Tax Setting in a Federal System: The Case of Personal Income Taxation in Canada

Transcription

1 Document de treball 2001/9: Tax Settng n a Federal System: The Case of Personal Income Taxaton n Canada Aleandro Esteller, Albert Solé Insttut d'economa de Barcelona Edfc Florensa Adolf Florensa, s/n Barcelona Tel.: Fax: E-mal: eb@pcb.ub.es pcb.ub.es/eb

2 TAX SETTING IN A FEDERAL SYSTEM: THE CASE OF PERSONAL INCOME TAXATION IN CANADA a,b Álex Esteller, Albert Solé c ABSTRACT: In a decentralsed tax system, the effects of tax polces enacted by one government are not confned to ts own ursdcton. Frst, f the tax base s moble, tax rate ncreases by one regonal government wll rase the amount of taxes collected by other regonal governments (horzontal tax externalty). Second, f both the regonal and the federal levels of government co-occupy the same felds of taxaton, tax rate ncreases by one layer of government wll reduce taxes collected by the other (vertcal tax externalty). Thrd, as Smart (1998) shows, f equalsaton transfers are present, an ncrease n the standard equalsaton tax rate provdes ncentves to rase taxes to the recevng provnces. In order to check the emprcal relevance of these hypotheses, ths paper estmates provncal tax settng functons wth data on Canadan personal ncome taxaton for the perod We fnd a sgnfcant postve response of provncal tax rates to changes n the federal ncome tax rate, the tax rates of competng provnces, and the standard equalsaton rate (only for recevng provnces). We also fnd that the reacton to horzontal competton s stronger n the provnces that do not receve equalsaton transfers. Keywords: fscal federalsm, tax competton, equalsaton grants, personal ncome tax JEL Classfcaton: H3, H21, H77 RESUMEN: En un sstema fscal descentralzado, los efectos de las polítcas mpostvas llevadas a cabo por un goberno no se agotan en su propa ursdccón. En prmer lugar, s la base mpostva es móvl, ncrementos mpostvos por un goberno regonal ncrementarán los recursos de otro goberno regonal (externaldad fscal horzontal). En segundo lugar, s los nveles de goberno regonal y federal comparten los msmos espacos fscales, ncrementos en el tpo mpostvo de un nvel de goberno reducrá la recaudacón del otro (externaldad fscal vertcal). En tercer lugar, como Smart (1998) ha demostrado, bao un sstema de nvelacón, un ncremento en el tpo mpostvo estándar de la transferenca ncentva aumentos en los tpos mpostvos de las provncas receptoras. Al obeto de comprobar la relevanca de estas hpótess, estmamos funcones provncales de facón de tpos mpostvos a partr del mpuesto canadense sobre la renta personal para el período Hallamos una sgnfcatva respuesta postva de los tpos mpostvos provncales ante cambos en los tpos mpostvos federales, los tpos mpostvos de las provncas competdoras y el tpo mpostvo estándar de la nvelacón (para las provncas receptoras), así como que la reaccón a la competenca fscal horzontal es mayor para aquellas provncas que no recban subvencones de nvelacón. Palabras clave: federalsmo fscal, competenca fscal, subvencones de nvelacón, mpuesto sobre la renta personal a Comments are welcome. The opnons expressed n the paper do not necessarly reflect the IEB's opnons. b An earler verson of ths paper was presented at the Conference of the Royal Economc Assocaton, St. Andrews, Unted Kngdom, Aprl 2000, at the 2 nd Summer Workshop on Appled Envronmental and Publc Economcs, ZEW, Mannhem, Germany, June 2000, and at the Conference of the Internatonal Insttute of Publc Fnance n Sevlla, Span, August Helpful comments by Thess Büettner, Bev Dalhby, Robert Inman and Stuart Landon are gratefully acknowledged. We have benefted from the fnancal support provded by the CICYT SEC , Mnstero de Educacón y Cultura, and as Consoldated Research Group n Fscal Federalsm and Regonal Economcs, 1999SGR00017 c Correspondng author: Albert Solé (asole@eco.ub.es) Dpto. Hacenda Públca- Facultad de Cencas Económcas (Unversdad de Barcelona) Avda. Dagonal, 690, Torre 4, pl. 2ª Barcelona (Span) Tel.: Fax:

3 1. Introducton The benefts of fscal decentralsaton are well known: manly, the greater senstvty to dverse regonal preferences (Oates, 1972) and the possblty of preference revelaton by household moblty (Tebout, 1956) 1. Nevertheless, aganst these benefts, the assgnment of fscal power to the decentralsed layers of government mght create neffcences that could more than compensate such benefts. A typology of these neffcences can be found n Gordon (1983) and, more recently, n Inman and Rubnfeld (1996), Dalhby (1996) and Wlson (1999). Many studes, both theoretcal and emprcal, have dealt wth specfc ssues: tax competton (e.g., Zodrow and Meskowsk, 1986), tax exportng (e.g., Brd and Slack, 1983), expendture spllovers (e.g., Case et al., 1993) and vertcal externaltes, both on the tax sde (e.g., Keen, 1998) and on the expendture sde (e.g., Dalhby and Wlson, 1999). In ths paper we concentrate on nteractons that may arse because of the decentralsaton of tax power to sub-natonal governments. In a decentralsed tax system, the effects of tax polces enacted by one government are not confned to ts ursdcton. Tax rate ncreases by one regonal government wll affect taxes collected by other regonal governments as well as by the federal government. Horzontal and vertcal tax externaltes are the man sorts of fscal externaltes among governments ponted out n the lterature and wll be the focus of our emprcal analyss. In addton, the federal nsttutonal framework of each country could allevate or exacerbate the effects of these externaltes. There are many dfferent examples of such nsttutons, but as our emprcal analyss uses data on Canadan taxes, we wll focus on the ncentve effects that the Canadan equalsaton program has on provncal tax polces. Let s have a more detaled look at these problems. Frst, vertcal tax externaltes arse as a consequence of the co-occupancy of the same feld of taxaton by both the regonal and the federal layers of government (Flowers, 1988; Boadway and Keen, 1996; Dahlby, 1996; or Keen, 1998). The 1 Some authors have ponted that decentralsaton may also help to ncrease accountablty and, thus, mprove preference revelaton at the ballot box; e.g., Salmon (1987) and Seabrght (1996). 2

