Documents de Travail du Centre d Economie de la Sorbonne

Transcription

1 Documents de Traval du Centre d Econome de la Sorbonne halshs , verson - Oct 008 Are Small countres leaders of the European tax competton? Ncolas CHATEAIS, Mathlde PEYRAT Mason des Scences Économques, 06- boulevard de 'Hôptal, Pars Cedex 3 ISSN : 955-6X

2 Are Small countres leaders of the European tax competton? Chatelas Ncolas CES TEAM Unversté Pars Panthéon Sorbonne and Insttut CDC pour la recherche halshs , verson - Oct 008 Peyrat Mathlde ESSEC Busness School and CES MATISSE Unversté Pars Panthéon Sorbonne MathldePeyrat@malxunv-parsfr Correspondng author Abstract: The am of ths paper s to develop a better understandng of the lterature dealng wth strategc fscal behavours of small EU countres usng estmatons of tax reacton functons of competng natonal governments Dervng a smple model of tax competton n a Nash and a Stackelberg game, we use panel data and tools from spatal econometrcs to examne the role of small countres n tax competton wthn the enlarged European Unon We fnd that nteractons are stronger among smaller EU countres than between larges ones and rates set n small countres nfluence those n bg countres Fnally, small countres located n the centre of the EU have more nfluence on tax polces choces of bg countres than small countres located n the perphery of EU Keywords: Strategc Interactons, Tax Behavours, Spatal Econometrcs, European Unon, Tax Competton, Small Countres JE Classfcaton: E6, F, F, F3, H30, H3, H73, H77, R

3 Introducton halshs , verson - Oct 008 Snce the Treaty of Maastrcht (99) and the establshment of real and nomnal convergence rules, several economc regulaton tools such as monetary or exchange polcy can no longer be used n cyclcal adjustments or n response to polces n other neghbourng countres Countres are now competng through tax and fscal polcy Indeed, the statutory corporate tax rate has dropped snce 995 n all European countres Ths competton arses from both the recognton of dfferent tax regmes wthn EU by most moble tax bases n ther locaton choce and the un-cooperatve commtment of some countres to attract these tax bases Fscal nteractons between countres are derved from the exstence of externaltes (postve or negatve) of tax and fscal polces among countres Consequently, choces of tax polces are not ndependent In addton, countres are competng to attract moble tax bases (captal, profts, hgh ncomes ) localzed n ther neghbours Knowng almost perfect spatal moblty of the tax base consttuted by captal and profts, each natonal government adopts non-cooperatve behavours when settng ther corporate tax rate Ths strategy ams to attract productve captal and / or profts n order to ncrease the producton and / or ncome tax In ths artcle, we test whether governments nfluence each other n ther choce of tax polces and specfcally n tax rate settng We wll try to determne whether small countres, defned by ther geographc sze, are tax competton leaders n Europe Ths s based on the hypothess of a sze effect n small countres whch can encourage the pursut of a tax rate lowerng strategy to attract moble taxable bases Indeed, the captal transfer from the bg country that results s relatvely hgh n proporton to ts producton or ts GDP Ths captal transfer offsets the ntal loss of tax revenue The ssue of strategc nteractons between local governments has been often treated n the lterature However between states, t s ncreasngly an mportant topc of research Thus, we can make reference to works from Altshuler and Goodspeed (00), Devereux, ockwood and Redoano (00), Redoano (003) who obtan a postve Nash reacton functon for European countres and those of Ruz (006), who concludes a lack of fscal nteracton n Europe Most of tax competton models and estmates that follow result from a Nash framework However, we can examne f some countres do not observe fscal and tax polces evolutons among ts neghbours n order to react n a second tme Thus, Gordon (99) and Altshuler and Goodspeed (00) deal wth strategc nteractons n a theoretcal Stackelberg world In contrast to these two artcles suggestng that the Unted States has played a leadng role towards Europe n tax competton especally wth the TRA (Tax Reform Act 986), we test emprcally the role of small countres towards larger n the EU-5 In addton, we nvestgate whch countres or whch types of small countres are leaders of the tax competton We dstngush two groups of small countres whch can nfluence bg countres Emprcally, t seems that small countres close to the centre of the European Unon, notably Belgum, have the most nfluence on large countres at the core and therefore are the real nstgators of tax competton Thus, large countres always adopt a poston n lne wth respect to these small countres also at the centre The artcle s organzed as follows We wll frst state tax competton n Europe, recallng the theoretcal role played by small countres n tax competton based on a quanttatve analyss (secton ) We wll then present a

