the burden of taxaton s among the hghest n the communty, yet gross prces are among the lowest. The dfferences n the preferred structure of cgarette taxaton have mpeded agreement on harmonsaton. The frst EU drectve ssued n 972 (Drectve 72/464/EEC) nstructed all member states to ntroduce a mxed tax structure. The specfc tax should be not less than 5% and not hgher than 75% of the total excse duty. The drectve was clearly n favour of predomnantly ad valorem taxaton; at that tme the majorty of EC members had an entrely 3 Moreover, Cnossen (992) argues that specfc taxaton s a better nstrument to nternalse the external costs that smokng mposes, snce t hts the cause of the costs drectly and does not tax tems that do not contrbute to the costs, such as wrappers, or even mtgate the effects of smokng, such as flters.

14 3 ad valorem tax structure. Shortly afterwards, Denmark, Ireland and the UK, countres wth predomnantly specfc taxaton, joned the Communty. A second drectve was approved n 977 (Drectve 77/805/EEC) accordng to whch the specfc tax should be between 5% and 55% of the total tax burden ncludng the VAT. Ths second stage was extended fve tmes untl 985, when t was extended ndefntely. After several years of dsagreement, n 992, t was agreed that the overall excse duty should be no less than 57% of the fnal retal prce of the most popular prce category (all taxes ncluded), and the VAT should be at least 5% of the fnal retal prce (nclusve of excses). These drectves mpled a mnmum overall tax level on cgarettes of 70% of the retal prce. The rato of specfc to total taxaton should be the same as n the 977 Drectve. From Table 2, t s apparent that there s a tendency for countres to locate toward ether of the extreme bounds on ths rato. 4. Data We compare the effects of specfc and ad valorem taxes on cgarette prces by regressng prce data from twelve European countres over sxteen years on correspondng tax data and controls for other determnants of prces (demand and cost condtons). For ten countres, the data cover the perod ; for Span and Portugal, The data are for prces and taxes n operaton at January of each year. 4 The prce data are for 000 cgarettes n the most popular prce category (MPPC), whch wll vary across countres and, potentally, also across tme. Varaton across countres can be dealt wth through country specfc effects. Varaton across tme s more dffcult to accommodate snce ths tme effect would not be common across countres. An nference problem would arse f swtches n the MPPC were 4 For two years we do not have data specfc to January. For 982 we use May data and for 995, July. Ths s unlkely to be a sgnfcant problem gven there s lttle ntra-year varaton n the tax and prce seres. For the perod , we have quarterly data. Estmates obtaned from annual and quarterly data showed lttle dfference.

15 4 correlated wth tax changes. Varous sources have been checked to dentfy any changes n the MPPC over the sample perod. There are two cases of large jumps n the prce of the MPPC whch appear, at least n part, to arse from a swtch of the leadng prce category - France and Greece In the former case, the problem has been dealt wth by specfyng dfferent group effects for the perods and In the case of Greece, there are nsuffcent data ponts after the swtch to allow a separate group effect for ths perod and so the Greek seres has been truncated at 993. The specfc tax s the monetary amount leved on 000 cgarettes and the ad valorem rate s the sum of the ad valorem excse duty and VAT expressed as a percentage of the tax nclusve retal prce. Sources for all of the data are gven n the Appendx. As a control for cost varaton, we nclude labour costs per worker n the manufacturng sector of the tobacco ndustry. 5,6 As wth all of the control varables, the data are for the year precedng the January date to whch the prce and tax data refer. GDP per capta s ncluded as a determnant of the level and prce elastcty of demand. Wth the excepton of GDP per capta, whch s denomnated n purchasng power standards, all monetary varables are converted to ECUs. The ECU exchange rate s ncluded, as an addtonal control, to avod spurous correlaton arsng from deprecaton or apprecaton of a currency. All monetary denomnated varables were deflated to 985 prces usng country specfc consumer prce ndces. 5 For France and Luxembourg data specfc to the tobacco ndustry were not avalable. Labour costs per worker across the whole of manufacturng ndustry were used nstead. The approprate data s absent for Ireland before 985. For ths perod, data specfc to food, tobacco and alcohol manufacturng are used. For many of the countres, labour cost data were not avalable for 996. We used a forecast based upon an assumpton of no real change n labour costs per worker between 995 and Unt labour costs n tobacco manufacturng were used as an alternatve cost control but labour costs per worker were found superor wth respect to sgnfcance and dagnostc tests. As a control for captal costs, real long term nterest rates were ncluded ntally but were found not to be sgnfcant and could be excluded wthout affectng the remanng coeffcents.

