Lars P. FELD, University of St. Gallen, SIAW and CESifo. Jean-Michel JOSSELIN, University of Rennes and CREREG

Transcription

1 "Tax mmckng among regonal ursdctons", n A. Marcano and J.M. Josseln (eds), From economc to legal competton. New perspectves on law and nsttutons n Europe, London : Edward Elgar, , 2003, wth Lars P. Feld and Jean-Mchel Josseln. Lars P. FELD, Unversty of St. Gallen, SIAW and CESfo Jean-Mchel JOSSELIN, Unversty of Rennes and CREREG Yvon ROCABOY, Unversty of Rennes and CREREG Abstract: Yardstck competton has recently ganed attenton n the analyss of fscal nteractons among regonal or local governments. Voters compare the fscal polcy of ther government to that n other or neghborng ursdcton to evaluate the fscal performance of ther representatves. Ths may lead the latter to adopt the fscal polcy of other ursdctons. The dfference wth tax competton s that such a copycattng may also lead to a convergence of tax rates at a level hgher than the compettve one. In ths paper, such a model of yardstck competton s developed and emprcally tested usng data for tax settng behavor of the French regons snce the decentralzaton process of 1986 to Keywords: Yardstck Competton, Taxaton, French Regons. JEL Classfcaton: H72, H73, D72, D78, H24, H25. Ths proect has been started whle the frst author vsted the Unversty of Rennes 1. Fnancal support from the Swss Natonal Scence Foundaton and the hosptalty of the faculty s gratefully acknowledged.

2 2 1. Introducton Snce Tebout (1956), the focus of models of fscal federalsm has been manly on the mechansms and propertes of market-lke competton between nsttutons or ursdctons. In a survey on the lterature, Wlson (1999) has ponted out that such a competton for moble producton factors between ursdctons may well lead to a race to the bottom n tax rates. However, authors lke Salmon (1987) and Besley and Case (1995) have used alternatve or complementary explanatons of publc decson-makng processes n a settng of fscal federalsm. A tool of labor economcs (Holmstrom, 1982) and of the economcs of the frm (Shlefer, 1985), yardstck competton thus has ganed attenton as a very nsghtful concept n the feld of publc economcs. In a world of mperfect and asymmetrc nformaton, voters have restrcted possbltes to assess the performance of the representatves n ther polty. Selfsh representatves am at gatherng poltcal rents and hence have ncentves to keep nformaton about ther opportunstc behavor hdden from voters. However, voters can draw nferences on poltcans behavor by comparng t to the performance of governments and parlaments n neghborng ursdctons. Other thngs beng equal, these neghbors serve as yardstcks for the voters evaluaton. A worse performance n ther own ursdcton compared to other ursdctons leads to a punshment of representatves by throwng them out of offce n the next electons. In such a concept, publc choce would not only be drven by nformaton gatherng from neghborng ursdctons, but also by mmckng behavor. Because representatves antcpate the yardstck mechansm, they are able to stay n power by adaptng to the polces of ther neghbors. As ths ntuton has progressvely made ts way, a number of sgnfcant artcles (Ladd, 1992; Case, 1993; Case, Rosen and Hnes, 1993; Besley and Case, 1995; Heyndels and Vuchelen, 1997; Fglo, Kolpn and Reed, 1999; Brueckner and Saavedra, 1999; Saavedra, 2000) has contrbuted to the better understandng of the mechansm of yardstck competton n publc decsons. 1 The emprcal relevance of the hypothess n the poltcal market has also been assessed, partcularly n the Amercan case (e.g. Besley and Case, 1995). The present paper follows ths wake and ts contrbuton s orgnal n three ways. Frst, we start wth a glance on the actual development of tax rates of regonal taxes n France snce the decentralzaton of 1986 n order to derve some nformaton as to whether a race to the bottom n tax rates can be observed (Secton 2). If a race to the bottom cannot be found n the data, other explanatons than the tradtonal tax competton reasonng have to be developed. In order to derve such an explanaton, the theoretcal part of the paper proposes an agency model n the sprt of Besley and Case (1995) whch depcts a mult-perod relatonshp be- 1. For a survey on mmckng behavor of competng ursdctons n the case of the U.S. welfare system see Brueckner (2000).