4 necessary condtons for the exstence of the externalty s that both layers of government have tax power over the same dstortonay tax base 2, and that they both neglect the mpact of ther actons on the amount of revenues collected by the other. Then, each layer of government wll establsh an neffcently hgh level of taxaton wth respect to the level that would have been establshed by a untary government. Two man emprcal predctons derve from these models: () combned tax rates wll be hgher n federatons where dfferent levels of government share the same tax bases, and () there wll be nterdependences among the tax rates set by the federal and regonal governments. Recently, some papers have looked for emprcal evdence on vertcal tax externaltes usng the second approach: Besley and Rosen (1998) for excse taxes n the US, Goodspeed (2000) for personal ncome taxes n a panel of OECD countres, Hayash and Boadway (2000) for corporate ncome taxes n Canada, and Esteller and Solé (2001) for personal ncome taxes n the US. In all the cases, the authors are able to fnd a tax nteracton between the lower layers of government and the federal one, though the sgn of the reactons are dfferent. However, ths last result s not surprsng, snce the theoretcal lterature has shown that the expected sgn of the nteracton s ambguous (Keen, 1998). Second, horzontal tax externaltes are probably the most well documented source of neffcency n a federal system of government. In partcular, we are nterested n the neffcency that comes out because of the moblty of the tax base among provncal governments. Under ths stuaton, all provncal governments compete to attract tax base to ther terrtory. Ths can only be acheved by lowerng the tax burden on the moble factor, whch wll provoke an neffcently low level of publc good provson n all the federaton (Zodrow and Meszkowsky, 1986; Wlson, 1986; Boadway and Wldasn, 1984). Two man emprcal predctons derve from these models: () ncreased tax competton wll cause tax rates to converge at lower levels, and () tax rates of competng regonal governments wll be nterdependent. Most of the papers have searched for evdence on tax competton by testng for horzontal tax nterdependences (see, e.g., Ladd, 1993, Case, 1993, Besley and Case, 2 It has to be stressed that tax-sharng arrangements.e., when the regonal government s enttled to a share of the revenues rased n ts terrtoral ursdcton do not conform wth what we mean by concurrent taxaton ; what s n fact necessary s that both levels of government share tax powers regardng that tax base. 3

5 1995, Brett and Pnske, 2000, Heyndels and Vuchelen, 1998, and Büttner, 2000) 3. All of them have found evdence of postve nteractons among tax rates of competng regons. Thrd, the nsttutonal framework can also affect the tax settng process of regonal governments. For nstance, tax settng decsons may be affected by the deductblty of the regonal tax lablty from the federal tax base (as n the US), or by the presence of a system of equalsaton grants (as n Canada). In the frst case, the tax deductblty lowers the regonal margnal cost of taxaton, whch tends to rase regonal tax rates (see Feldsten and Metcalf, 1987, for an emprcal applcaton). In the second case, the equalsaton grant also reduces the margnal cost of publc funds, and so tends to rase regonal tax rates as well (Smart, 1998). Ths means that the tax rate wll be hgher n the recevng provnces under the equalsaton program that under an equal-yeld lump-sum transfer. We wll account for ths n the emprcal analyss by lookng for the effect of the equalsaton status on the level of tax rates. Thus, the am of ths study s to test the emprcal relevance of each one of the three cted sources of tax nteracton: horzontal and vertcal tax externaltes, and equalsaton transfers. We test these hypotheses wth data on personal ncome taxaton n the Canadan provnces durng the last two decades. The structure of the paper s as follows. In the next secton, we provde a smple theoretcal analyss of provncal tax settng decsons wth the sole purpose of settng a benchmark for selectng approprate varables to nclude n the econometrc model. In the thrd secton, we test the dfferent hypotheses developed wth data on Canadan personal ncome taxaton for the perod Fnally, the last secton offers the man conclusons. 2. Theoretcal background 2.1 Vertcal Tax Externaltes In order to set up the hypothess we want to test n the emprcal analyss, n ths secton we present a smple model of personal ncome taxaton n the presence of tax base overlappng, based on Boadway 3 See also Saavedra (2000) and Wheaton (2000) for evdence on welfare competton among U.S. states. 4

6 and Keen (1996). Followng ths paper, we wll suppose that each layer of government maxmses the ndrect utlty functon of a representatve taxpayer subect to ts budget constrant and takng as gven the behavour of the other layer of government (.e., they behave Nash). For the provncal government, the maxmsaton problem s as follows, max t υ( w ) + h( g) + H ( G) subect to g = t.w.l ( w ) + n n R where w n =w(1-t-t) s the net wage rate; w s the gross wage rate; t and T are the provncal and federal personal ncome tax rates; g and G are the provncal and federal publcly provded prvate goods; and l(w n ) s labour supply 4. Therefore, t.w.l(w n ) are the provncal revenues comng from ncome taxaton whle R are exogenous provncal revenues (e.g., lump-sum grants). Then, the frst order condton of the maxmsaton problem s 5 : h' ( 1 t( l' / )) = 1/ l (1) where we have supposed the margnal utlty of prvate ncome s equal to one. We defne the left hand sde of the above expresson as the margnal beneft of publc goods (MB), whle the rght hand sde s defned as the margnal cost of publc funds (MCPF). Note that only the provncal tax rate enters nto the calculaton of the MCPF. Ths means that the provncal government underestmates the true margnal cost of taxaton, snce the ncome tax burden s n fact t+t. Because of ths vertcal tax externalty, the provncal government wll set a tax rate that s too hgh from the socal pont of vew. From the frst order condton (1), we obtan the provncal tax settng functon: t=f(t, w, R ) (2) Ths tax-settng functon s the bass for the econometrc model we wll estmate n secton 3. The 4 Labor supply, l=l(w n ), and consumpton demand x=x(w n ) are obtaned from the utlty maxmsaton problem of the representatve consumer, that s by maxmsng u(x,l) + h(g) + H(G) subect to x=w n l. 5 For functons of a sngle varable, partal dervatves are ndcated by '. 5