4 theoretcal model (secton 3) that we econometrcally estmate (Secton 4) Fnally, we wll present our fndngs (secton 5) to conclude on the mportant role of small countres n the European tax competton Tax competton wthn the European Unon Faced wth the exstence of a wde varety of taxaton systems wthn the European Unon, frms and partcularly multnatonals (MNF), may fnd benefcal to ntegrate ths varable n ther decson-makng strategc processes (localzaton, nvestments) Knowng that, some natonal governments can opt for an aggressve tax strategy n order to attract these productve enttes, or at least ther profts, and thus compel other governments to partcpate n the competton game due to hgh captal moblty In the European Unon case, t seems that small countres are at the orgn of tax competton and that large countres react to ths aggressve strategy halshs , verson - Oct 008 Why small countres practse tax competton? Exstence of a sze effect In the tradtonal lterature, the noton of sze s often defned by the market sze For some small countres, the tax polcy has to compensate for the small sze of the domestc market If a company s located n a large country, t can take advantage of the relatvely large domestc market at ts dsposal to explot ncreasng returns to scale In contrast, the market sze beng relatvely restrcted n a small country, companes can not thus beneft of these scale economes and prefer to locate near the large market Therefore, governments of small countres would beneft from havng a less heavy tax system n order to offset ther ntal market sze handcap However, on a European scale, the domestc market sze s not a very relevant argument Indeed, snce tarff barrers are free or very low and transport costs are reduced, companes have free access to the European market The relatvely small market sze s not thus a convncng argument to justfy a decrease n tax rate However, another effect lnked to ts small sze may encourage a country to practce a tax rate reducton strategy A relatvely low tax rate results n a captal transfer towards the country settng ths rate For a small country, the captal transfer that result s relatvely large n proporton to ts producton or ts GDP, whch largely offsets the ntal loss of tax revenues On the contrary, larges countres have nterest n less reactng to the tax competton because cuttng taxrates do not attract suffcent captal relatve to ther producton and ntal losses of tax ncomes are not compensated by productve transfers When a change of the tax rate occurs, dfferent consequences wll sustaned accordng to the sze from the country concerned Ths change has an mpact of wdth dfferent on the captal per capta rato dependng on the sze of the country Ths s due to greater captal elastcty to tax rate n small countres Thus, at the Nash equlbrum n a two country model, the small country wll be encouraged to set a tax rate lower than n the bg foregn country For example, Bucovetsky (99) and Wlson (99) have proposed a theoretcal framework ncorporatng two jursdctons havng a dfferent number of nhabtants Orgnally, no captal transfer would be excepted f the two localtes adopt the same tax rate The exchange wll thus take place only because the two jursdctons have set dfferent tax rates As n prevous models (Tebout 956), captal taxaton s used to fnance a publc good and leads to a captal market equlbrum nvolvng equal after-tax returns Consequently, tax rates set n a localty determne captal stock n the other one, and thus ts revenue from taxes Contrary to a theoretcal framework wth two 3

5 dentcal jursdctons, the captal margnal productvty elastcty to tax rate s not any more equal n both localtes Indeed, the amount of work avalable s lower n the small jursdcton, the elastcty s therefore greater Thus, captal flght wll be hgher n a small country than n a large f they adopt an equal tax rate ncrease Ths externalty s even more mportant as the jursdcton sze s small In concluson, beng more nfluenced by tax competton, the small localty may fnd a benefce n establshng ts tax rates below those set n larges ones Internatonal tax competton challenges mply to refer any more to varous tax bases elastctes n tax polcy elaboraton The theory (Bucovetsky, 99 and Wlson, 99) predcts that at equlbrum, large countres of the European Unon, havng captal elastcty lower than smaller countres, choose hgher moble bases tax rates than smaller members Ths proposal has been emprcally verfed by Huznga-Ncodème (006), whose regressons show a sgnfcant and robust relatonshp between the corporate tax burden and the sze of ther home country measured by the logarthm of GDP halshs , verson - Oct 008 Tax rates Reductons n small countres In ths secton we wll pont out an overall reducton n statutory corporate tax rates snce 995 n Europe We wll show a more pronounced declne n small countres of the European Unon Before, we justfy the choce of statutory rate compared wth other ndcators of tax pressure n order to deal wth tax competton Imperfect ndcators of tax pressure The theoretcal or statutory tax rate s the most frequently used ndcator n economcs to measure the tax burden because t s relatvely smple to evaluate However, ths ndcator may lack of effcency because t does not take nto account the tax base or potental tax shelter In addton, the calculaton of the tax base s rather dffcult to assess accurately To remedy ths shortcomng, t was created and ntroduced new tools to assess the tax burden: mplct tax rates and effectve tax rates or smulated tax rates In general, mplct tax rates ( backward lookng ) are used to analyze the ex-post corporate tax burden and tax dstrbuton effects But they do not take nto account the process of nvestment locaton choces or frm s margnal nvestments The calculaton of effectve tax rates ( forward lookng ) may allevate ths problem The ndcator used n ths artcle, s the statutory corporate tax rate whch s generally used n the context of nternatonal comparsons The standard rate s the sum of the hghest federal tax rate plus any more local taxes (eg regonal) Because of ts smplcty and avalablty, ths tax rate plays an mportant sgnal effect It s a determnng factor wth regard to the frm locaton and proft transfers among varous enttes of a multnatonal company based n several countres The MNF wll ndeed try to reduce declared benefts n countres where the nomnal tax rate s hgh and transfer them n countres where the nomnal tax rate s lower However, the statutory tax rate observed n varous countres gve only a partal vew of the actual tax pressure exerted on frms 4

6 To conclude, the statutory tax rate s a good tax assessment element of a country It s a very mportant element n nternatonal comparsons and whch nfluences consderably tax compettveness felt by frms Even f the felt tax compettveness dffers from the real compettveness, t s often the frst whch nfluences drect nvestment decsons and, more generally, the localzaton of moble factors of producton The nomnal tax rate crteron can act as a sgnal effect because the taxaton systems scrubland leads to dffcult decson makng process for frms We can also establsh comparsons between varous tax rates n Europe (table ) Thus, when we classfy countres accordng to varous ndcators of fscal pressure (statutory, EATR, EMTR), the rankng obtaned s relatvely dentcal to statutory tax rates A smlar classfcaton s obtaned f we take nto account effectve tax rates calculated by Devereux, Grffth and Klemm (00) Thus, n ths artcle we assume that nomnal tax rates are prncpal vectors of tax competton: ther reductons are analysed as sgnal effects and the study of these rates gves excellent nformaton of a government tax polcy and the tax burden whch t exerts on frms halshs , verson - Oct 008 Table : Country classfcaton wth respect to tax pressure level (005) Rankng Country Statutory tax rate EATR EMTR Germany Italy 5 6 France 3 3 Greece Malta Span 6 3 Netherlands Austra Belgum uxemburg Denmark Unted Kngdom 7 5 Fnland Czech Rep Sweden 5 Estona Portugal Slovena Poland 9 0 Slovaka 0 3 Hungary 9 5 Cyprus 5 4 atva 3 3 thuana Ireland 5 9 sources: OECD, European Commsson, EATR et EMTR are from ZEW EATR: Effectve average tax rate, EMTR: Effectve margnal tax rate 5