3 3 tween some decsve voter and ts agent, the elected representatve (Secton 3). Yardstck competton wth neghborng ursdctons nduces an nterference n ths agency settng. The agent n one ursdcton wants to take advantage of shocks that are not observed by prncpals. The clamed shock on the cost of provson of the publc goods would then be forwarded n an ncrease n the tax rate wthout endangerng the prospects of re-electon. However, the ablty of the prncpal to observe the stuaton n neghborng ursdctons affected by smlar shocks threatens to allevate the possble strategc use of prvate nformaton by the agent. Our model here substantally dffers from that of Besley and Case (1995) snce opportunstc or bad behavors are not consttuent of the nature of the agents (they beng bad or good n the model of Besley and Case). Opportunsm s here smply one of the possble strateges that can be checked by comparson wth the behavor of agents n other ursdctons. The thrd contrbuton of the artcle s to provde an econometrc test wth data on French regons (Secton 4). The French decentralzed level s partcularly nterestng, because régons, départements and local ursdctons have non-neglgble tax settng power snce the decentralzaton of Thus, the analyss n ths paper s (to our knowledge) the frst emprcal test of tax mmckng n a country engaged n a sgnfcant devoluton process. Secton 5 offers concludng remarks. 2. The Development of Tax Rates n the French Regons Snce the decentralzaton of 1986, the French sub-natonal admnstratons are allowed to set tax rates n a wdened band fxed by the central government n Pars. The taxes that can be vared are the taxe d habtaton as a tax on housng, the taxe professonelle as a local busness tax, and the taxes foncères sur les proprétés bâtes et non-bâtes as two property taxes on propertes wth and wthout buldngs. The taxe d habtaton s ndependent from the property status. It s also leved on apartments of renters and thus bascally a tax on revealed demand for housng as an ncome elastc good. Fgure 1, where the development of the taxe d habtaton s drawn for the seven regons Alsace, Bourgogne, Bretagne, Centre, Franche Comté, Ile de France and Rhône-Alpes, reveals an nterestng pattern. Immedately after the decentralzaton of 1986 the varance of the taxe d habtaton ncreases consderably. The coeffcent of varaton of ths tax for all 22 regons ncreases from n 1985 to n After the decentralzaton that allowed regons more dscreton n tax rate changes, a slght convergence s observable for the sx regons n fgure 1. For all 22 regons the coeffcent of varaton decreases from n 1987 to n 1995 whch corresponds to a reducton of about 51 percent. Hence, n the ten years after decentralzaton, tax rates of the taxe d habtaton have consderably converged. Most mportantly, and n contrast to the predctons of the theory of tax competton, no race to the bottom occurs, but these tax rates converge to a hgher and stll ncreasng level.

4 Fgure 1: Tax Rates of the taxe d habtaton n Selected French Regons, 1984 to 1995 A smlar pattern of tax rate convergence between the French regons s observable n the case of the local busness tax. Fgure 2 llustrates ths convergence for the same seven regons as before. For all 22 regons, the coeffcent of varaton ncreases from n 1985 to n 1987 and decreases afterwards to 19.92, whch corresponds to a reducton of 51 percent as well. Lke the taxe d habtaton, the local busness tax ncreases contnuously durng ths tme. Whle a convergence of tax rates s also observable for both property tax rates, convergence s less pronounced than n the two prevous cases, as fgures 3 and 4 ndcate. For all 22 regons, the coeffcent of varaton decreases from n 1987 to n 1995, a reducton of 30 percent n the case of the property tax on property wth buldngs, and decreases from n 1987 to n 1995, a reducton of 28 percent n the case of the property tax on property wthout buldngs. Both tax rates ncrease as well contnuously over tme. Gven ths development of tax rates at the French regonal level, an explanaton on the bass of tradtonal tax competton models does not seem to be approprate. An alternatve explanaton s necessary n order to be able to explan such a race to the top.