7 comparatve statc of expresson (2) allows us to state that hgher wage rates wll translate nto lower or hgher tax rates 6 and that hgher exogenous resources wll allow provnces to set lower rates. Also, we expect there wll be nterdependences n tax settng (.e., t/ T 0). However, we are not able to unambguously state the sgn of the vertcal nteracton. We can decompose the reacton of the provnce n front of changes n T nto two effects. On the one hand, an effect over the margnal beneft of publc goods (MB): an ncrease n the federal tax rate decreases revenue collected by the regon and, therefore, the level of the provncal publc good provded. To restore the prevous level of utlty derved from the publc good, the provnce has to ncrease ts tax rate 7. On the other hand, there s an effect over the margnal cost of publc funds (MCPF). After the ncrease n the federal tax rate each addtonal unt of tax revenue wll bear a dfferent margnal cost. If the MCPF decreases, ths wll renforce the prevous effect and the tax nteracton wll be postve. However, f the MCPF ncreases, ths effect wll pont to a reducton n the provncal tax rates, thus counterbalancng the effect over MB. Nevertheless, we are not able to state whether the margnal cost wll decrease or ncrease, but t wll depend on the functonal form of the labour supply wth respect to the tax rate 8. Takng both effects nto account we cannot unambguously sgn the reacton of the provnces n front of changes n the federal tax rate. Thus t s necessary to perform an emprcal analyss to know the drecton of the vertcal tax nteracton. 6 There are two effects at work. Frst, for a gven level of publc good provson, a hgher wage means that the same revenue could be rased wth a lower tax rate. Second, a hgher wage may ncrease the demand of publc servces, f these are normal goods. 7 As long as the margnal utlty of the provncal publc good s decreasng. 8 Keen (1998) found the functonal relatonshp of the labour supply wth respect to the tax rate under whch we can sgn t/ T n the case the government follows a revenue maxmsng behavour (so leavng asde the effects through the margnal beneft of publc funds). Hence, accordng to hs fndngs, log convexty s a necessary and suffcent condton for t/ T>0. It can be shown that n our case that condton expresses as: (l /l )<(l /l). 6

8 2.2 Horzontal Tax Competton Another source of tax nteracton n a federal system comes from competton n taxes among provncal governments. Any ncrease n a provncal tax rate wll generate an outflow of tax base to other provnces n order to avod the ncrease n the tax burden. Thus, n ths case, the dstortonary nature of the tax comes out because of the potental mgraton of the tax base. Usually, n order to analyse ths source of tax nteracton, the lterature has focused on captal taxaton (Zodrow and Meszkowsky, 1986; or Wlson, 1986). However, as our emprcal analyss deals wth ncome taxaton, we wll suppose that labour s the moble factor to be taxed by the provnces and the federal government (e.g., see Wldasn, 1991). In any case, horzontal tax competton results n an neffcently low level of taxaton, snce provncal governments are n a prsoner s dlemma (Boadway and Wldasn, 1984) due to ther fears of losng tax base 9. In order to derve the tax settng functon of the provnces n the presence of horzontal tax competton, we wll present a rather rudmentary model based on three smplfyng assumptons. Frst, we wll suppose that the populaton of each provnce, N, s composed by two groups. One group s formed by those who supply labour and bear neglgble mgraton costs (n ). The other group s composed by those who also provde labour wthn the provnce, but bear nfntely hgh moblty costs, so they are mmoble, n ~. Second, we suppose mmoble workers are unsklled, whle moble workers are sklled. Therefore, mmoble workers earn a wage rate, moble workers, w ~, lower than the wage earned by w. Thrd, we wll suppose that those ctzens who are moble do not get any beneft from the publc goods provded by the regons 10. The locaton decsons of the group of moble ctzens are made accordng to the condton: 9 Ths concluson holds as long as we adopt a welfarst approach. Otherwse, the results could be reversed. For nstance, f provncal governments are a Levathan, horzontal tax competton can be welfare enhancng (see Brennan and Buchanan, 1980, and Edwards and Keen, 1996). 10 These assumptons are akn to those used by Wldasn (2000), wth the excepton that he also consders the taxaton of moble captal. 7

9 ' ' f (1 t T ) = f (1 t T ), (3) Where the gross wage rate s now equal to the margnal productvty of sklled labour, f '(n ) = w, and f ''(n )<0. Then, accordng to expresson (3), a moble taxpayer wll choose her place of lvng n such a way that her net wage rate n s equal to the net wage rate she could acheve n any other regon of the federaton. Note that an ncrease n the tax rate of provnce wll produce an outflow of sklled workers tryng to recover ther prevous net wage rate, whch wll happen untl condton (3) s restored. The counter-balancng effect that avods an nfnte outflow of moble workers from provnce to the rest of provnces s the new hgher equlbrum gross wage rate there for sklled workers 11. From (3), we can obtan the mgraton functon of moble taxpayers lvng n : n = f(t, t -, T) (4) where the subscrpt ndcates the set of competng provnces of, and where comparatve statc analyss show that n / t <0, n / t - >0 and the sgn of n / T s ambguous (see Appendx). An nterestng feature of ths model s that the mpact of a change n the tax rate on the mgraton of moble populaton dffers accordng to the sze of the provnce; that s, n t 0 for bg provnces, whle n t > 0 for small provnces. Therefore, those bg regons wll provde a level of publc good provson closer to the frst-best than small provnces 12. Addng the constrant mposed by the moblty of the tax base (4), the maxmsaton problem of the provnce can be re-formulated n the followng way: ( w n + w ~ n ~ ) l + R and n = f ( t, t, ) max υ( w ~ ) + h( g) + H ( G) subect to g n ~ = t T t n 11 See Feldsten and Vallant (1998) for an emprcal analyss of ths effect for US states. 12 See the Appendx for a proof. However, t can be shown that small regons are better-off under tax competton than wth dentcal tax rates (see Wellsch, 2000, pp ). 8

