Does Tax Competition Tame the Leviathan? (2019) PDF

Journal of Public Economics 177 (2019) 104037 Contents lists available at ScienceDirect Journal of Public E...

Journal of Public Economics 177 (2019) 104037 Contents lists available at ScienceDirect Journal of Public Economics journal homepage: www.elsevier.com/locate/jpube Does tax competition tame the Leviathan?夽 Marius Brülhart a, b, * , Mario Jametti c, d a Department of Economics, Faculty of Business and Economics (HEC), University of Lausanne, Lausanne 1015, Switzerland b Centre for Economic Policy Research, London, United Kingdom c Institute of Economics (IdEP), Università della Svizzera italiana, Lugano 6904, Switzerland d CESifo, Munich, Germany A R T I C L E I N F O A B S T R A C T Article history: We study the impact of tax competition on equilibrium taxes and welfare, focusing on the jurisdic- Received 3 August 2016 tional fragmentation of federations. In a representative-agent model of fiscal federalism, fragmentation Received in revised form 11 June 2019 among jurisdictions with benevolent tax-setting authorities unambiguously reduces welfare. If, however, Accepted 11 June 2019 tax-setting authorities pursue revenue maximization, fragmentation, by pushing down equilibrium tax Available online 9 August 2019 rates, may under certain conditions increase citizen welfare. We exploit the highly decentralized and heterogeneous Swiss fiscal system as a laboratory for the estimation of these effects. While for purely direct- JEL Classification: democratic jurisdictions (which we associate with relatively benevolent tax setting) we find that tax rates H2 increase in fragmentation, fragmentation has a moderating effect on the tax rates of jurisdictions with H7 some degree of delegated government. Our results thereby support the view that tax competition can be D7 second-best welfare improving by constraining the scope for public-sector revenue maximization. Keywords: © 2019 Elsevier B.V. All rights reserved. Tax competition Optimal taxation Government preferences Fiscal federalism Direct democracy 1. Introduction of the international policy agenda as well. The main opposing argu- ments are straightforward. Advocates of tax harmonization think of Is tax competition good or bad for the wellbeing of society? This governments as essentially benevolent maximizers of social welfare, has been a question of concern to federal states for as long as tax- whose ability to offer the optimum level of public goods is under- payers have been free to settle in whatever part of their country mined by the erosion of their tax base. Conversely, those who view they pleased. And as some lucrative tax bases have become highly tax competition as a force for good consider governments as self- mobile across national borders, tax coordination has risen to the top interested revenue maximizers, whose voracity may be constrained by tax competition. These are stock arguments in debates concern- ing tax coordination, such as on the taxation of e-commerce across 夽 We thank Wojciech Kopczuk (the editor), two anonymous referees, Robin US states, on harmonization of value added taxes and corporate Boadway, Sam Bucovetsky, Olivier Cadot, Bev Dahlby, Michael Smart, Thomas von Ungern-Sternberg, John Wilson, and workshop and conference participants at the taxes in the European Union, or on the definition of “harmful tax Institute for Fiscal Studies, the Paris School of Economics, and at the Universities competition” by the OECD. of Barcelona, Essex, McMaster, Queen’s (Kingston), St. Gallen, Toronto, Warwick and Research in this area abounds. Economic theory provides ele- York (Toronto) for helpful comments. We are grateful to Hansueli Bacher, Eleonora gant statements of the conditions under which tax competition Lanzio and Edoardo Slerca for excellent research assistance, and to Monika Bütler, Lars may be a force for good or a force for bad. Edwards and Keen Feld, Guido Koller, Raphaël Parchet and Rinaldo Signoroni for the generous provision of data. Financial support from the Swiss National Science Foundation (grants 147668, (1996), for example, show that the net welfare effect of tax com- 159348 and 166618) is gratefully acknowledged. petition hinges on the relative magnitude of two parameters: the * Corresponding author at: Department of Economics, Faculty of Business and marginal excess burden of taxation and the government’s marginal Economics (HEC), University of Lausanne, Lausanne 1015, Switzerland. E-mail addresses: [email protected] (M. Brülhart), ability to divert tax revenue for its own uses. Such parameters, [email protected] (M. Jametti). however, elude precise measurement. Empirical work has there- URLs: https://www.hec.unil.ch/mbrulhar (M. Brülhart)https://people.lu.usi.ch/ fore focused on indirect approaches, based on observable variables. jamettim/ (M. Jametti). The most prominent strategy, initiated by Oates (1972, 1985), is https://doi.org/10.1016/j.jpubeco.2019.06.005 0047-2727/© 2019 Elsevier B.V. All rights reserved. 