5 Fgure 2: Tax Rates of the Local Busness Tax n Selected French Regons, 1984 to Basc Theoretcal Framework Such a model s developed n ths secton. The framework of the model conssts of two ursdctons provdng publc goods fnanced through local taxaton. In each of them, an agency mechansm descrbes the workng of collectve decsons. The decsve voter s the prncpal whle the local government s mandated as the agent. Rather than strctly focussng on the largely dscussed prncpal agent relatonshp, the model examnes how decsons wthn a gven agency are nfluenced by the observaton of what happens n the other communty. If the prncpals have the opportunty to compare the actons of ther respectve agents, how wll t shft ther strateges? How wll the agents take ths possble comparson nto account? What wll be the results of the correspondng game? In order to answer these questons, the frst step of the analyss conssts n the descrpton of the assumptons of the model (3.1). In ths settng, the agents wll develop strateges (3.2) whose nteractons wll provde the outcome of the game (3.3).

6 Fgure 3: Property Tax Rates (Wth Buldngs) n Selected French Regons, 1984 to Assumptons of the Model The frst assumpton s that of a non cooperatve framework. Nether the prncpals nor the agents can buld coaltons. In partcular, the agents cannot try collectvely to deceve the prncpals, for nstance by conscously shrkng n the same way so as to blur any sgnals of cheatng that the prncpals could catch. The latter, n the same way, cannot create a common agency by whch they would coordnate the control of the agents. The game has two stages, wth an electon takng place at each of them. The two ursdctons are represented respectvely by agents and and the prncpals wsh to obtan at each stage the same quantty of publc good g = g g. By assumpton, ths quantty s equvalent to = the publc expendture. To fnance t, each local government (each agent) reles on the taxaton of a base B that can take two values accordng to the current budgetary condtons. The frst one s low, B = B, and t corresponds to a negatve shock, whch may occur wth probablty p. The second value of the tax base s hgh, B = B, wth probablty (1-p) for ths postve shock. The agents do observe ths varaton n ther tax bases but the prncpals do not. However, the latter know that both ursdctons are smlarly affected by the shock whatever ts drecton.

7 Fgure 4: Property Tax Rates (Wthout Buldngs) n Selected French Regons, 1984 to 1995 Dstnctons between the two ursdctons thus reduce to tax rate polces t and t, whch helps concentrate the analyss on possble mmckng behavors. Agent, for nstance, elected n perod one, s by constructon of the model reelected n perod two f t t. He fals to be reelected f t happens that t > t. Snce both ursdctons face the same budgetary condtons, the opportunsm of agent s exposed whenever the agent n the other communty s honest. The range of values for the tax rates s assumed to be such that, n each ursdcton, t [ t, t ] wth g = t B = t B. The maxmum rate t allows ust to fnance publc expendtures g n unfavorable budgetary crcumstances, whlst good economc condtons make t possble to use the lowest rate t to fnance g. An external and exogenous authorty s assumed to have the capacty of controllng and preventng any attempt to pck a rate outsde [ t, t ]. We have seen that, elected n the frst perod, an agent can be reelected n the second one only f the tax rate he plays s lower than or equal to the one played by the agent of the other ursdcton. By assumpton, the number of mandates s lmted to two. Durng the second one,

8 8 havng nothng to lose, the reelected representatve s assumed to systematcally behave strategcally by always choosng the hghest tax rate t, whatever the budgetary condtons. F- nally, there s no dscountng from one perod to the other. If an agent ntends to maxmze hs expected gan over the two perods, then whch rate must he choose n the frst one? 3.2. The Agents Strateges We consder for nstance the strateges of agent who must determne the tax rate t that wll provde the best possble answer to a gven tax rate t from agent. Two cases may arse. Frst case: strategy of reelecton: t t ε wth ε 0 = Agent s n ths case reelected. Let EP 1 denote hs expected payoff over the two perods. It can be calculated as follows. In frst perod, the tax base takes ts low value wth probablty p for a gan of ( t ε )B t B. Wth probablty (1-p), the shock on budgetary condtons s favorable and the correspondng gan amounts to ( t ε )B t B. In the second perod, the agent s reelected but whether he obtans a strctly postve gan depends on the nature of the shock. In the stuaton of an unfavorable envronment, the representatve cannot take advantage of hs poston, even f he uses the upper value of the tax rate. Provdng g yelds t B but costs as much. On the contrary, a favorable shock provdes a postve gan of t B t B = t B t B = t B, ths wth probablty (1-p). On the whole, the expected payoff s: EP1 = p[( t ε) B t B (1 p) t B] (1 p)[( t ε ) B t B (1 p) t B] If the elected representatve maxmzes ths payoff, then he solves the program max EP1 = max EP1, whch mples ε = 0. He models hs behavor on that of hs counterpart t ε n the other ursdcton and thus obtans: EP1 = p[ t B t B (1 p) t B] (1 p)[ t B t B (1 p) t B] whch amounts to: EP = t EB g (1 p) t B 1 where EB ) = pb ( 1 p B represents the expected tax base n the communty. Ths strategy of reelecton can be contrasted wth the other opton consstng n playng straghtforward opportunsm from the frst perod onwards, even f ths leads to non reelecton n the second perod. Second case: strategy of non-reelecton: t t ε wth ε > 0 = At frst glance, t may seem surprsng that a poltcan would seek defeat n an electon. Everythng else equal, t s nevertheless a ratonal strategy f t brngs about a hgher expected