10 Note that the new budget constran consders that the tax base of the provnce s now ( w n w ~ n ~ )l and that the publc good s now consumed only by the mmoble factor. + 13, The frst order condton of the maxmsaton problem s: h' = ' n f 1+ ~ ' n ~ f 1 n t n t n f f '' ' (5) Observe that the MCPF, the rght-hand sde of (5), ncreases wth the mgraton response to tax rate changes ( n / t ). Ths fact llustrates the horzontal tax externalty. However, the fact that the outflow of ctzens captalses n a hgher equlbrum wage rate for the group of "moble" ctzens that stay n (accounted by f' '/f' ) tends to lower the MCPF. The mportance of these two effects ncreases wth the rato of the amount of tax base produced by moble workers to the tax base produced by mmoble workers, (.e., n f / n ~ ~ f ). From the frst order condton (5) we can obtan the provncal tax settng functon: t =f(t, t -, N, n ~, R ) (6) That s, the tax rate chosen by a provnce depends on the tax rates of competng provnces (t - ). However, the sgn of the horzontal tax nteracton s ambguous, as n the case of the vertcal tax externalty. A change n the tax rate of competng provnces prompts two dfferent effects. The revenue collected by wll ncrease after the ncrease n the tax rate of any other provnce, due to the postve externalty comng from the nflow of tax base 14. Therefore, the MB of publc goods wll decrease and, to restore the prevous level of utlty derved from the publc good, the provnce wll have to decrease ts tax rate. Therefore, the sgn of the reacton of provnce n front of varatons n the tax rate of any other provnce depends on the varaton produced on the MCPF, that could be ether 13 The number of hours of work suppled by each worker, l, s consdered as fxed n order to solate the dstortons comng from mgraton decsons. 14 That s, g t = ( n t ) t f ' l(1+ N ( f '' / f ' )) 0. 9

11 postve or negatve. Ths makes the determnaton of the sgn of the horzontal nteracton also an emprcal ssue. The are other facts that dfference the tax-settng functon n (6) from that n (2). Frst, although expresson (6) also ncludes the federal tax rate (T), ts nterpretaton s somewhat dfferent. The effect of T only accounts for the fact that, when regons are dfferent, the federal tax rate has some ncdence on mgraton flows, as expresson (4) ndcates (see Appendx). The effect of T derved from vertcal externaltes can not work n ths settng because the labour supply per worker has been supposed as fxed [see footnote (13)]. However, n a more complete settng, allowng for both types of dstortons (.e., fscal moblty and reducton n hours worked) both horzontal and vertcal externaltes may be present. Second, tax rates wll be hgher the hgher s the number of mmoble taxpayers ( n ~ ). Thrd, note that the gross wage rate (w) does not appear anymore n the tax-settng functon. Ths happens because the wage rate s now endogenous; n fact, t s the factor that re-equlbrates the labour market n front of tax dfferentals. Fourth, note that tax rates wll also depend on the populaton of each provnce (N ): bg provnces seem to be mmune to tax base losses and are less affected by changes n the tax rates of the rest of provnces (.e., n / t 0 and n / t 0, see Appendx). Also, changes n the tax rate of bg provnces prompt bgger mgraton changes n the other provnces. Three dfferent emprcal predctons derve from ths fact: () we expect bg provnces to set hgher tax rates than small ones, () bg provnces wll react less than small ones n front of changes n tax rates of compettors, and () the reacton n front of a tax rate change of a gven provnce wll be hgher the bgger s ths competng provnce. We wll take ths nto account n the emprcal specfcaton. 2.3 Equalsaton Grants The tax settng behavour of the provnces may also change as a result of the ncentves provded by a system of equalsaton grants, as Smart (1998) showed. These ncentves arse because the equalsaton 10

12 grant does not only provoke an ncome effect as any lump-sum grant, but also a prce effect. Ths latter effect reduces the margnal cost of publc funds of the recevng provnces and, ceters parbus, forces them to ncrease ts tax rate. Analytcally, the Canadan equalsaton grant, E, can be expressed as: E = t ( wl wl) 0 f wl > wl and E = 0 f wl wl where wl s the standard equalsaton tax base, that can be defned as the tax base wth respect to whch the tax capacty of each provnce s compared. Hence, f the tax base of a provnce s below such level, t wll be compensated n proporton to t, the average tax rate of the entre federaton 15. Otherwse, the provnce s nether enttled to any transfer through the equalsaton system nor has to contrbute to the system. The equalsaton grant system modfes the maxmsaton problem of secton 2.1 for the recevng provnces n the followng way: max t υ( w ) + h( g) + H ( G) subect to g = t w l( w ) + t ( wl - wl( w )) + R n n n The frst order condton of ths problem can be expressed as follows: h' ( 1 ( t t )( l' / )) = 1/ l (7) Thus, ceters parbus, the equalsaton grant reduces the margnal cost of publc funds, and so tends to ncrease the provncal personal ncome tax rate above the level that would have been set wth a lumpsum grant (Smart, 1998). From the expresson above, we obtan the provncal tax settng functon for equalsaton-recevng provnces: t = f (T, w, t, wl, R ) (8) 15 We consder the varables t and w. l changes n provncal tax rates could affect them (see, e.g., Courchene and Beavs, 1974). For these as fxed n the analyss, though some authors have consdered that effects to be relevant, the provnce has to be relatvely bg. Ths s not the case for the provnces recevng equalsaton, wth the excepton of Quebec. However, note that when Quebec ncreases ts tax rate, the average tax rate rases, but the standard tax base shrnks (f the tax s dstortonary), producng an ambguous net effect on the revenues of ths provnce. 11