7 Statutory tax rates declnes n small countres Snce 990, the average corporate tax rate n the enlarged European Unon countres dd not cease fallng, from 389% n 990 to 58% n 006 (Fgure ) Fgure : Evoluton of Statutory corporate tax rate n EU5 halshs , verson - Oct ,0 40,0 38,0 38,0 36,0 36,0 34,0 34,0 3,0 3,0 30,0 30,0 8,0 8,0 6,0 6,0 sources:oecd, European Comsson servces 4,0 4, To explan the statutory tax rates fall n Europe, we can splt the EU n two sub-groups accordng to ther sze The partton was done accordng to two crtera that are geographcal area and populaton Thus we get a straghtforward fronter between countres: - arge countres: Germany, France, Great Brtan, Italy, Span and Poland - Small countres: Belgum, the Netherlands, uxembourg, Ireland, Portugal, Greece, Austra, Fnland, Denmark, Sweden, Slovena, Slovaka, Hungary, Czech Republc, Estona, thuana, atva, Malta and Cyprus Ths partton s nuanced when we take GDP as the choce crtera, especally for Poland, but ths anomaly can be explaned by the past weghng on economc structures of ths country Thus t s qute remarkable (fgure ) that small countres set corporate tax rate lower than those establshed n large ones In addton, t should be noted a stronger and earler tax rates decrease n small countres than n larges ones Ths tax strategy s characterzed by an upper pont s gap n 006 and has undergone a sgnfcant change over the last decade Fgure : Evoluton of Statutory tax rates n small and large countres 4,5 40,0 37,5 35,0 3,5 30,0 7,5 5,0,5 arge countres Small countres sources:oecd, European Comsson servces, own calculatons 4,5 40,0 37,5 35,0 3,5 30,0 7,5 5,0,5 6

8 Table : Corporate statutory tax rates arge countres 395% 383% 349% Small countres 36% 3% 34% Indeed, accordng to the table (), the tax rates gap between these two groups of countres was already on average of 7 ponts n 995, and reached more than ponts n 006 Accordng to table (3), small countres are at the orgn of the European tax competton Indeed these ones have ntated a very sgnfcant tax-cuttng process and sharply ntensfyng durng the twenty-frst century For example, small countres have reduced ther tax rates by more than 8% between 995 and 006, whle the declne n larges countres hardly establshed at % Ths declne has been partcularly strong snce the begnnng of 000 wth a loss of 77 ponts n corporate tax rates n small countres It appears that larger countres have been forced to respond by reducng ther rates also by 36 ponts halshs , verson - Oct 008 Table 3: Corporate tax gap arge countres - pts - 36 pts - 46 pts Small countres - 5 pts - 77 pts - 9 pts Map : Corporate statutory tax rate evoluton between 995 and 005 However, the map () shows the exstence of strong dspartes wthn the group of small countres but also n that of larges ones Ths tax rates declne was partcularly strong n Ireland, Czech Republc, Slovaka, Portugal, Estona, thuana and Cyprus Thus, the most strkng declne s attrbutable to Ireland (Fgure 3a), whose rate has experenced a vertgnous fall from 40% to % n just over 0 years, showng a very clear tax strategy on behalf Irsh governments However, ths declne s relatve because t has been accompaned by a tax base broadenng and the removal of MNF 7

9 specfc arrangements Thus, even f statutory tax rates have fallen, the mplct tax rates declne s less mportant Slovaka has also pursued a very aggressve corporate tax polcy (Fgure 3a), decreasng ts tax rate from 40% to 9% n just over 0 years Portugal and the Czech Republc have also changed ther tax rates n the same drecton, but wth a less pronounced declne (Fgure 3a) We can also refer to the sharp declne n tax rates n Estona (Fgure 3b) f we take nto account the taxaton of profts put to reserves (0%) nstead of dstrbuted profts (3%) We can ask whether n the long term corporate tax rates could be at zero Accordng to some economsts as Artus (006a), Mntz (99) and Wechenreder (005), ths hypothess can not be excluded halshs , verson - Oct 008 We can already make a very mportant remark: tax competton s more a small countres tax strategy rather than those of new entrants (0 CEEC) Indeed, we notced that t s not only new entrants who have lowered ther tax rates, but the majorty of small EU countres Proof of ths s the sharp declne n Ireland and Portugal tax rates (Fgure 3a) The proposal s corollary mples that the corporate tax rate wll not go up n new small entrants countres at the end of the convergence process unless ths strategy s neffectve and leads to budgetary problems Fgures 3 (a and b): Corporate tax rates evoluton n small countres Czech Republc Ireland Portugal Slovaka sources: OECD, European Comsson Servces Cyprus Estona thuana sources:oecd, European Comsson servces We can also observe dfferent strategc behavours wthn the group of larges countres In general, facng the noncooperatve tax behavour of smaller countres, the reacton of large countres was moderate Durng the last decade, the trend has been downward, more or less, n all large countres For example, France and, to a lesser extent, Italy and Great Brtan (Fgure 4) have experenced a slght decrease n tax rates, wth an ncrease of rates n 997 n France due to the change of government and n Italy n 005 due to encountered budgetary dffcultes Faced wth competton from smaller countres, Span has been able to mantan ts tax rate to 35% (Fgure 4) thanks to the good health of ts economy Nevertheless, there should be from 007, a declne of both the household ncome tax and the 8