9 t Player s reacton curve t = t 9 gan than what the other opton can offer. If t t ε wth ε > 0, agent s not reelected and = he of course cannot expect any reward durng a second mandate. At the end of the frst perod, the payoff s: EP = p[( t ε) B t B ] (1 p)[( t ε ) B t B ] 2 Solvng max EP2 = max EP2, amounts to choosng ε = t t and hence the maxmum rate t ε t, whch brngs an expected payoff of EP2 = (1 p)[ t B t B ] = (1 p) t B. Those strateges are of course played non cooperatvely wth the other agent. The actual payoffs to the representatves then depend on the possble outcomes of these nteractons. Fgure 5: Nash equlbra of the game of mmckng t Player s reacton curve g / EB t - t - g / EB t t 3.3. Interactons and Mmckng We frst construct the reacton functons n order to compute the Nash equlbra of the game between the agents n a second step. Reacton curves The strategy of reelecton, namely the case where t t ε wth ε 0, ratonally leads to = mmckng. Should ths strategy be preferred to the second one ( t t ε wth ε > 0 ) where = the agent plays all hs cards durng the frst perod? In the latter case, the representatve cannot clam a second mandate and he ratonally chooses the maxmum rate t. The strategy of ree-

10 10 lecton s preferable f EP 1 > EP2, that s to say f t g EB. The elected agent n ursdcton wll thus adopt a mmckng behavor ( t = t ) f the rate set n the other communty s greater than or equal to the rate requred to fnance publc expendtures, takng nto account the expected tax base. On the contrary, f t < g EB, then EP 2 > EP1 and player should play the maxmum rate Nash equlbra t = t. As s llustrated n fgure 5, the Nash equlbra of ths symmetrcal game can be computed as follows. For values of the tax rate n the nterval [ t, g EB], the Nash equlbrum s n t = = t t. If ursdcton plays a rate t g EB maxmum rate t < ˆ <, then agent had better play the t = t snce he knows that the electors wll systematcally sancton any rate hgher than that of the neghbor. The reacton of agent then conssts n playng the maxmum ~ rate as well. If the latter were to play t such that g EB < ~ t < t, then agent would adust ~ hs own rate so as to obtan t = t. The computaton of the Nash equlbra of the game renforces the ntuton behnd the assumpton of mmckng. At least, ths assumpton cannot be reected. The combnaton of the reacton curves of the players dsplays an nfnte number of mmckng equlbra such that t = t for [ t, t ] [ g EB, t ]. Non opportunstc behavors represented by t = t = g EB do provde a Nash equlbrum whch conssts n playng the rate that fnances publc expendtures for the expected tax base, but t s only one of the possble equlbra. Moreover, ths honest equlbrum s domnated by all the others n terms of Pareto effcency (f the doman of valdty of ths crteron s restrcted to the set of players). Any move along t = t from = t = t g EB onwards s Pareto mprovng and leads to a unque Pareto effcent stua- ton n t = t = t. Ths provdes a strong ncentve to colluson, whch s precluded by the non cooperatve settng of the model but whch real-world games cannot fully prevent. 4. An Econometrc Analyss for French Regons The theoretcal model suggests that the payoffs of representatves n each regon depend on tax rates of competng ursdctons. In the case of the French regons, ths means that the level and the changes n the taxe d habtaton, the local busness tax and both property taxes n regon are a functon of the respectve tax levels and tax changes n the other regons. In the econometrc specfcatons used here, we assume that those taxes n regon are nfluenced by the (unweghted average of) taxes n geographcally neghborng regons only. 2 For n- 2. Wth ths specfcaton, we follow Case (1993) n her frst analyss of copycattng of tax polces. We also performed robustness tests as to whether the extent of tax mmckng would be reduced f we ncluded the wegthed average of tax rates of all French regons usng the nverse of the dstance between the regonal captals (n klometers) as a weght. The results wth respect to tax mmckng are largely the same.