13 There are two man dfferences between ths tax-settng functon and the one n (2). The frst one refers to the ncluson of the standard tax base ( w l ).The effect of ths new varable on the tax rates set by the provnces s smlar to the effect of the other exogenous revenues ( R ). The second refers to the ncluson of the average tax rate (t ); the sgn of the reacton of the provncal tax rate to changes n ths varable ( t/ t ) s ambguous, snce there are two effects at work. Frst, an ncrease n t wll lower the margnal cost of publc funds, so forcng an ncrease n the provncal tax rate (Smart, 1998). However, there wll also be an ncome effect that wll reduce the margnal beneft of publc goods (as any lump-sum grant), so promptng a decrease n the provncal tax rate. It s necessary to perform an emprcal analyss n order to ascertan the sgn of the reacton. However, ths s not the only hypothess regardng the mpacts on tax settng of equalsaton grants that we wll take nto account n the emprcal analyss. Note that we have treated the effects of that grant n solaton from the two knds of externaltes (vertcal and horzontal) analysed n the prevous sectons. In fact, some authors have suggested that equalsaton payments may have some potental to allevate or exacerbate these externaltes. For example, the capacty of equalsaton transfers to mtgate tax competton s an old argument of the lterature (see, e.g., Reschovsky, 1980). The reason of the reduced ncentves to compete s that under ths knd of programs the revenue gans comng from the attracton of tax bases to a ursdcton must be shared wth all the federaton. The relatonshp between equalsaton and vertcal externaltes s less clear. However, as Hayash and Boadway (2000, p.3) argue, the fact that recpent provnces are nsulated (at least partly) from the adverse consequences that rasng ther tax rates has on ther tax bases mght ntensfy the vertcal fscal externaltes. Two addtonal emprcal predctons derve from the dscusson n the last paragraph: frst, the mpact of horzontal tax competton wll be lower for the recevng provnces, and second, the mpact of vertcal tax competton may be hgher n the recevng provnces. 12

14 3. Emprcal mplementaton 3.1 Country, tax and perod: Canadan Personal Income Tax To test the dfferent tax nterdependence hypothess developed n secton 2, we wll use data correspondng to the personal ncome tax n Canada durng the last two decades. The Canadan personal ncome tax s partcularly nterestng for our purpose because there are not many countres where regonal governments have tax powers over that feld. In Canada, the personal ncome tax s the man tax both for the federal and provncal governments. For the federal government the share of revenues comng from that source was 43.7% n 1982 and ncreased to 47.2% n The correspondng fgures for the provncal governments are lower 21.5% n 1982 and 25.9% n 1993, snce they also obtan revenues from sales and natural resources taxaton and from varous types of transfers. The share of personal ncome tax revenues gong to the provncal governments roughly 30% and also the average provncal ncome tax burden 6.72% of Personal Income n 1982 and 8.74% n 1993 are hgh by nternatonal standards. These facts make the Canadan personal ncome tax a natural focus for an emprcal study on vertcal and horzontal nteractons. There s also some concern among Canadan economsts over externalty problems n the feld of ncome taxaton. For example, the ncrease n provncal ncome tax rates durng the last two decades (and the ntroducton of surcharges on hgh ncome taxpayers) has been nterpreted by some authors as evdence of excessve taxaton caused by nteractons between layers of government (see, e.g., Dalhby, 1996). Many papers have shown the emprcal relevance of personal ncome tax rates and other fscal factors as determnants of nterprovncal mgraton (see, e.g., Wner and Gauther, 1982 or Day, 1992), showng thus that horzontal tax competton may also be of concern. Boadway and Flatters (1982) have shown that ths fscally nduced mgraton could end up n an neffcent allocaton of labor across provnces, provdng a ustfcaton for equalsaton transfers. Nevertheless, analyses of tax nteractons among provncal or local governments are very rare; Hayash and Boadway (2000) on provncal busness ncome taxes and Brett and Pnkse (1999) on local property taxes are the 13

15 exceptons. Ths fact contrasts wth the growng U.S. lterature on personal horzontal tax nteractons (see, e.g., Case, 1993; Besley and Case, 1995; and Brueckner and Saavedra, 2001). The advantage of focusng on the Canadan case also rests n the possblty of testng the ncentve effects ntroduced by the equalsaton program. There are few countres n the world where such a transfers system operates ontly wth a hgh degree of provncal tax autonomy and tax base overlap between federal and regonal governments. Also the exact desgn of the program helps us to devse the emprcal test. The Canadan equalsaton system transfers funds from the federal government to those provnces whose capactes to rase revenues fall below a natonal standard. The amount of transfer s calculated by applyng the average provncal tax rate,.e., the standard tax rate, to the dfference between the standard tax base and the tax base of the provnce, whle the rch provnces do not receve equalsaton funds. Thus, the transfer system s not zero sum,.e., the rch provnces do not fll the revenue capacty gap of the poor ones. Therefore, the ncentve to rase tax rates s experenced only by the recevng provnces and, as Smart (1998) demonstrates, ncreases wth the standard or average tax rate. In addton to the equalsaton program, other nsttutonal features of the Canadan federal system should be taken nto account when nterpretng the evdence on personal ncome tax nteractons. In partcular, the emprcal test of vertcal tax externaltes should consder the very hgh degree of harmonsaton of the provncal ncome tax wth the federal one. For nstance, Quebec s the only provnce whose personal ncome tax does not conform at all to federal defntons. In the rest of the provnces, the personal ncome tax conssts of an applcaton of a basc proportonal rate over the federal tax lablty. Ths basc rate ranged n 1993 from the hghest 69% n Newfoundland to the lowest 45,5% n Alberta. Ths fact constrans provncal ncome taxes to have the same dstrbutonal mpact that the federal one. But snce the begnnng of the eghtes the provnces have tred to gan degrees of freedom n tax settng usng surcharges over the tax lablty of hgh ncome taxpayers The frst provnces to make use of these surcharges where Saskatchewan, Brtsh Columba and Mantoba; Ontaro, Alberta and Prnce Edward started n 1986, 1987 and 1988, and New Brunswck, Nova Scota and Newfoundland n 1991, 1993 and 1996 (source: Provncal Budget Roundup, Canadan Tax Foundaton). 14