10 corporate tax rate (expected to declne by 5 percentage ponts between 007 and 0) It was n Germany that the tax rate has most strongly fallen (Fgure 4), declnng from 566% n 995 to 389% n 00 Ths spectacular declne can be explaned by the fact that Germany had the hghest corporate tax rate n Europe Snce 00, Germany dd not lower ts rate, but t should not contnue snce a fall of ths rate s planned for 007 to 30% The other only large country havng sgnfcantly reduced ts tax rate s Poland (Fgure 4) ke Germany, Poland had a rather hgh tax rate (40%) n 995 (larger than France) Moreover, a very mportant crteron not to omt s the dstance to the European centre Indeed, Poland beng on the perphery, t wll tend to lower ts tax rate to offset agglomeraton forces and attracts frms Thus, we observe that small countres mpose a downward pressure on large countres whch are constraned to respond by reducng n turn, ther tax rates on moble bases Fgure 4: Corporate tax rates evoluton n large countres halshs , verson - Oct France Germany Italy Span Poland Great Bretan sources: OECD, European Comsson servces The theoretcal tax rate evoluton n all EU countres durng the last decade lead to a corporate tax rate gap n favour of small countres (map ) Map : Corporate tax rates levels n 006 9

11 3 Tax competton Model 3 Smple tax competton model We use a smple tax competton model to obtan natonal governments reacton functons Reacton functons developed thereafter, based on Bruckner s theoretcal model (999, 003), wll treat only the competton on corporate tax rate, and not ncome, consumpton or property tax rates, because t seems that tax competton prmarly affects frms Thus, we wll establsh two types of reacton functons accordng to the game taken nto account: - A Nash equlbrum where all countres set ther tax rates smultaneously; - A Stackelberg equlbrum where a country or a group of countres set ther tax rates frst and other EU countres adopt a Follower behavor halshs , verson - Oct 008 For smplcty, we are buldng a model of an economc area wth two countres, generalzable wth n countres, where captal (K ) nvested n the country, =, s perfectly moble and labour ( ) s consdered as a fxed producton factor A prvate consumpton good X s produced combnng producton factors accordng to a constant returns to scale technology Ths producton functon s gven by F(K, ), wrtten f(k ), where k s the number of captal per worker unts For smplcty, we assume the number of workers n each country s equal to ts populaton The economy s composed of two consumer goods, a prvate good X and a publc good G A tax s leved on nvestment n each country, wth t the tax rate per unt of captal n the country The frm proft functon n the country s defned as follows: Π = f ( k ) r k t k (3) Wth r, the common nterest rate n the area The proft maxmzaton s equvalent to equalze the margnal productvty to the nterest rate plus the tax rate: f '( k ) = r (3) t The captal net tax rate of return s equal to the margnal productvty less taxes As captal s perfectly moble, net returns should be equal n both countres: f '( k ) t = f '( k t = r (33) ) Where r s the unform net return Then, we note K the captal amount avalable n the area, thus obtanng the followng condton: K = K (34) K 0

12 Countres captal endowments are equal and are equally dstrbuted among workers, leadng to ndvdual endowments gven by k* = K / ( ) For smplcty, we specfy an dentcal frm producton functon among countres: f α k ² k) = β k where β, α > 0 (35) ( Thus, equatons (33) and (34) determne k, k, and r accordng to tax rates t and t halshs , verson - Oct 008 t t k = k * α ( ) t t k = k * α ( ) It s clear that the captal elastcty to tax rate s negatve, k / t (36) (36 ) = < 0 Ceters parbus, a tax rate α( ) ncrease n country causes captal flght towards another country to equalze the net return The rate of return dervatve to the tax rate, r / t, s also negatve, ndcatng the net return s reduced by hgher tax rates We defne a utlty functon for resdents, U, havng homogeneous preferences n each country Ths utlty functon U (X ; G ) s a trade-off between consumpton of a prvate good X and a publc good G Consumers ncomes (R ) are composed of labour ncomes (wages), w = f k ) k f '( k ) and captal endowments returns, rk* The country resdents budget constrant s: ( x = w rk * (37) We assume the publc good s fully fnanced by tax on captal We thus obtan the government fscal constrant: g = t k (38) Substtutng w n (37), and usng (38), we can rewrte the utlty functon n the followng form: [ f ( k ) k f '( k ) r k*; t k ] U (39) Resdents of country choose t to maxmze (39), takng nto account tax rate effects on k and r In addton, the tax rate s regarded as a constant Wthout loss of generalty, we consder the utlty functon U as addtve (lnear preferences):

13 g x g x U = ) ; ( (30) Now the analytcal framework s presented, we must consder dfferent types of games that countres wll partcpate 3 Nash Equlbrum Each government seeks to maxmze the representatve agent utlty functon under both publc and prvate budget constrant Thus, we maxmze (30) subject to constrants (37) and (38) k t k r k f k k f MaxU * ) '( ) ( τ = (3) Nash equlbrum tax rates t and t are located at the ntersecton of countres and reacton functons Substtutng (36) nto (3), the equaton can be solved n order to obtan the reacton functon of country : * ) ( ² ) ( ² k t t λ = (3) Where: = ) ( ² ) ( ² )² ( α λ In the same way, we obtan the country reacton functon: * ) ( ² ) ( ² k t t λ = (3 ) Where: = ) ( ² ) ( ² )² ( α λ Rearrangng (3) and (3), we obtan * / / k t s s t λ = (33) * / / k t s s t λ = (33 ) Where ( ) s = and ( ) s = are each populaton shares n the total area populaton halshs , verson - Oct 008