11 11 stance, representatves of the Île de France consder the tax polces n the regons of Haute- Normande, Pcarde, Champagne-Ardenne, Bourgogne and Centre when they decde upon changes n tax rates. By takng the unweghted average of neghbors tax rates, we weght each neghbor of each regon equally. In our example, the regon of Pcarde does not have more nfluence on tax rate decsons n the Île de France than the regons of Bourgogne or Centre. In addton to neghbors taxes, a few other economc and demographc varables may have an mpact on tax polcy at the French regonal level. Regonal tax polces do not only depend on the capacty to generate regonal revenue by taxes, but also on grants from other ursdctons, n partcular from the French central level. Thus, we nclude natonal grants n the analyss. We do not nclude spendng varables or bond fnancng n order to prevent (addtonal) endogenety problems from emergng. Moreover, we ntroduce standard control varables lke average regonal household ncomes, the regonal unemployment rate as a proxy for busness cycle developments and the accompanyng necessty to ncrease socal welfare spendng and fnancng, and demographc factors, lke populaton densty, the share of the populaton younger than 20 years and older than 60 years. The estmaton equaton s: 3 t β t β tˆ β X = 0 t 1 1 t 1 2 t 1 ε, where t (t t-1 ) s a [NT 1] vector of N regons tax changes observed for T years (lagged by one perod); X t-1 s an [NT k] matrx of k observable regons economc and demographc characterstcs of the precedng perod, t ˆ t 1 s a [NT 1] vector of the average of regons geographc neghbors taxes of the prevous perod for T years and ε s an error term whch s assumed to be normally dstrbuted wth zero mean and constant varance. The parameter β1 ndcates to what extent a regons tax polcy s nfluenced by the tax polcy n neghborng regons. We estmate the model n a fxed effects specfcaton frst for the levels of the four dfferent tax rates and second for the dynamc model wth the changes of tax rates as the dependent varable whch s equvalent to subtract t t-1 from both sdes of the equaton and mposng the restrcton that the coeffcent of the lagged endogenous varable s equal to one. Table 1 contans the results for the fxed effects model of tax mmckng among 22 French regons. The model explans the tax rates of the taxe d habtaton, the local busness tax and both property taxes qute well. About 90 percent of the varance of the dependent varables are explaned by the model. The hgh value of the coeffcent of determnaton s not surprsng because the lagged endogenous varable s ncluded n the estmaton model and s hghly 3. See agan CASE (1993) pp Snce adequate nstruments are not avalable, we cannot perform TSLS estmatons although there clearly s an endogenety problem. On the one hand, a regon s tax settng behavor s nfluenced by taxes of neghborng regons. On the other hand, the tax settng behavor of neghborng regons s also nfluenced by the tax settng behavor of ts neghbors, one of whom s the regon under consderaton.