2 M. Brülhart and M. Jametti / Journal of Public Economics 177 (2019) 104037 to study the relationship between government size and “decen- governments’ intrinsic desire to overtax and the vertical externalities tralization”, where decentralization is understood alternatively as pushing towards excessively high taxes.3 the share of sub-federal government in consolidated revenues or An empirical evaluation of this prediction requires extraneous expenditure (centralism), or as the number of sub-federal govern- information on the benevolence of government. Our second main ments (fragmentation).1 This approach draws its working hypothesis contribution is to exploit an empirical setting that allows us to from Brennan and Buchanan’s (1980) description of governments distinguish a priori between government objectives across juris- as revenue-maximizing Leviathans, whose tax raising powers could dictions. We compile a detailed new data set of local taxation in be held in check by decentralization. Negative partial correlations Switzerland, which offers a propitious laboratory for research on between government size and decentralization were therefore inter- tax competition thanks to the exceptional institutional diversity and preted as evidence in support of the Leviathan view, and, implicitly fiscal autonomy of Swiss sub-federal jurisdictions. With its three at least, of the conjecture that tax competition is a force for good. It hierarchically nested layers of government (central, cantonal and has come to be recognized, however, that this approach faces a major municipal), Switzerland can be considered a federation of federa- identification problem, because negative correlations between gov- tions, thus allowing identification from variation between as well as ernment size and decentralization are also predicted by a model of within federations (cantons). Another feature of the Swiss data is that horizontal tax competition among fully benevolent governments — they allow us to classify jurisdictions by the benevolence of their gov- in which case tax competition is welfare reducing.2 Hence, regressing ernments, where we take the intensity of direct-democratic control government size on decentralization does not allow conjectures on in matters of local taxation as a proxy measure for benevolence. We underlying government objectives or on the welfare consequences of thereby have empirical counterparts for all the variables that appear tax competition. Recognizing the interpretational ambiguity beset- in the theoretical prediction we wish to test. ting much of the existing empirical literature, Epple and Nechyba We estimate the impact of government benevolence on the (2004, p. 2463) noted that “the work stimulated by Oates addresses relationship between local tax rates and the relative smallness the issue of whether spending falls with increased competition, but of jurisdictions, controlling for differences in revenue needs, loca- does not address the issue of whether resources are used more effi- tional attractiveness and systemic idiosyncrasies either through ciently as competition increases”. Similarly, Wilson and Wildasin jurisdiction-level controls or, in the most demanding specifications, (2004) concluded their survey with the observation that “more work through jurisdiction-level fixed effects. Our central empirical finding is needed to incorporate reasonable political processes into tax com- is that, among jurisdictions with relatively benevolent governments, petition models, leading to sharper empirical distinctions between smaller ones set higher equilibrium tax rates, but that this relation- good and bad tax competition”. ship disappears in jurisdictions with greater scope for governmen- In this paper, we seek to advance towards that aim through two tal revenue maximization. Hence, our estimation results coincide main contributions. First, we address the difficulty of distinguishing with the theoretical prediction. Our empirical specification allows good from bad tax competition in a way that is tied rigorously to the us to interpret this finding as evidence that tax competition has a theory. We derive a reduced-form relationship which involves only moderating impact on equilibrium taxes because governments are observable variables and maps monotonically into welfare effects, Leviathans, and the underlying theory identifies this as evidence drawing on a model of fiscal federalism in the vein of