14 s Second order condtons to get a maxmum n the optmzaton program requres ( / s ) < 0 or > Thus, equaton s (3) denomnator s negatve Therefore, the reacton functon slope, equal to s ( / s) ( / s), s negatve when < < s and postve when > s In the same way for the second country, n order to maxmze ts optmzaton program, we need to have s > and condtons on the s sgn of the reacton functon slope are symmetrc, that s, negatve when < < s and postve f > s Ths theoretcal result ndcates that country wll decrease (ncrease) ts tax rate n response to an ncrease n country when the margnal publc good utlty s small (large) The ntersecton of the two reacton functons (33) and (33) gves the Nash equlbrum Equatons (33) and (33 ) show that reacton functons both depend on country preferences and ts sze Occultng halshs , verson - Oct 008 the case where = s, the slope s dfferent from zero Thus, wth econometrc estmates, the emprcal test to verfy the strategc nteractons exstence wll consst n checkng the get parameter sgnfcance Otherwse, ths would tend to prove the absence of strategc nteracton between European countres Suppose now both countres are of equal sze, e that =, we obtan followng reacton functons: 4α ( ) t = t k * 4 4 4α ( ) t = t k * 4 4 (34) (34 ) Thus, f both countres have same preferences, = =, both countres wll react n the same proporton to changes n tax rates n the neghbourng country Ths two countres model s also vald for an N countres economc area We emprcally test ths model n the case of the European Unon where we strongly suspect tax strategc behavours The confrmaton of ths theory wll consst n testng the sgnfcance of the reacton functon slope n each country If ths test s sgnfcant, t wll clearly ndcate the presence of strategc behavours wthn the European Unon 33 Stackelberg leader extenson In contrast to the theoretcal Nash framework where all countres are consdered on an equal footng, the Stackelberg model wll afford to hghlght the presence of leader(s) element(s) n the tax competton Thus, we derve the model n a leader / follower framework, where the leader country set ts tax rate frst wthout takng nto 3

15 account those n other countres Then other follower countres wll set ther own takng nto account the tax rate set n the frst country For smplcty, we are always thnkng n the context of a two countres model where we wll adopt a two stages sequental approach (backward nducton): - Step : We determne the follower country s reacton functon - Step : We determne the leader country s reacton functon Step : Country, the follower, chose ts tax rate t dependng on the leadng country tax rate t Country chooses ts rate t to maxmze ts utlty functon U (X ; G ) Ths makes t possble to characterze the calculated prevously best reacton functon gven by (33) t follower / s leader = t λ k * / s (35) halshs , verson - Oct 008 Step : Country, the leader, chooses t to maxmze ts utlty functon, takng nto account the country best reacton functon (BR) n ts utlty functon maxmzng program, we get: Wth t leader [( AS = S ( S) A = > ( S ) )( λs α ) α [S ( A) S ²( A)²] ) α ] k * (35 ) The leader country sets ts tax rate to maxmze ts nhabtant s welfare and follower countres act as n the case descrbed prevously n the Nash equlbrum From these two theoretcal results (tax rates fxed n a Nash and Stackelberg framework), we wll seek to emprcally verfy the earler advanced argument where small EU countres practse tax competton 4 Emprcal methodology 4 Econometrc specfcaton The emprcal estmaton underlyng theory s a strategc tax competton model (Brueckner 999, 003) n a game where countres set ther corporate tax rates smultaneously (Nash reacton functon, Wldasn, 988) or followng the leader country s tax rate (Stackelberg reacton functon) Emprcally, we wll estmate reacton functons n both cases for EU countres We are n lne wth the few abundant lterature dealng wth estmaton of natonal reacton functons (Devereux, ockwood and Redoano, 00, Redoano, 003, Altshuler and Goodspeed, 00, Ruz, 006) 4

16 We econometrcally estmate the two followng equatons:, = X ϕd ηtt ε (4) τ t δ β ωjτ j, t θ, t j τ, t = δ β ωjτ j, t θx, t ατ, t ϕd ηtt ε (4) j Where represent country, t, the tme, τ s the home country tax rate, τ j,t, are neghbourng countres tax rates, τ,t- s the lagged leader tax rate, X s an exogenous vector of domestc soco-economc control varables, ω s a weghtng matrx dscussed later and α, β, θ, η, ϕ are parameters to estmate, d beng potental ndvdual effects and T, an halshs , verson - Oct 008 eventually trend necessary for the estmate, and ε,, the error term We detal the exogenous control varables n a later part Thus, slopes of Nash (4) and Stackelberg (4) reacton functon are respectvely β and α Accordng to the theoretcal model, these slopes can be negatves because of the desre to mantan publc expendture, but we expect that these slopes are rather postves because of the hgh captal elastcty to tax rates 4 Econometrc estmaton We chose to use panel data to carry out ths emprcal study Most of studes on tax competton were realzed wth cross secton data The panel method has several advantages; t enables to take nto account the tax competton dynamcs between countres wthout elmnatng effects related to cross secton nformaton In addton, the panel reduces the rsk of colnearty among explanatory varables, snce these varables are expressed n two dmensons Fnally, ths approach makes t possble to cure the presence of tme nvarant omtted varables (unobservable ndvdual correlated wth some of explanatory varables) Accordng to the homogenety tests procedure presented by Hsao (986) for the followng regresson: Y,t = α β X,t ε,t The consdered panel s then a panel wth ndvdual effects: the 6 vectors β can be consdered as dentcal between countres whle constants α dffer among the 6 countres Parameters α are consdered as determnstc constants specfc to each 6 countres These parameters can correspond to omtted structural varables that may explan tax rates choce n dfferent countres The fxed effects model provdes convergng estmates, even when there s a correlaton between explanatory varables and unobservable characterstcs But when estmatng (4) and (4), we were confronted wth several problems often present n spatal econometrcs (Anseln, 988): - Endogenety of τ,t - The possble spatal dependence of error terms 5