12 12 sgnfcant n each of the four equatons. The Durbn-Watson test statstc ndcates that there s no autocorrelaton left n the resduals. Three of the four tax rates are sgnfcantly nfluenced by respectve neghbors tax rates at the one percent sgnfcance level. The hgher the tax rate of the taxe d habtaton n the neghborng regons, the hgher s the tax rate of ths tax n the regon under consderaton. The same holds for the local busness tax and the property tax on property wth buldngs, but not for the property tax on property wthout buldngs. The sze of the coeffcents reflects that tax mmckng occurs to dfferent extents for dfferent tax rates. It s quanttatvely most mportant n the case of the local busness tax, followed by the tax on housng and the property tax on property wth buldngs. As to the other varables, usually one demographc varable has a sgnfcant mpact on tax rates, and n the case of the housng tax and the property tax (wth buldngs), average ncome has a sgnfcantly postve mpact at the one or ten percent sgnfcance level respectvely. Other varables, notably unemployment, populaton densty and federal grants do not have any sgnfcant mpact n any of the equatons. Table 1 about here The results of the dynamc model of tax mmckng reported n Table 2 corroborate these fndngs. The tax rates of the housng tax, the local busness tax and the two property taxes sgnfcantly ncrease f geographc neghbors ncrease ther respectve tax rates as well. Lke n the case of the statc model, the change of the local busness tax rate of the neghbors has the strongest mpact quanttatvely, followed by the housng tax and the two property taxes. Agan, other factors, lke changes n the demographc structure, n natonal grants, average ncome or n the unemployment rate, do not have any consstent mpact n the four equatons. Table 2 about here All n all, the results ndcate that the levels and changes n tax rates at the French regonal level are postvely nfluenced by the tax polcy n neghborng regons. Bascally, ths evdence s compatble wth tradtonal tax competton models as well as wth yardstck competton models. Gven the fact that tax rates of the four regonal taxes consdered n the analyss are steadly ncreasng over tme snce the decentralzaton, a yardstck competton model allowng for convergence n ncreasng taxes s more consstent wth the estmates. We thus obtan some ndrect evdence for the opportunstc Nash equlbrum arrved at n the theoretcal model. 5. Concludng Remarks In a settng of yardstck competton, voters compare the fscal polcy of ther government to that n other or neghborng ursdcton to evaluate the fscal performance of ther representatves. Ths may generate a convergence of fscal polces. The dfference wth tax competton s that such a copycattng may also lead to a convergence of tax rates at a hgher level. In ths paper, the yardstck competton model of Besley and Case (1995) s augmented by de-

13 13 fnng opportunsm here as one of the possble strateges that can be checked by comparson wth the behavor of agents n other ursdctons, nstead of opportunstc or bad behavors beng a consttuent of the nature of the agents. The theoretcal model mples that tax rates n a ursdcton under consderaton are nfluenced by tax rates n neghborng ursdctons and that ths may ether lead to a convergence to optmal tax rates or to neffcently hgh tax rates. Ths mplcaton of the model s tested econometrcally wth panel data of the 22 French regons from 1984 to The French decentralzed level s partcularly nterestng, because French régons, départements and local ursdctons have non-neglgble tax settng power snce the decentralzaton of Moreover, the relevant regonal tax rates have converged on a hgher level than the startng pont n each case. The econometrc panel results show that tax polcy at the French regonal level s determned by the tax polcy of states that are geographcal neghbors. If a neghborng regon ncreases tax rates, the regon under consderaton has the possblty to do the same. Together wth the overall convergence of tax rates at hgher levels, ths s ndrect evdence for an opportunstc behavor of representatves n competng regons when addtonal poltcal restrants lke an nspecton procedure n the form of referenda s not avalable to voters. References BESLEY, T. and A.C. CASE (1995), Incumbent behavor: Vote-seekng, tax-settng, and yardstck competton, Amercan Economc Revew 85, BRUECKNER, J.K. (2000), Welfare reform and the race to the bottom: Theory and evdence, Southern Economc Journal 66, BRUECKNER, J.K. and L.A. SAAVEDRA (1999), Do local governments engage n strategc property tax competton?, Unpublshed Manuscrpt, Unversty of Illnos at Urbana- Champagn. CASE, A.C. (1993), Interstate tax competton after TRA86, Journal of Polcy Analyss and Management 12, CASE, A.C., H.S. ROSEN and J.R. HINES JR. (1993), Budget spllovers and fscal polcy nterdependence. Evdence from the states, Journal of Publc Economcs 52, FIGLIO, D.N., V.W. KOLPIN and W.E. REID (1999), Do states play welfare games?, Journal of Urban Economcs 46, HEYNDELS, B. and J. VUCHELEN (1997), Tax mmckng among Belgan muncpaltes, Natonal Tax Journal 51, HOLMSTROM, B.R. (1982), Moral hazard n teams, Bell Journal of Economcs and Management Scence 13, LADD, H.F. (1992), Mmckng of local tax burdens among neghborng countes, Publc Fnance Quarterly 20,