16 5 5. Results Dfferences n the nature of the cgarette ndustry across Europe were dscussed n secton 3. Whle some of these dfferences can be dealt wth n estmaton through the ncluson of country effects, others affect not only prce levels but the tax responsveness of prces. Indeed, accordng to the theoretcal results, dfferences n market power and conduct should be reflected n tax-prce relatonshps. In the context of a lnear n levels specfcaton estmated by OLS wth country specfc ntercepts, the restrcton of homogenety n slope coeffcents s decsvely rejected [F=62.30 (p=0.0000)]. We therefore look for sub-sets of countres across whch ths restrcton has greater valdty. Country specfc tax shftng parameters, calculated from ndvdual country prce regressons, are presented n Table 3. 7 Gven small sample szes, these estmates cannot be expected to be partcularly accurate. However, they are useful n dentfyng mportant dfferences n tax-prce relatonshps across countres. In most cases, the estmates suggest undershftng of the ad valorem tax. In three countres (Denmark, the Netherlands and Portugal), ths undershftng s sgnfcant. In only one case (Italy) there s evdence of sgnfcant overshftng of the ad valorem tax. In contrast, sx countres show sgnfcant overshftng of the specfc tax, wth only one (Netherlands) ndcatng sgnfcant undershftng. The theoretcal predcton that the prce effect of the specfc tax exceeds that of the ad valorem s confrmed n all but two cases (Germany and UK). In nether of these two theoretcally nconsstent cases does the dfference between the tax effects reach statstcal sgnfcance. On the other hand, there are sx cases n whch the specfc tax has a sgnfcantly greater mpact on prce than the ad valorem. 7 See notes to Table 3 for a descrpton of the estmaton procedure. Regresson coeffcents are avalable from the authors.

17 6 Estmates of the overshftng of the specfc tax n France and Luxembourg are very large and result n extremely large ratos of specfc to ad valorem effects. Such results probably reflect peculartes n the market for cgarettes n each of these countres. In France, the market s led by a state producer, whereas cross-border shoppng has a very large mpact on the market n Luxembourg. 8 These features mght be expected to result n complex relatonshps between tax and prce and to render the theoretcal model we are nterested n testng napplcable. It s notceable that the rato of the two tax effects s also large for two (Portugal and Span) of the remanng countres n whch the state has monopoly control of domestc producton. Ths rato s also large n the case of Greece, where domestc manufacturers, usng domestcally grown tobacco, have a large share of the market. Italy, the fnal country wth state producton, s dstngushed by estmates of a large degree of overshftng of both taxes. As mght be antcpated, results from ndvdual country regressons suggest state producton (France, Italy, Portugal and Span), producton from domestcally grown tobacco (Greece) and a very large amount of cross-border shoppng (Luxembourg) affect tax-prce relatonshps. We therefore concentrate on estmates derved from a group of countres wthout these features (Belgum, Denmark, Germany, Ireland, Netherlands and UK Group ). As llustrated by Table, these countres dsplay a degree of homogenety wth respect to market structure. Wth the excepton of Denmark, they are all domnated by a small number of multnatonals and there s smlarty across the countres n the most popular type of cgarette. The rato of the tax effects, whch accordng to the mperfect competton model s the mark-up parameter, s estmated to be larger n Denmark than n the others from ths group. Ths mght reflect the greater degree of market concentraton n ths country. In the 8 Up to 80% of cgarettes sold n Luxembourg are purchased by non-resdents (European Bureau for Acton on Smokng Preventon, 995).