17 4 Varables endogenety The presence of strategc nteractons mples that tax rates, τ,t, of dfferent countres are jontly determned Indeed, the theoretcal model mposes a jont determnaton of tax rates n Nash equlbrum, and ths smultanety nduces that the explanatory varable s endogenous what makes the use of OS nconsstent There are three methods to solve ths endogenety ssue: - Estmate (4) wth the method of Maxmum kelhood (M) - Usng the method of Instrumental Varables or the Generalzed Method of Moments (Anseln 988) - Ths endogenety problem can be avoded by assumng that nteractons take place wth one or more delay In ths case, τ j,t would be replaced by τ j,t-n n (4) where the number of lags wll be determned halshs , verson - Oct 008 We chose to estmate the equaton (4) wth the Generalzed Method of Moments Ths method s also consstent wth the second potental econometrc problem that may be the errors spatal dependence 4 Spatal dependence of errors The spatal autocorrelaton s the correlaton of a varable wth tself (autocorrelaton) attrbutable to the data geographc organzaton (spatal) (Grffth, 99) To check the absence of errors autocorrelaton, t s mportant to carry out ths test n order to elmnate fallacous estmaton rsks: - Rsk of havng estmated standard devatons bas, estmators remanng however unbased - Rsk of common shock presence producng a correlaton between tax rates of a country and those of neghbours and that ths correlaton wrongly leads to conclude that neghbourng rates affect the rate of a gven country, whereas actually t s not the case Ths s a well-known rsk n spatal econometrcs (Anseln 988) The Moran statstcal test s the most used n the detecton of spatal autocorrelaton (see Anseln & al, 996; Anseln and Kelejan, 997) Thus, the Moran statstcs calculaton allows to determne for each country f the tax rate s fxed randomly compared to the geographcal proxmty (n ths case, the statstc s not sgnfcant), and therefore f there s a spatal autocorrelaton (postve or negatve) wth the settng of tax rates between countres Ths confrms the exstence of postve spatal autocorrelaton between seres To overcome these econometrc ssues, two estmaton methods are possble The frst conssts n regressng the equaton (4) wth the method of Maxmum kelhood takng nto account the error structure descrbed n (43) Ths method has been explored by Case et al (993) Another approach s to use the Instrumental Varables method The estmaton method chosen was that of Generalzed Moments (GMM) Kelejan and Prucha (998) have shown that wth the presence of spatal dependence of errors, nstrumental varables method, and therefore the Method of Moments provde a robust estmate of β Wth several exogenous structural varables t seemed more conducve to use ths method 6

18 We must now take nto account the neghbourng countres nfluence of on the domestc country by buldng weghtng matrces 43 Weghtng Matrces In realty, a country has generally more than one neghbour, t s mportant to order the nfluence of ts neghbours on the domestc country by buldng weghtng matrces These weghtngs wll be used to measure the mpact of a neghbour on a country by assgnng t a value Thus, we wll evaluate the nfluence for each par of states Weghts of the weghtng matrx ω are constructed to take nto account the nfluence of tax rates of each country consdered as a neghbour on the home country s tax rate In order to dentfy all possble nfluences, we opted for the constructon of 3 weghtng matrces: halshs , verson - Oct Geographc weght : o Dstance matrces: d Dstance between countres : ω j = d d Where ω d j are the j elements of the weghtng matrx W d and d j s the geographcal dstance between country and country j Unlke most studes, we do not take the dstance between captals of each country, but the weghted dstance calculated by Mayer and Zgnago (CEPII), consderng several ctes n the country n order to take nto account the geographc dstrbuton of populaton wthn the country d Dstance to the centre (Belgum-Brussels) : ω c = d d As wth the prevous matrx, the dstance from the centre s weghted by the geographcal dstrbuton wthn each country j j j c j c - Economc weght : gdp o GDP : ω jt = ( GDP GDP ) ( GDP GDP ) t jt jt t jt The dstance weghtng ω jd, where geographcally closest countres have a larger mpact, seems to be the most obvous However, wth the New Geographc Economy contrbuton, t seemed mportant to buld a matrx of weghts from the centre, supposed to be Brussels n our study, to take nto account agglomeraton forces n the tax rate settng process Indeed, more a country s close to the centre, less t wll be nfluenced by tax choces of perpheral countres Ths does not call nto queston our ntal hypothess that small countres are tax competton nstgators, because those are mostly at the European Unon perphery 7

19 We also add a noton of economc neghbour to take nto account the sze of countres and ther possble smlar behavours Indeed, large countres havng much heaver economc structures than smaller countres, t seems clear that they do not have the same capactes of reacton, ether n tme and magntude of changes Thus, we consder a GDP weghtng matrx It should be noted that ths matrx may vary over tme We note that each matrx s normalzed to, thereby encouragng comparsons and ndex constructon such as that of Moran used to test the autocorrelaton 43 Data and varables halshs , verson - Oct 008 We estmate models (4) and (4) usng tme seres data of 5 European Unon countres, before the ntegraton of Bulgara and Romana n January 007, plus Iceland over the perod We have delberately chosen to begn the perod of study n 995 as the majorty of countres of the former Warsaw Pact and the Sovet Unon have started ther converson towards a market economy only a few years after ther emancpaton The explaned varable, τ j,t, s represented by top statutory tax rates These data are derved from the OECD and the European Commsson database Ths choce reflects the mportance of ths rate n the tax pressure evaluaton by frms prevalng n a country Moreover, t represents more than 40% of ncome from taxaton (Redoano, 003) We concentrate only on the corporate tax rate because t seems t s the prncpal tool of tax competton between states; tax competton on hgh ncomes beng more margnal, and unproven (Devereux, ockwood, Redoano, 004) We use statutory tax rates rather than effectve (smulated) tax rates because frst ones appear to be more relevant as sgnal effect to the frms attenton Indeed, Devereux, ockwood, Redoano (004) have shown that the effectve margnal corporate tax rate (EMTR) s not relevant n order to reflect strategc nteractons Ths s n lne wth our expectatons, because ths rate s more used to note the after tax return of a margnal nvestment Geographc weghtngs are carred out from dyadc data calculated by Mayer and Zgnago (CEPII) We use weghted dstances to take account of the populaton geographcal dstrbuton wthn the country Economc weghts are computed from GDP data avalable on the DRI database Fnally, nstrumental varables and control varables take nto account for each country: - The share of publc expendture n GDP, - The defct, - The domestc demand - The drect nvestment as a % of GDP - The unemployment rate, - GDP, - The GDP per capta, - The share of over 60 years n the populaton The presence of these varables s warranted to reflect the structural constrants weght present n natonal economes Indeed, a relatvely large country wth mportant publc expendture wll be nclned to respond less quckly and less ntensely to lower tax rates n other countres Moreover, the presence of varables such as GDP, GDP per capta and the defct as a % of GDP makes t possble to take nto account postve and negatve shocks 8