16 More recently some provnces have begun to set a flat tax rate over net taxable ncome 17. Ths means that the drect mpact on provncal revenues of a federal tax reform s dfferent for those groups of provnces. On the one hand, for Quebec, there s not any drect mpact on revenues collected. On the other hand, for the rest of the provnces, all the changes n federal effectve tax rates (comng from changes n the defnton of the tax base, changes n deductons and tax credts, and changes n statutory tax rates) wll affect ther revenues. Moreover, the mpact on revenues wll be even hgher for the provnces that use a hgh ncome surcharge 18. Nevertheless, the wnd-fall gans (or losses) arsng from ths knd of drect effect we have related above should not have any nfluence on the optmal tax rates derved from the theoretcal tax model developed n secton 2. Ths s due to the fact that a government that maxmses the utlty functon of the representatve ctzen would react to the changes passed by the federal government adoptng the necessary legal actons to keep effectve tax rates at the optmum. If the estmaton results ndcate that the provnces do not react to the wndfall gans or losses, ths would suggest the exstence of some sort of fscal lluson 19. However, n some cases t would be dffcult to dsentangle the passve adustment from the actve reacton due to the vertcal tax externalty. That would happen f both reactons go n the same drecton. The only way to test f the reacton s due to economc behavour or to fscal 17 The use of the flat tax began n Saskatchewan n 1985, was followed n 1986 by Mantoba and Alberta and has been also set n 1996 n Brtsh Columba and Ontaro (source: Provncal Budget Roundup, Canadan Tax Foundaton). 18 The sze of ths drect mpact (dt/dt) can be expressed n the three cases as: 0 (non-conformng), b (fully harmonzed), and b+α.s (hgh ncome surcharge); where we have defned t=provncal tax burden, T=federal tax burden, b=provncal basc rate, s=provncal surcharge, α=share of federal revenues comng from hgh ncome taxpayers. Let suppose, for example, that b=0.50 and s=0.30; f α=0.20, then the surcharge wll ncrease the drect mpact from 0.50 (fully harmonzed) to Several recent analyss on the effects of personal ncome tax reforms n the U.S. document substantal emprcal evdence of ths knd of automatc nterdependency, and thus of the exstence of certan degree of fscal lluson n tax settng behavour (see, e.g., Gold, 1991, Ladd, 1993, and Tannelwald, 1991). 15

17 lluson s to allow the slope of the reacton functon to dffer among the three groups of provnces prevously defned. Fnally, n the Canadan case the perod of analyss should be carefully chosen n order to ensure that the test of tax nteractons makes sense. The emprcal analyss s carred out wth data on the ten provnces durng the perod The frst year of the perod we analyse s 1982 because n the prevous years a feature of the Canadan federal system was the substtuton of cash payments for addtonal tax room (manly n personal ncome taxaton) 21. In ths settng the data would show an nterdependence between federal and provncal tax rates that s not drectly related to the vertcal tax externalty hypothess 22. The nterest of the perod analysed s also ustfed by the substantal changes n the structure of the Canadan Federal Personal Income Tax. The 1981, 1987 and 1991 tax reforms reduced the number of tax brackets from 13 to 10 and 3, respectvely. The top federal margnal tax rate was also reduced form 43% to 34% n 1981 and to 29% n 1987 (Doak, 1996), but some surcharges were ntroduced n the followng years (Howard et al., 1995). Although these changes translate nto a lesser extent to 20 We exclude Yukon and the NorthWestern Terrtores from the analyss because, though they are enttled to enter the feld of ncome taxaton, they have a specal status n the federaton (Perry, 1989). 21 The Tax Collecton Agreements ( ) represented the begnnng of a trend toward ncreased fscal responsblty. The new provncal ncome taxes of 1962 were accommodated wth an abatement n the federal ncome tax that was ncreased n subsequent years (1965 and 1966). Later on ( ), under ncreased pressure for fscal autonomy, an addtonal abatement (n exchange for some specfc grant programs) was gven to Quebec and then extended partly to the other provnces. The Tax Transfers Agreements ( ) saw the substtuton of the tax abatement for the reducton n federal tax rates, and an addtonal cesson of tax room n exchange for a reducton n specfc grants (Establshed Programs Fnancng, n 1976). In the arrangements of more emphass was gven to cash payments, and no further tax changes were made. See Perry (1989, chapter 16) for a more detaled analyss of all these nsttutonal changes. 22 Note, for nstance, that the coeffcent of correlaton between average effectve federal and provncal ncome tax rates s for the perod and 0.91 for the perod (source: Revenue Canada, Taxaton Statstcs). Ths correlaton does not of course tell us anythng about causaton, but the negatve sgn for the frst perod seems to reflect the cesson of tax room. 16

18 modfcatons of average effectve tax rates (Daves and Zhang, 1996), ths varable also shows perods of decrease and ncrease durng the sample analysed, provdng thus enough varaton to carry out the emprcal analyss. 3.2 Emprcal framework Our emprcal purpose s to estmate the magntude of the reacton of provncal ncome tax rates to exogenous changes n federal ncome tax rates, changes n ncome tax rates n competng provnces and changes n the standard ncome tax rate used n the equalsaton program. Ths amounts to estmatng the slope of the reacton functon of a representatve provncal government vs a vs the federal government and the rest of the provnces. We take nto account that all these tax decsons are made smultaneously and, thus, that both the federal tax rate and the tax rates of the competng provnces are potentally endogenous. However, we postpone the dscusson of ths ssue to the next secton. The basc specfcaton to test these hypotheses s: t t k = 1 Tt + α2 X ω X t t + α3 Xt t X deq,t + αk X Zt 1 + α0 k α X + ε (9) t Where t,t and T,t are the provncal and federal tax rates, and subscrpts and t ndcate provnce and year. We use the average effectve tax rates as the defnton of our tax varables. These tax rates are constructed as the rato of ncome tax revenue to the Personal Income Component of GDP 23. Data for ths calculaton has been obtaned from the CANSIM database (see Table 1). We have chosen personal ncome nstead of taxable ncome to avod the effects that changes n the defnton of the tax base between provnces (bascally dfferences between Quebec s and federal tax laws) and over tme have on the average effectve tax rate. Ths procedure s supported by the way ncome tax reductons are usually carred out. Although n many cases the change n effectve tax rates has followed statutory tax changes or changes n deductons and credts, n many others ncome tax reforms have also changed the defnton of tax bases. Therefore, dong the calculaton n that way means that we do not accept as 23 Ths ncludes salares, wages, and supplementary labour ncome; nterest and mscellaneous nvestment ncome; and net ncome of farmers and other unncorporated busnesses. 17