18 7 nterests of effcency, we choose to nclude Denmark n the core group countres and comment on the senstvty of the results to ts excluson. Estmates from poolng the remanng countres (France, Greece, Italy, Luxembourg, Portugal and Span - Group 2) are also presented. The peculartes of the markets n these countres make them less nterestng from the pont of vew of testng the model of mperfect competton, however, good estmates of the average degree of tax shftng across these countres are of nterest n ther own rght. Estmates of prce regressons for Group and 2 countres are presented n Table 4. The wthn groups (WG) estmator s used - Hausman tests reject the random effects specfcaton. Tme effects are sgnfcant and ncluded for Group but not Group 2. RESET tests favoured a levels specfcaton for Group and a log transformaton of all varables for Group 2. Quadratc terms are ncluded where they were found to be sgnfcant. 9 All varables take the antcpated sgns n each regresson, wth the excepton of the negatve wage effect n Group 2. The hgh R 2 values, whle not unusual for ths type of data and analyss, mght suggest problems of non-statonarty. Gven the length of the tme seres, no formal testng for unt roots s undertaken. However, t s reassurng that, at least for Group, estmaton n frst dfferences gave very smlar results to those presented. The Durbn-Watson values are also reassurng n ths respect. A dynamc specfcaton was tred, through the ncluson of a lagged dependent varable, but was not found to be approprate n ether case. The assumed homogenety of the slope coeffcents s rejected for Group 2 but cannot be rejected at the % level of sgnfcance for Group. 20 Smlarty n the cgarette ndustres across Group countres appears to gve rse to smlar tax-prce relatonshps and justfes poolng of the data across these countres. Whle the wthn groups estmator allows for 9 Interacton effects were not found to be sgnfcant. 20 The restrcton s rejected at 5% sgnfcance for Group. Provded varaton n the parameters s random, the WG estmator gves consstent estmates of the mean (across country) vector of parameters (Hsao, 986, p. 32).

19 8 correlaton between country fxed effects and the regressors, there remans the potental for endogenety of the tax varables. For example, the EU rules on the level and structure of cgarette taxaton may lead to dependence of the taxes on prces. Countres at, or close to, the lower lmt on total taxes as a percentage of the retal prce (70%) must rase taxes n response to a prce ncrease. Further, beng close to the lower (5%) or upper (55%) threshold for the specfc tax as a proporton of total taxes wll requre a shft n the balance of taxaton followng certan prce movements. Hausman tests, based on comparson between WG and two-stage WG estmates n whch the tax varables are nstrumented, ndcate the null of exogenety cannot be rejected for ether group. 2 Ths s perhaps to be expected for the Group countres, gven only one (Germany) had a tax burden very close to the 70% threshold. Further, wth the excepton of Belgum, these countres have a balanced structure of ad valorem and specfc taxaton. Only prce falls, not the more lkely prce rses, cause problems for such countres attemptng to keep wthn the upper lmt on the tax structure rato. Only two countres (Ireland and UK) were very close to ths lmt anyway. Tax ncdence, market power and conduct parameters calculated from the coeffcents on the tax varables and usng sample mean values are presented n Table 5. The estmated parameters dffer across the two groups of countres. In Group, there s sgnfcant undershftng of both types of taxes. A unt ncrease n tax arsng from a change n the ad valorem rate results n an ncrease n prce of 0.72, whereas a unt ncrease n the specfc tax ncreases prce by The dfference n the prce effects s statstcally sgnfcant. These results are consstent wth the theoretcal predctons n secton 2: under mperfect competton 2 Gven cgarette taxes are set wth some regard to the state of the macroeconomy, the followng were selected as nstruments: real growth rates of prvate consumpton and GDP, the general government defct/surplus as a percentage of GDP and the unemployment rate. Gven the quadratc specfcaton, followng Kelejan (97), levels, squares and crossproducts of all the exogenous varables are used as nstruments.

20 9 there need not be full-shftng of commodty taxes and a specfc tax wll have a greater mpact on prce than an ad valorem. The theory also suggests that the rato of the prce effects of the two taxes s an estmate of the prce-cost mark-up (). The mark-up s estmated to be There are no prevous estmates avalable for Europe wth whch to compare ths estmate. Applebaum (982), followng a structural approach, estmates a mark-up of 2.84 n the U.S. tobacco ndustry. Snce we have no quantty data correspondng to the prce data we employ, that s, the MPPC, we use (2), rather than (3), to estmate the numbers equvalent of frms. Pont estmates and 95% confdence ntervals are gven n Table 5 for varous values of the prce elastcty of demand ( e). Estmated compettveness s lower the hgher the assumed value of the prce elastcty. The lterature provdes a wde range of estmates of the latter, wth some clusterng around a value of -0.4 (Chaloupka and Warner, 998). At ths value, our estmate of the numbers equvalent of frms s.4 for Group. Ths les wthn the range of estmates of the lower bound on the numbers equvalent of frms n the U.S. cgarette ndustry calculated by Sullvan (985). From the fgures provded n Table, t s apparent that, n general, there are fve or sx frms operatng n each of the markets ncluded n Group. At a market prce elastcty of -0.4, the 95% confdence nterval for the numbers equvalent does not nclude 6, suggestng frms n these markets are behavng n a manner whch s more compettve than the equvalent of Cournot. Assumng hgher values for the prce elastcty, the equvalent of Cournot behavour could not be rejected. However, even assumng a untary prce elastcty, the confdence nterval does not nclude, allowng rejecton of the hypothess of cartel behavour. 22 The pattern and statstcal sgnfcance of the results for Group are not changed f Denmark s excluded. It s reassurng, gven the near monopoly supply n Denmark, that ths excluson results n a fall n the estmate of the mark-up.