20 on natonal economy whch would affect consequently the tax rate level n the country These data are derved from the Datastream database For our estmates, we have verfed that all these seres were statonary, whether they are dependent or explanatory varables When that dd not happen, we make them statonary to make sure of our estmates accuracy 5 Results halshs , verson - Oct 008 In ths part, we test countres reacton delays to other European countres tax polces We demonstrate that there s a group of leaders n the tax competton but also succeed n characterzng ths group of leadng countres The frst step s to demonstrate that small countres, taken as a whole, are tax competton leaders Then we establsh a dstncton accordng to the geographcal locaton of these countres wthn Europe So we separate the group of small countres nto two groups: ntegrated (located n the centre of Europe) and dstant (located at the perphery) It appears that only the ntegrated group plays a sgnfcant role n corporate tax rate settng of larges countres astly, the thrd stage s to determne whether a small country, taken ndvdually, plays a leadng role n the tax competton 5 Exstence of strategc behavours n the EU Whle regressng each country tax rate on tax rates of other countres n the European Unon, we found the presence of strategc nteractons wthn the European Unon (Chatelas and Peyrat 008) Table (4) also shows that countres observe ther neghbours fscal polces and respond the followng year Indeed, takng account of a tax reform n a neghbourng country, the home country wll develop a new tax schedule by changng ts rates the followng year n order to stop captal flght We note that the domestc country's reacton s more than proportonal compared to the tax rates declne n countres wth a common border (estmated coeffcents are respectvely 35 and 84 for dstance to the centre and blateral dstances matrces) Ths can be explaned by the fact that the country reacts to tax rates falls n other countres but t also antcpates ts neghbour s reacton to ts own lower tax rates and / or tres to attract taxable ncomes Countres react more strongly to ther closest geographcal neghbours tax rate snce a tax rates fall can jeopardze access to the market known as home market bas Indeed, a frm may locate n the neghbourng country, takng advantage of a more attractve tax system, whle havng access to the domestc market at a lower cost The dstance to the centre crteron also appears to be mportant n the decsonmakng process and n reacton Countres, ncludng small dstant ones, wll be attentve wth core countres fscal polces and wll tend to react by lowerng ther tax rates more sharply n order to compensate for agglomeraton forces present n the centre However, for estmates made wth the GDP weghtng matrx, we found smlar coeffcents wth one lag (00) or none (08) Fnally, we can note that estmates wth two lags of delays are not sgnfcant Ths suggests that countres react quckly, usually n the followng year Ths delay s an average, some countres wll react faster, theoretcally small ones, whle others wll take longer to respond, probably larger ones 9

21 Table 4: Reacton functon estmates wth and wthout lags for all countres of UE for Dependent varable : dτ (corporate tax rate) GMM wth ndvduals fxed effects Coeffcents Estmates Explanatory Varables: Wcentre Wdstance Wgdp dτj 0,97** 0,95*,08*** (,5) (,85) (3,9) dτj(),35**,84*,00** (,05) (,85) (,) R² 0,6 [0,] 0,3 [0,08] 0,5 [0,4] Sum squared resd 889 [940] 9 [097] 899 [909] DW,9 [,0],0 [,6],03 [,00] J-stat 7,37 [,46] 3,55 [4,84] 7,7 [8,97] t-stat n bracket * statstc sgnfcance level at 0%, ** à 5% et *** à % Regressons statstcs between [ ] concern estmatons wth one lag halshs , verson - Oct 008 Ths frst emprcal part of the study allows establshng the presence of strategc behavours wthn countres of the European Unon In addton, t appears that countres wll react more to ther neghbours tax polces after one year Accordng to the theory developed n the precedng parts and ntal estmates, we can assume that there are leadng countres n the tax competton and that they are small countres We thus now try to detect the nfluence of small European countres on tax rates set n large countres We carry out estmates for the group of large and small countres whch enable to hghlght the greatest reactvty of small countres compared to large ones In other words, we sought to determne whether small countres mplement aggressve tax strategy vs-à-vs bg countres 5 Small countres: leaders of the tax competton n Europe Accordng to the theory developed prevously, we expect that small countres of the European Unon apply aggressve lowerng tax rate strategy vs-à-vs larges countres n order to attract taxable ncomes The fact that countres wat to note others tax polcy before makng ther own reforms suggest the exstence of leaders n the fallng tax rates process Accordng to table (5), Chatelas and Peyrat (008) fnd that tax nteractons are stronger between small countres than n all European countres or among large countres These nteractons are weak or even non-exstent wthn the group of larges countres (see Table 5) The presence of strategc nteractons beng establshed, we show that small European countres are at the orgn of ths tax competton We contnue wth panel estmates A group of 8 small countres wll play the Stackelberg leader role Ths group s defned as follows: Belgum, Denmark, Czech Republc, Ireland, Portugal, Slovaka, thuana and atva The average tax rate of ths group was calculated usng three weghts: the dstance from the centre, the weght of GDP and populaton 0