19 accurate, relable or stable the defnton of tax base gven by federal or provncal governments n Canada. There are two man reasons for the use of average effectve tax rates nstead of usng, for example, statutory margnal tax rates 24. Frst, gven that personal ncome taxes n Canada are progressve, no sngle margnal tax rate accounts for the ncentves faced by every household. Even provncal ncome taxes, once a basc proportonal rate appled over the federal tax lablty, s now charactersed by a varety of surcharges, flat rates and deductons that make them farly more complex. The lmtaton of the emprcal analyss requres us to descrbe tax polces n terms of a sngle summary statstcs average effectve tax rates. Some authors defend ths poston (see, e.g. Mendoza et al., 1994) and argue that average tax rates are a good way to take nto account the effectve, overall tax burden of a tax. In addton, other analyses of fscal nteractons use smlar measures of tax rates (see, e.g., Goodspeed, 2000, for the personal ncome tax, and Hayash and Boadway, 2000, for the busness ncome tax). Second, average tax rates are often employed to represent nterursdctonal dfferences n taxaton. If tax bases respond to taxes because of nterprovncal moblty of workers, t s prmarly the average tax rate, and not the margnal tax rate, that nfluences mgraton decsons (see, e.g., Carrol and Wasylenko, 1994). However, there s a man problem n the utlsaton of average effectve tax rates. Observed average tax rates reflect two types of nfluences. They have a polcy component, arsng from delberate choces by government offcals and based on nformaton avalable to them, but they also have an unexpected one. That s, for example, actual provncal average tax rates are the result of the tax parameters that were chosen by publc offcals (basc provncal rate, surcharges, flat tax, tax credts), gven ther antcpaton of the sze of the base and the structure of the federal ncome tax. However, actual tax rates may also reflect, for example, the effect of exogenous factors affectng the level and 24 Some authors have advocated the use of margnal tax rates computed drectly from the statutory tax functon (Barro and Sahasakul, 1986), or usng the top margnal tax rate (Tannewald, 1991, and Mullen and Wllams, 1993). 18

20 dstrbuton of personal ncome. We consder how to deal econometrcally wth ths ssue n secton 3.3. Horzontal tax nterdependences are controlled for n our emprcal analyss wth the ncluson n equaton (9) of a weghted average of the tax rates of competng provnces, where t t s the tax rate set by provnce and ω s the weght of regon n the set of competng regons of regon. To compute these weghts (ω ) we need to consder the ssue of whch provnces are most lkely to be the ones from whch one provnce can expect to loose or gan employment. The easest way to handle ths problem s to consder that dstance matters for horzontal competton. In fact, some papers that estmate the mpact of tax dfferentals on mgraton n Canada (see, e.g., Wner and Gauther, 1983, and Day, 1992) have also found a hgh and sgnfcant mpact of dstance on nterprovncal flows of resdents. As Day (1992) ponts, ths could be nterpreted as evdence that people prefer to move short dstances f at all 25. The use of dstance to compute these weghts has also been common practce n recent analyses of horzontal tax nteractons among U.S. states (see, e.g., Case, 1993; Besley and Case, 1995). Followng ths lterature, we consder the contguous provnces as the source of mmgraton and emgraton of resdents. Ths means that the Atlantc provnces belong to the same set of recprocal compettors and are also nfluenced by Quebec. The compettors of Quebec are the Atlantc provnces and Ontaro. The compettors of Ontaro are Quebec and Mantoba. Fnally, each of the Western provnces (to the excepton of Brtsh Columba) has two potental compettors, one on each sde. Note, however, that the set of compettors of some provnces s composed of very dfferent provnces: n some of them the man populaton centre s located really far away whle n others s very close, some are very bg and some very small. In order to consder further the mpact of dstance and to account for the possble asymmetrc mpact of bg and small neghbours 26, we weght each 25 Ths author carefully quantfes the movng costs among provnces, takng nto account the travel cost of people and belongngs, and the average tme needed to move between provnces tmes the foregone wage (see Day, 1992, p. 17). 26 Ths s consstent wth the fact that n the tax competton model of secton 2, the mpact of the change n the tax rate of a competng provnce s hgher the hgher ts sze (see Appendx). 19

21 contguous provnce by populaton and nverse dstance. Thus the weghts attached to each provnce (ω ) are: ω = w w c w here w =. d n Where c =1 f provnces and are contguous, d =dstance n mles between the man populaton centre of provnces and, and n =resdent populaton of provnce. Note that the weghts (ω ) are standardsed and, thus, sum up one. Ths procedure s common n ths knd of analyses (Anseln, 1988) and allow us to nterpret the estmated coeffcents as the response to the average tax rates of competng provnces. 27 The effects of equalsaton grants are accounted for n equaton (9) by ncludng the standard ncome tax rate used for equalsaton purposes ( t t ) nteracted wth deq,t, a dummy varable equal to one f the provnce s enttled to receve equalsaton n that year. Three provnces dd not receve equalsaton payments throughout the perod analysed (Alberta, Brtsh Columba and Ontaro), one provnce (Saskatchewan) receved payments only n the fscal years and to , and the rest have been enttled to the equalsaton program all the years. The average provncal tax rate of the federaton ( t t ) has been computed for each provnce after the excluson of ts own tax rate, n order to avod endogenety problems. Although t t only changes yearly, the fact that only a fracton of the Canadan provnces receve equalsaton transfers, allows the varable ( t t ) to have cross-secton varaton. As we stated n secton 2, the parameter α 3 can be ether postve or negatve, dependng on whch effect (ncome or prce) domnates. 27 We also tested other weght defntons, but they dd not perform as well as ths one (contguty alone, contguty wth nverse of dstance, contguty wth populaton) or that prove totally neffectve (add a second order set of contguous provnces). 20