14 14 SAAVEDRA, L.A. (2000), A model of welfare competton wth evdence from AFDC, Journal of Urban Economcs 47, SALMON, P. (1987), Decentralsaton as an ncentve scheme, Oxford Revew of Economc Polcy 3, SHLEIFER, A. (1985), A theory of yardstck competton, Rand Journal of Economcs 16, TIEBOUT, C.M. (1956), A pure theory of publc expendtures, Journal of Poltcal Economy 64, WILSON, J.D. (1999), Theores of tax competton, Natonal Tax Journal 53,

15 15 Table 1: Model of Tax Mmckng among 22 French Regons, 1986 to 1994, Fxed Cross Sectonal Effects (GLS) Dependent Varable Tax rate of the taxe d habtaton Tax rate of the local busness tax Property tax rate (wth buldngs) Property tax rate (wthout buldngs) Average household ncome (mllon French francs) 0.004** (3.17) (1.04) 0.002(*) (1.90) (1.64) Unemployment rate (-0.76) (-0.70) (-1.62) (0.20) Populaton densty (-1.10) (-1.15) (-1.61) (0.56) Share of populaton younger than 20 years (-1.05) (-0.50) * (-2.26) 0.390** (3.44) Share of populaton older than 60 years 0.052** (3.15) (1.21) 0.044* (2.25) 0.689** (6.59) Federal grants (n FF) (0.28) Lagged endogenous varable 0.511** (7.34) Tax rate of the taxe d habtaton of neghborng regons 0.144** (4.05) (0.34) 0.390** (4.34) (-0.26) 0.591** (8.73) (-1.54) 0.665** (11.21) Tax rate of the local busness tax of neghborng ursdctons 0.228** (5.11) Property tax rate (wth buldngs) of neghborng ursdctons 0.080** (3.10) Property tax rate (wthout buldngs) of neghborng ursdctons (-0.30) R SER D.-W d.f The numbers n parentheses are the absolute values of the estmated t-statstcs. **, * or (*) show that the estmated parameter s sgnfcantly dfferent from zero on the 1, 5, or 10 percent level, respectvely. SER s the standard error of the regresson. D.-W. s the Durbn-Watson test statstc on autocorrelaton of the resduals and d.f. are the degrees of freedom of the t-statstcs.

16 16 Table 2: Dynamc Model of Tax Mmckng among 22 French Regons, 1987 to 1995, Fxed Cross Sectonal Effects (GLS) Dependent Varable Change of the tax rate of the taxe d habtaton Change of the tax rate of the taxe professonnelle Change of the property tax rate (wth buldngs) Change of the property tax rate (wthout buldngs) Average household ncome (mllon French francs) (-0.73) (0.43) (-0.93) * (-2.59) Unemployment rate (-0.01) (1.08) (0.89) (0.17) Populaton densty 0.004* (2.03) (1.12) 0.004** (3.44) 0.034** (3.74) Share of populaton younger than 20 years (0.24) (0.76) (0.62) 0.294** (3.00) Share of populaton older than 60 years (0.88) (-0.11) (1.06) 0.471** (5.34) Federal grants (n FF) (-1.41) Change of the tax rate of the taxe d habtaton of neghborng regons 0.124** (4.22) (-1.20) * (-2.22) * (-2.05) Change of the tax rate of the local busness tax of neghborng ursdctons 0.274** (8.68) Change of the property tax rate (wth buldngs) of neghborng ursdctons 0.121** (5.99) Change of the property tax rate (wthout buldngs) of neghborng ursdctons 0.076* (1.99) 2 R SER D.-W d.f The numbers n parentheses are the absolute values of the estmated t-statstcs. **, * or (*) show that the estmated parameter s sgnfcantly dfferent from zero on the 1, 5, or 10 percent level, respectvely. SER s the standard error of the regresson. D.-W. s the Durbn-Watson test statstc on autocorrelaton of the resduals and d.f. are the degrees of freedom of the t-statstcs.