22 Table 5: Reacton functon estmates for small and large EU countres over Dependent varable : dτ (corporate tax rate) GMM wth ndvduals fxed effects Coeffcents Estmates Explanatory Varables: Wcentre Wdstance Wgdp dτj small countres,05**,07**,3*** (,04) (,07) (4,0) dτj large countres 0,70* 0,78* 0,5 (,78) (,87) (0,79) R² 0, [0,05] 0, [0,03] 0,3 [0,5] Sum squared resd 708 [90] 875 [67] 703 [4] DW,89 [,08],00 [,04],05 [,90] J-stat 5,79 [0,95],7 [,97] 4,38 [4,59] t-stat n bracket * statstc sgnfcance level at 0%, ** à 5% et *** à % Regressons statstcs between [ ] concern estmatons wth for the large countres group (Germany, France, Unted Kngdom, Italy, Span and Poland) halshs , verson - Oct 008 Results of the table (6) show estmates of the followng equaton conducted wth the three weghtng matrces (dstance from the centre, dstance and weght of blateral GDP): τ,t = δ β ω j τ j,t θx,t ατ,t ε (4) j Table 6: Reacton functon estmates for large EU countres over wth all small countres as leaders Dependent varable : dτ (corporate tax rate) GMM wth ndvduals fxed effects Coeffcents Estmates Explanatory Varables: Wcentre Wdstance Wgdp dwτ 0,48 0,06 0,87 (,) (0,4) (,3) Dleader(),8**,76**,48** (,95) (,6) (,96) Constant,3*,39*,57 (,7) (,80) (,65) R² 0,07 0,09 0,03 Sum squared resd DW,08,,06 J-stat,6,9,08 t-stat n bracket * statstc sgnfcance level at 0%, ** à 5% et *** à % Estmates wth dstance from the centre weght and blateral dstances weght produce conclusve results (unlke those obtaned wth the GDP weght, estmated coeffcents are sgnfcantly dfferent from zero) The detaled results below were acheved wth weghtng by the dstance from the centre Most of large countres beng n the European Unon core, t appears that they are more nfluenced by the change n tax rates of small countres close to the centre

23 However, t should be noted that there s no strategc behavour among major countres when we add the group of small countres leaders of Stackelberg n the regresson Ths may ndcate that the pressure that small countres mpose on large ones s strong enough to that competton between major appears neglgble (and s not revealed n econometrc regressons) In concluson, we show that small countres mpose a (downward) pressure on large countres corporate tax rates It would be nterestng to make a dstncton between small countres to see those havng the most nfluence on larges ones Indeed, we should remember that we weghted the tax rate of the eght small countres group by ther dstance from the centre (small countres close to the core and thus close to large ones have the most weght) We assume that small countres close to the centre and those at the perphery do not have the same mpact on tax rates settng n bg countres 53 Small countres close to the centre are the real tax competton leaders halshs , verson - Oct 008 The mportance of the small countres group n tax rate settng has been demonstrated We refne ths analyss by dstngushng small countres close to the centre and those at the perphery Indeed, small countres are very numerous (0 small aganst 6 bg ) n the European Unon, we make a dstncton wthn ths category Our assumpton s that small countres close to the centre have the most nfluence on core major countres and consequently they are the real nstgators of ths tax competton Bg countres always adopt a poston of follower wth respect to these small countres ntegrated to the centre rather than vs-à-vs small perpheral countres We splt our sample of small countres nto two groups: ntegrated ones close to the Economc heart and dstant ones - ntegrated : Ireland, Belgum, the Netherlands, uxembourg, Denmark, Austra, Slovena and Czech Republc - dstant : Sweden, Fnland, Estona, atva, thuana, Hungary, Slovaka, Greece, Portugal, Cyprus, Malta, and Iceland Ths separaton was done only on geographcal crtera, and the absence of Scandnavan countres n the ntegrated category s justfed The 6 bg remanng countres (Germany, France, Great Brtan, Span, Italy and Poland) are the thrd group of countres The so-called ntegrated countres Table (7) summarzes estmates made wth the group of small ntegrated countres Small countres have ndeed a central nfluence on tax rate establshed n large countres (for both geographcal weghtng matrces) Bg countres always react wth two years late and always strongly wth a coeffcent greater than Small countres ntegrated to the centre plays a leadershp role vs-à-vs bg countres n tax rate settng It should be noted n ths case, there s the presence of strategc nteractons among major countres However, rates n large neghbourng countres are less mportant than rates set n small ntegrated countres But remember that large countres such as France and Span have not yet done so ther tax reform n 006, whch are planned for comng months

24 Table 7: Reacton functon estmates for large EU countres over wth small ntegrated countres as leaders Dependent varable : dτ (corporate tax rate) GMM wth ndvduals fxed effects Coeffcents Estmates Explanatory Varables: Wcentre Wdstance Wgdp dwt 0,90***,00**,0** (,54) (,65) (,4) Integrated eader (),57*,7*,4 (,8) (,8) (,50) Constant,34,0,8 (,54) (,55) (,3) R² 0,5 0, 0,09 Sum squared resd DW,97,98,0 J-stat,3,78,8 halshs , verson - Oct 008 t-stat n bracket * statstc sgnfcance level at 0%, ** à 5% et *** à % Thus small ntegrated countres are leaders n the bg countres corporate tax rate settng Now let us consder f we can demonstrate whether the same was true for small countres located at the perphery Small perpheral countres How large countres do react n dealng wth changes n tax polces of small perpheral countres? Table (8) shows there s a weak nfluence of small dstant countres or even a lack of nfluence f we refer to the dstances matrx Estmated coeffcents are 053 for weghtng by the dstance from the centre and 064 for that by the GDP Table8: Reacton functon estmates for large EU countres over wth small perpheral countres as leaders Dependent varable : dτ (corporate tax rate) GMM wth ndvduals fxed effects Coeffcents Estmates Explanatory Varables: Wcentre Wdstance Wgdp dwt -0,08 0,48 0,68 (-0,) (,4) (,38) Perphery leader() 0,53* 0,8 0,64* (,75) (0,6) (,7) Constant 0,3-0,05 0,87 (0,5) (-0,0) (,0) R² 0,6 0,8 0,4 Sum squared resd DW,9,,5 J-stat 0,79,64 5,80 t-stat n bracket * statstc sgnfcance level at 0%, ** à 5% et *** à % These coeffcents are sgnfcantly lower than those found for the group of small countres taken as a whole or for the group of small ntegrated countres We deduce from t that small perpheral European countres have very lttle nfluence on the tax rate settng n large countres 3