22 The other varables ncluded n equaton (9) are a vector of other control varables Z, ndexed by k. These controls are needed because tax nteractons are ust some of the factors that nfluence tax settng. Tax decsons are carred out wthn a very complex nsttutonal process that accounts for the preferences of voters, electoral nterests of poltcans, and economc, nsttutonal and cultural constrants. Many of those nfluences may be correlated wth the tax rates of the federal and competng provncal governments, thwartng any drect nference about ts effects on provncal tax rates. As a result of ths, and n order to solate the effects of tax nterdependences, t s necessary to control for all the other relevant varables that affect the regonal tax-settng process. In order to avod endogenety problems all these varables are ntroduced nto the equaton wth one lag. We nclude several groups of control varables (see Table 1 for defntons and statstcal sources): Economc resources. Rcher populatons wll demand more regonal publc goods and, n consequence, wll tolerate hgher ncome taxes; we nclude Personal ncome per capta to control for ths effect. However, the effect of an ncrease n ncome on the tax rate s uncertan, due to the fact that the same level of revenue s now sustanable at a lower tax rate. Also, the amount of grants receved by the regonal government can have an effect on the ncome tax rate. Although rcher governments that s, governments recevng more grants from the federal government wll also spend more, they wll return part of the amount receved to ts ctzens; so we expect that ncome taxes wll be lower, the hgher the amount of transfers receved. To account for ths effect, we nclude three varables that control for the amount and type of transfers receved. These are: the per capta amount receved from General purpose transfers (manly equalsaton payments), the per capta amount receved from Other transfers 28, and the revenue from Natural resources per capta. Note that t s necessary to ntroduce these varables wth a lag. Ths s due to the potental endogenety of per capta ncome (.e., f the tax 28 One could argue that some federal specfc grants are matchng whle they are mplctly consdered as lumpsum. However, year-to-year changes n the structure of matchng grants affect all provnces smlarly; hence ther mpact may be controlled by tme effects. 21

23 s n fact dstortonary, the base shrnks when the provncal tax rate ncreases) 29 and equalsaton receved (.e., the grant ncreases when the tax base shrnks) 30. Expendture needs. Populatons wth hgher shares of potental users of publc servces and/or hgher cost of delverng those servces wll need hgher levels of expendture and, therefore, wll be burdened more heavly through ncome taxes. We nclude as explanatory varables the sze of the Populaton and ts ncrease n the last fve years (% Populaton) 31. We also ntroduce n ths group of varables the sze of varous groups of potentally ntensve servce users: the proporton of populaton over 65 (% Populaton over 65) and under 15 (% Populaton under 15) and the proporton of populaton unemployed (% Unemployed). Poltcal envronment. Although many poltcally motvated models of publc polcy generaton suggest that partes converge at the same platforms regardless of ts deology, many others suggest that f poltcans are polcy-motvated and do not only care about wnnng electons the polces mplemented need not be the same 32. We nclude a dummy that accounts for the deology of the regonal executve (.e., dleft=1 f the executve s on the left wng of the provncal poltcal arena 33 ) 29 Recall also that w has been consdered endogenous n the horzontal externalty case, and thus has not been ncluded n that tax-settng functon. 30 Recall that the tax-settng functon presented n secton 2 for the equalsaton case ncludes the sze of the standard tax base ( w~ ~ l ) as a control. However, n order to avod problems derved from the colnearty wth other varables (manly the payments receved from the remanng categores consdered by the equalsaton program) we add the equalsaton payment n ncome taxaton to other categores payments, but one perod lagged to avod endogenety problems. 31 As we stated n secton 2.2, the sze of the populaton also reflects the fact that bgger provnces are able to set hgher tax rates due to ts market power. 32 See Alesna and Rosenthal (1995), chapter 2, for a survey. 33 The partes consdered on the left are the New Democratc Party and the Part Québécos. 22

24 and a varable that accounts for the fragmentaton of the government (.e., dmnorty=1 f the government s n mnorty n the legslatve) 34. The coeffcent α 0, represents a provncal tme-nvarant ndvdual effect. The estmaton of a fxed effects model by OLS wll gve us consstent estmates of the tax nteracton parameters whenever the provncal effects are correlated wth the explanatory varables ncluded n the equaton (Mundlak, 1978, Hausman and Taylor, 1987). Provncal fxed effects represent specfc crcumstances of each state that stay relatvely constant durng the analysed perod: characterstcs of the local poltcal market, specfc dfferences n the cost of local publc servces (e.g., geography, clmate) or a permanent nflow of revenue from other tax resources. Otherwse, f these were correlated wth the varables ncluded n the emprcal model, the obtaned parameters would be nconsstent (Holtz- Eakn, 1986). Of course, f some of the regressors of the equaton were endogenous, these statstcal propertes of the fxed effects model would no longer hold. Ths basc specfcaton s then extended to account for the possble nteracton among the Equalsaton Program and ntensty of the two knds of externaltes analysed. To account for ths possblty the federal tax rate (T t ) and the tax rates of the competng provnces (Σ ω x t ) are nteracted wth a dummy that ndcates whether a provnce s enttled or not to receve equalsaton transfers (deq t ). t,t = α 11 X T + α X 3,t + α deq 12,t X t deq X t +,t k x T k,t + α α X Z 21 k,t 1 X + α 0, ω X t + ε,t,t + α X 22 deq,t X ω X t,t (10) The parameters α 11 and α 21 ndcate the strength of vertcal and horzontal tax nteractons, respectvely, n the provnces not recevng equalsaton transfers. The sum of parameters (α 11 +α 12 ) 34 The results of some emprcal studes suggest that dvded governments are more vulnerable to redstrbutve pressures, and ths fact results n dffcultes to undertake reforms to cut spendng or defcts; see, for example, Roubn and Sachs (1989) and Alt and Lowry (1994). In the case of ncome taxaton, dvded governments could fnd more dffcult to elmnate loopholes from the tax code. 23