21 20 The results for Group 2 ndcate sgnfcant overshftng of both taxes. Overshftng of the specfc tax s partcularly marked - a unt ncrease n tax s estmated to rase prce by more than two. The dfference n the prce effects of the two taxes s sgnfcant at 0% but not 5%. Gven the presence of state producers wthn ths group, t mght be argued that the results should not be nterpreted accordng to the theory of secton 2, whch assumes proft maxmsaton. If such an nterpretaton s made, the results suggest a mark-up of.47. A hgher mark-up for ths group of countres than for Group s consstent wth a pror expectaton gven knowledge of dfferences n market structure. The numbers equvalents of frms estmates are smaller than for the frst group of countres, suggestng less compettve behavour. The equvalent of Cournot behavour could not be rejected for a value of the prce elastcty as low as Assumng a prce elastcty at, or above, -0.6 would not allow rejecton of cartel behavour. Whle the potental napplcablty of the theoretcal model to Group 2 countres must be acknowledged once more, t s nterestng that the estmate of low compettveness s consstent wth a pror expectaton gven the very hgh concentraton n these countres. 6. Conclusons Ths paper had three prncple ams. Frst, to test predctons from recent theory of commodty tax ncdence n mperfectly compettve markets. Second, to ntroduce and apply a new method of estmatng market power and conduct. Thrd, to nform polcy debates on tax harmonsaton n Europe and the use of cgarette taxaton as an nstrument of health polcy. The results reveal that commodty taxes are not always fully shfted onto consumers. For a group of northern European countres wth smlar market structures and qualty of cgarettes (Group ), there s evdence of undershftng of both ad valorem and specfc taxes, wth sgnfcant dfferences between the two. In a remander group of manly southern European

22 2 countres, there appears to be overshftng of both taxes, wth, agan, a sgnfcantly greater effect of specfc taxaton. Whle these results are consstent wth the predctons of the mperfect competton model dscussed n secton 2, qualty effects could also be responsble for the specfc tax havng a larger mpact on prce than the ad valorem. However, accordng to Kay and Keen (99), nether undershftng of both taxes, nor overshftng of both, s a plausble scenaro under the qualty model. Also, t s dffcult to dentfy changes n qualty n these markets, to whch undershftng of the ad valorem n Group and overshftng of the specfc n Group 2 could be attrbuted. Imperfect competton s a more persuasve explanaton. 23 For Group, for whch the theoretcal model has greater relevance, the results allow rejecton of both market extremes - perfect competton and cartel behavour. More specfcally, frms behavour n these markets would appear to be no less compettve than the equvalent of Cournot and are probably more compettve than ths. If the results from Group 2 are nterpreted wthn the context of the theoretcal model, they suggest less compettve behavour than n Group, wth the equvalent of Cournot not beng rejected and, at perhaps mplausbly hgh prce elastctes, cartel behavour not beng rejected. Our emprcal fndng that the specfc tax has a greater mpact on prce makes dfferences n preferences for the form of cgarette taxaton across Europe understandable. If northern governments want hgh prces, to satsfy the health lobby, and hgh profts, to please the multnatonals, specfc taxaton s the preferred opton. Governments n southern Europe are less exposed to these lobbes and favour ad valorem taxaton n order to mantan the prce advantage to the domestc products. Our emprcal confrmaton of the dfferental effect of the two types of tax suggests there wll be lttle progress n harmonsng the structure of 23 Another explanaton for the undershftng n Group s cross-border shoppng. Whle ths s a growng ssue, t unlkely to have been sgnfcant for the greater part of the perod covered n ths analyss. Also the problem s reduced by the excluson of Luxembourg from ths group.