Keen and Kot- of beneficial tax competition. We thereby overcome the interpreta- sogiannis (2002). In this model too, the difference between welfare- tional ambiguity of prior empirical work. improving and welfare-reducing tax competition hinges on largely Our paper contributes to some additional issues raised in the lit- unobservable structural parameters. However, we can establish the erature. One recurrent theme in empirical research following Oates following prediction: if the relationship between states’ “smallness” (1985) concerns the definition of “decentralization”, the metric for and the equilibrium state tax rate is positive for states that have the intensity of tax competition. We argue that fragmentation is relatively benevolent governments, and if, other things equal, this the appropriate measure: while, to the extent that governments same relationship turns negative for states that have less benevolent are benevolent, relative sizes of federal and sub-federal government governments, then the latter effect can be interpreted as evidence budgets are endogenous, the number of jurisdictions, and thus the of welfare-increasing “Leviathan taming” in these states. The intu- relative size of a representative jurisdiction, can more plausibly be ition is straightforward. The smaller a state j, the less it internalizes treated as exogenous with respect to citizens’ fiscal preferences.4 We the externalities created by its choice of tax rate on the tax base of therefore model the intensity of tax competition via differences in other jurisdictions, both horizontally (i.e. for the other states in the states’ smallness, in terms of their population share. federation, whose tax base shrinks if state j lowers its tax rate) and By allowing for fiscal interdependencies not only among same- vertically (i.e. for the federal government, whose tax base increases level governments but also among different hierarchically nested if state j lowers its tax rate). Smallness therefore exacerbates dis- government layers, our analysis furthermore takes account of the tortions created by externalities. Dominant horizontal externalities fact that the standard model of purely horizontal tax competi- lead to state taxes that are too low, while vertical externalities push tion is increasingly inappropriate as a framework for analyzing towards state taxes that are too high. If smallness is positively corre- non-coordinated tax setting in many real-world contexts. Both lated with state tax rates set by relatively benevolent governments, fiscal decentralization from national to sub-national governments this implies that the dominating externality pushes towards equilib- and (to an as yet lesser extent) delegation of fiscal competencies rium state taxes that are too high. If smallness is at the same time from national governments to supranational institutions are evident negatively correlated with tax rates among Leviathan states, all else global trends.5 equal, this implies that smallness (i.e. tax competition) must be a The configuration studied in this paper is therefore not specific good thing for the citizens of those states, as it countervails both their to the Swiss case. In general, vertical externalities are more likely 3 For a graphical representation of the different predicted relationships between 1 Important later contributions to this literature include Nelson (1987), Wallis and smallness and equilibrium tax rates, see Fig. 1 in Section 2.3. 4 Oates (1988), Zax (1989) and Forbes and Zampelli (1989). For a survey, see Feld et al. Fragmentation represents the standard approach for modeling the intensity of (2010). tax competition in theories of fiscal federalism and Leviathan governments (see, e.g., 2 This has in fact first been pointed out by Oates (1985, footnote 2) himself, as he Epple and Zelenitz, 1981; Keen and Kotsogiannis, 2003; Eggert and Sørensen, 2008). 5 stated that “other sorts of models besides Leviathan could produce such an outcome”. See Epple and Nechyba (2004) for a survey of the theoretical and empirical See also Hoyt (1991). literature on fiscal decentralization.  