23 22 cgarette taxaton across Europe, provded governments contnue to pursue dfferent objectves. Further, the fndng that the effect of a gven tax vares across Europe makes harmonsaton even less lkely. Estmates of the dstrbutonal effects of taxaton are typcally generated under the assumpton that commodty taxes are fully shfted. Our estmates show that ths assumpton does not always have emprcal valdty. Under the assumpton of full shftng, cgarette taxaton has been found to be regressve. Gven the overshftng found n the south, concerns, f any, over such regressvty should be ntensfed. On the other hand, cgarette taxes mght not be as regressve as s thought n the north, gven evdence of undershftng, partcularly of the ad valorem tax. In both cases, the emprcal fndngs suggest a more careful analyss of tax ncdence. A smlar warnng apples to analysts of cgarette taxaton as an nstrument of health polcy.

24 23 Appendx - Data sources The prce and tax data are taken from the Summary of Tax Structures on Cgarettes n E.C. Member States obtaned from the European Commsson (D.G. XXI) Excse Duty Tables and the Confederaton of European Communty Cgarette Manufacturers. Total labour costs and employment n the tobacco (manufacturng) ndustry were suppled by Eurostat from ther DEBA database. GDP at market prces n current prces and current Purchasng Power Standards per capta were obtaned from varous Eurostat sources ( Eurostat Natonal Accounts ESA: aggregates 970-9, Eurostat Yearbook 996, Eurostat NEWCRONOS Database). The seres was converted to 985 prces usng the Consumer Prce Index (CPI). Natonal CPIs were obtaned from Eurostat s NEWCRONOS database. Prce and specfc tax data were deflated usng the CPI specfc to January each year. Other varables were deflated usng the CPI for the approprate year. In 997 Eurostat changed from usng Natonal CPIs to ts new Harmonsed Indces of Consumer Prces (stll country specfc but calculated usng a common methodology). Prce and tax data for January 997 were deflated usng ths new CPI seres. ECU exchange rates were obtaned from Eurostat s NEWCRONOS database.

25 24 REFERENCES Anderson, S.P., A. de Palma and B. Kreder, 997, Tax ncdence n dfferentated product olgopoly, mmeo, Unversty of Vrgna. Applebaum, E., 982, The estmaton of the degree of olgopoly power, Journal of Econometrcs, 9, Ashenfelter, O. and D. Sullvan, 987, Nonparametrc tests of market structure: An applcaton to the cgarette ndustry, The Journal of Industral Economcs, 35, 4, Baker, P. and V. Brechlng, 992, The mpact of excse duty changes on retal prces n the UK, Fscal Studes, 3, 2, Baltag, B.H. and D. Levn, 986, Estmatng dynamc demand for cgarettes usng panel data: The effects of bootleggng, taxaton and advertsng reconsdered, The Revew of Economcs and Statstcs, 68, Barnett, P.G., T.E. Keeler and T-W Hu, 995, Olgopoly structure and the ncdence of cgarette excse taxes, Journal of Publc Economcs, 57, Barzel, Y., 976, An alternatve approach to the analyss of taxaton, Journal of Poltcal Economy, 84, 6, Besley, T.J. and H.S. Rosen, 994, Sales taxes and prces: An emprcal analyss, mmeo, Prnceton Unversty. Bhargava, A., L. Franzn and W. Narendranathan, 982, Seral correlaton and the fxed effects model, Revew of Economc Studes, 49, Bresnahan, T.F., 982, The olgopoly soluton concept s dentfed, Economcs Letters, 0, Bresnahan, T.F., 989, Emprcal studes of ndustres wth market power, n R. Schmalensee and R.D. Wllg, Handbook of Industral Organsaton, Volume II, Elsever, pp Breusch, T.S. and A.R. Pagan, 979, A smple test for heteroskedastcty and random coeffcent varaton, Econometrca, 47, Bulow, J.L. and P. Klemperer, 998, The tobacco deal, Brookngs Papers on Economc Actvty, forthcomng. Bulow, J.I. and P. Pflederer, 983, A note on the effect of cost changes on prces, Journal of Poltcal Economy, 9,, Carlton, D.W. and J.M. Perloff, 994, Modern Industral Organzaton, 2nd Edton, Harper Collns, New York. Chaloupka, F.G. and K.E. Warner, 998, The Economcs of smokng, The Handbook of Health Economcs, forthcomng. Cnossen, S., 992, Intrgues around the tobacco excse n the European Communty, Internatonal Tax Revew, 2, Coats, R.M., 995, A note on estmatng cross-border effects of state cgarette taxes, Natonal Tax Journal, XLVIII, 4,