M. Brülhart and M. Jametti / Journal of Public Economics 177 (2019) 104037 3 to dominate the smaller is the sub-federal fiscal share. Average 2.1. A small open federation revenues of our sub-federal jurisdictions (municipalities) amounted to 69% of corresponding federal (canton) revenues in our sample We consider a federation with a central government and N ≥ 1 period 1990–2009, which is a relatively high sub-federal fiscal share fiscally autonomous sub-federal states j. These states are alike in all in international comparison.6 The scope for vertical externalities respects bar their size and their governments’ preferences. Each state should therefore be rather greater in many other federations. In addi- is populated by Mj investor-firms.8 Hence, the federation’s total pop- tion, even the “Leviathan” governments in our Swiss data are subject N ulation is given by Mj = M. The tax base is represented by a to direct-democratic controls via voluntary referenda, which means j=1 that elected officials still enjoy comparatively little leeway to pursue production factor K that is perfectly mobile among states as well as their self-serving aims. Other nations’ sub-federal jurisdictions likely between the federation and the outside world. K can represent any exhibit greater scope for both vertical externalities and revenue max- mobile factor, including labor, but for simplicity we refer to it as imization than Swiss municipalities, and hence our results imply that “capital”. there is even greater scope for Leviathan-taming tax competition in Investor-firms determine the within-federation, per-firm supply many other federations. (S) and demand (K) of capital. Firms use an identical concave produc- Finally, our study is related to a literature that seeks to establish tion technology F(Kj ), with F > 0 and F = c < 0 in the relevant how different democratic institutions shape policy outcomes. The range of Kj , implying that the slope of the demand for capital does not impact of direct democracy represents one of the key themes in this depend on the tax rate. The net-of-tax rate of return q is determined research area. In a comprehensive literature survey, Besley and Case in a federation-wide capital market. Capital is taxed by federal and (2003, p. 45) put the central insight as follows: “the possibility of state governments at rates T and tj respectively, with tj = T + tj. We initiatives forces greater agreement between voter preferences and denote the vector of state tax rates by t, with elements tj. The vector policy outcomes, assuming that representatives elected to the leg- of equilibrium state tax rates is denoted by t∗. islature have views that are out of step with the citizens at large”. Profit maximization determines per-firm capital demand Kj =   In the same vein, Matsusaka (2005) concludes that “direct democ- K q + tj , with Kj = 1/c. State j s aggregate capital demand is sim-   racy works”, precisely because it mitigates agency problems between ply Mj Kj. Rent pj = p Kj , defined as the difference between the voters and potentially Leviathan governments. Our contribution is value of production and the rental cost of capital, is distributed to to explore the effect of direct democracy on local taxation via its residents. interaction with fiscal externalities. This causal link has not, to our Each investor is endowed with e units of capital, of which Sj is knowledge, been studied before. invested within the federation and the remainder is invested in the The paper is organized as follows. Section 2 presents the theoret- rest of the world (ROW).9 Without loss of generality, returns in the ical model underlying our analysis and derives testable predictions. ROW are normalized to zero, which implies that q can take negative In Section 3, we discuss our estimation strategy and describe the values if the rate of return is lower in the federation than in the ROW. empirical setting. Regression results are reported in Section 4, and Preferences over private goods are given by Section 5 offers a concluding summary and discussion. private     2. The model Wj = u (1 + q) Sj + e − Sj + pj , (1) The theoretical framework informing our estimation strategy is a “small open federation” variant of the model developed by Keen and where u( ) is an increasing and concave function, implying a home Kotsogiannis (2002). We extend their model by allowing for hetero- bias in investment. Domestic and foreign incomes being perceived as geneous government objectives and state sizes while retaining the imperfect substitutes, differences in the rate of return between the assumptions that private agents hold identical preferences and that federation and the ROW can exist even with perfect capital mobility. there is a single mobile tax base.7 The investment decision implies per-investor capital supply Sj = After setting up the model we characterize its equilibrium. For S (q), which turns out to be identical across states. Capital supply the subsequent development, we can use the theoretical results from state j is thus given by Mj S(q). For analytical convenience, we of monotonic comparative statics in supermodular games, which assume that inward investment from outside the federation is zero.10 requires only a minimum of structure. Our analysis proceeds in three     Market clearing implies that j Mj S(ho) = j Mj K q + tj and deter- steps: i) we show that if (equilibrium) tax rates increase with intensi- mines the equilibrium rate of return in the federation. The effect on fied tax competition (smaller size of jurisdictions), then this situation the rate of return of a change in state j s tax rate (using Kj = 1/c) is is unambiguously welfare reducing for any type of government objective; ii) if we observe raising tax rates with more competition for relatively benevolent governments, it must also be that tax rates are falling for some less benevolent governments; iii) finally, if points ∂q Mj K  K =    = pj  < 0, (2) i) and ii) apply, we can then translate the tax effect of size into a ∂ tj Mk (S − K ) (S − K  ) welfare effect of size. k 6 Data from the OECD’s Fiscal Decentralization Database (own-source tax revenue). For comparison, in 2016 US state and local revenue corresponded to 49% of federal tax revenue, while the relative size of local tax revenue to state tax revenue amounted to 70%. The corresponding shares in Germany were 47% and 36% respectively. 7 8 The model follows a tradition of modeling tax competition as a simultaneous- We will use the terms “investor”, “firm” and “agent” interchangeably. move game among N symmetric states with constant returns, an immobile and a 9 In the interpretation of K as labor, e represents endowments of time which can perfectly mobile factor, and one-for-one technology for transforming private into pub- be “invested” in the labor market. Note that one might equivalently interpret the licly provided goods (Zodrow and Mieszkowski, 1986). Heterogenous government model in terms of an intertemporal savings decision, as in Keen and Kotsogiannis objectives and state sizes have been considered independently by, among others, Keen (2002), by relabeling “investment in ROW” as first-period consumption and “domestic (1998) and Bucovetsky (1991). Important theoretical precursors to the work of Keen investment” as investment for second-period consumption. 10 and Kotsogiannis (2002, 2003, 2004) include Wrede (1996, 1999), Dahlby (1996) and Allowing for two-way international investment flows would complicate the Fuest (2000). model without changing any of our qualitative results. 4 M. Brülhart and M. Jametti / Journal of Public Economics 177 (2019) 104037 M of taxation, however, also depends on the determination of federal where pj = Mj is the population share of state j. Similarly, the effect on q of a change in the tax rates of all states is taxes. We address this issue in the empirical analysis by controlling for T.  Mj K  Using Eq. (2), and the fact that p = ∂∂tp = −K, we can write ∂q  j K the first-order condition, evaluated at equilibrium and implicitly j q ≡ =    =  ∈ [−1, 0). ∂t Mj (S − K ) (S − K  ) determining state tax rates, as j Hence, ∂∂ tq = pj q. This implies that the change in the net-of- j   tax rate of return with respect to a change in one state’s tax rate is ∂ Yj    1      Hj ≡ = 1 − lj Sp q − Kj pj q + 1 + Cg Kj + tj K  pj q + 1 independent of the distribution of the federal population among the ∂ tj t=t∗ (1 + q ) j other states (as the distribution of pi=j , does not feature in ∂∂ tq ).      j +CG TK  pj (q + 1) + lj Kj + tj K  pj q + 1 = 0, (5) Publicly provided goods are produced with constant returns and distributed equally to all investor-firms, and no tax revenue is   wasted. This implies per-capita budget constraints gj = tj K q + tj 1    for the state governments, and G = M j TMj K q + tj for the federal 1 where 1+q = u (see Eq. (3)). government.11 This first-order condition implies Publicly provided goods enter agents’ utility function. Total indi- rect utility for an investor in state j can be written as          Wj = u ((1 + q) S(q)) + (e − S(q)) + p q + tj + C gj ; G , (3) ∂ tj K j  ∂ gj   =  = Kj + tj K  (q + 1) ≥ 0. (6) ∂t  ∂ t t=t∗   t=t∗ where C gj ; G is increasing and concave in both arguments. 2.2. Government preferences and citizen welfare At equilibrium, tax revenues increase with a symmetric rise in The existing literature identifies two polar cases: benevolent states’ tax rates, and overall changes in equilibrium state tax rates governments and purely revenue-maximizing (Leviathan) govern- have a weaker effect on capital supply than a tax increase by a single ments. We allow also for intermediate cases. This is captured by the state. Hence, state governments find themselves on the upward slop- following per-capita objective function of state governments: ing part of their Laffer curve, and state tax rates are monotonically related to changes in tax revenues and public spending. This implies     that our empirical approach based on tax rates is consistent with Yj = 1 − l j Wj + l j tj K q + t j , (4) Oates-type