26 25 Corts, K.S., 998, Conduct parameters and the measurement of market power, Journal of Econometrcs, forthcomng. Cremer, H. and J-F Thsse, 994, Commodty taxaton n a dfferentated olgopoly, Internatonal Economc Revew, 35, Delpalla, S. and M.J. Keen, 992, The comparson between ad valorem and specfc taxaton under mperfect competton, Journal of Publc Economcs, 49, European Bureau for Acton on Smokng Preventon, 995, Tobacco and health n the European Unon - An overvew, Brussels. Genesove, D. and W.P. Mulln, 998, Testng statc olgopoly models: conduct and cost n the sugar ndustry, , RAND Journal of Economcs, 29, 2, Gerosk, P.A., 988, In pursut of monopoly power: Recent quanttatve work n ndustral economcs, Journal of Appled Econometrcs, 3, Hall, R.E., 988, The relatonshp between prce and margnal cost n U.S. ndustry, Journal of Poltcal Economy, 96, Harrs, J.E., 987, The 983 ncrease n the federal cgarette excse tax, Tax Polcy and the Economy,, 87-. Hyde, C.E. and J.M. Perloff, 995, Can market power be estmated?, Revew of Industral Organzaton, 0, Hsao, C., 986, Analyss of Panel Data, Cambrdge Unversty Press. Johnson, T.R., 978, Addtonal evdence on the effects of alternatve taxes on cgarette prces, Journal of Poltcal Economy, 86, 2, Kay, J.A. and M.J. Keen, 983, How should commodtes be taxed?, European Economc Revew, 23, Kay, J.A. and M.J. Keen, 99, Product qualty under specfc and ad valorem taxaton, Publc Fnance Quarterly, 9, Kelejan, H., 97, Two stage least squares and econometrc systems lnear n parameters but nonlnear n the endogenous varables, Journal of the Amercan Statstcal Assocaton, 66, Lau, L., 982, On dentfyng the degree of compettveness from ndustry prce and output data, Economcs Letters, 0, Panzar, J.C. and J.N. Rosse, 987, Testng for monopoly equlbrum, Journal of Industral Economcs, 35, Poterba, J.M., 996, Retal prce reactons to changes n state and local sales taxes, Natonal Tax Journal, 39, Seade, J., 985, Proftable cost ncreases and the shftng of taxaton: Equlbrum responses of markets n olgopoly, Unversty of Warwck Dscusson Paper no Shapro, M., 987, Measurng market power n U.S. ndustry, Natonal Bureau of Economc Research Workng Paper No Skeath, S. E. and G. E. Trandel, 994, A Pareto comparson of ad valorem and specfc taxaton n noncompettve envronments, Journal of Publc Economcs, 53, 53-7.

27 26 Sobel, R.S. and T.A. Garrett, 997, Taxaton and Product qualty: New evdence from generc cgarettes, Journal of Poltcal Economy, 05, 4, Stern, N.H., 987, The Effects of taxaton, prce control and government contracts n olgopoly and monopolstc competton, Journal of Publc Economcs, 32, Sullvan, D., 985, Testng hypothess about frm behavour n the cgarette ndustry, Journal of Poltcal Economy, 93, 3, Sumner, D.A., 98, Measurement of monopoly behavour: An applcaton to the cgarette ndustry, Journal of Poltcal Economy, 89, 6, Sumner, M.T. and R. Ward, 98, Tax changes and cgarette prces, Journal of Poltcal Economy, 89, 6,

28 27 Table : Market Shares of Fve Leadng Cgarette Frms, 992/93 Top 5 Frms Total of Top 5 Total of Multnatonals State producer BELGIUM 30% % 94.8% 0% DENMARK 78.8 n.a. n.a. n.a n.a GERMANY GREECE SPAIN FRANCE IRELAND 2 n.a. n.a. n.a. n.a. n.a ITALY n.a. n.a. n.a LUXEMBOURG n.a. n.a. n.a. n.a. n.a. n.a. n.a. 0 NETHERLANDS n.a. > PORTUGAL U.K n.a. >92.5 > Source: European Bureau for Acton on Smokng Preventon (995). Notes. Shares of total sales held by 5 top domestc producers. The fgure for multnatonals s mports plus cgarettes produced for MNCs under lcense. 2. The MNCs account for almost the whole market.