specifications, which use tax revenues or public spending as the dependent variable. with l j ∈ [0, 1]. For l j = 0, the government’s objective function coincides with Given the homogeneity of agents, social welfare is characterized the utility of the state’s residents, whereas l j = 1 represents a pure by Wj. Analysis of the symmetric version of this model (e.g. by Keen Leviathan. Hence, larger values of l characterize governments with a and Kotsogiannis, 2002, 2003) has shown that, except for knife-edge stronger taste for revenue maximization.12 configurations, independent state-level tax setting leads to socially State governments maximize Yj taking into account agents’ suboptimal equilibrium state tax rates: a symmetric change in all choices, factor-market clearing and the budget constraints. They hold tax rates can be welfare improving. The equivalent in our setup is a Nash conjectures over other states’ tax rates and the federal tax rate. marginal change in t. Using the fact that, for state j, the other states’ We do not model the tax setting of the federal government explicitly, neither in terms of objective function nor timing, but assume that tax rates enter the welfare function only indirectly via their effect on the federal tax rate is independent of the distribution of investor- q, we can express the effect of such a change as firms across states, which is sufficient for isolating the welfare effects of state taxes separately from the determination of T.13 In other words, our results hold conditional on T. The overall welfare effect ∂ Wj 1   = Sq − Kj (q + 1) + Cg Kj + tj K  (q + 1) + CG TK  (q + 1). ∂t ( 1 + q) 11 (7) This model abstracts from vertical transfers. In the empirical part, we shall take account of this by controlling for vertical fiscal equalization. 12 Keen and Kotsogiannis (2003) model the Leviathan by assuming that some exoge- nously given fraction of tax revenues is used for expenditure that benefits only the government itself. Adopting this modeling approach would not change our relevant For less than pure Leviathans (l < 1), subtracting Eq. (5) from results. Eggert and Sørensen (2008) represent Leviathans as pursuing vote maximiza- tion through rents offered to public-sector employees, who have a positive weight in Eq. (7) yields an expression that lends itself to economic interpretation: the social welfare function. It turns out that this leads to qualitatively equivalent con- clusions regarding the desirability of horizontal tax competition to those identified by Edwards and Keen (1996) and therefore to those implied in our model as well. Keen (1998) combines benevolent and Leviathan motives by positing the objective function ⎡ ≶0 ⎤ (1 − l ) v (q + t) + C(tK, G), where v() represents citizens’ utility from a private good,     >0 0 implies that tax rates are below the social optimum. The first set of brackets contains three terms. The first of these dominant vertical externalities), which can be gleaned from terms may be called a tax exporting effect, due to the fact that the correspondence of bracketed terms in Eqs. (8) and (9), and in this setting, unlike in the symmetric model, capital supply and 2. the intensity of Leviathan preferences (with stronger Leviathan demand in state j are not necessarily equal. This effect pushes equi- preferences weakening the relative importance of vertical librium tax rates above or below the social optimum, depending on externalities). S S whether Kj > (1+q ) or Kj < (1+q). The second term represents the horizontal tax externality, arising from the interaction among state Another useful comparative static result is the effect of the governments, and driving equilibrium tax rates below the social opti- Leviathan parameter l on equilibrium tax rates. This is given by mum. The third term represents the vertical tax externality, which ∂H ∇tl∗j = −Ht−1 ∗ Hlj. Again focusing on ∂ l j , we obtain results from the use of the same tax base by the federal and the j state governments. This effect pushes equilibrium tax rates above the social optimum. Finally, the second brackets contain what we  ∂ Hj 1      call the Leviathan effect, representing the deviation from optimal rev- =− Sp q − Kj pj q + 1 + Cg Kj + tj K  pj q + 1 enue collection induced by Leviathan government preferences. The ∂ lj (1 + q) j Leviathan effect implies that the higher is l j the greater is the scope      +CG TK  pj (q + 1) + Kj + tj K  pj q + 1 > 0. (10) for excessively high state tax rates. 2.3. State size Given that the second square bracketed term is positive, inspection of Eq. (5) shows that the term in curly brackets must be nega- While we cannot obtain closed form expressions for the equilib- ∂H tive at equilibrium, which establishes the sign of ∂ l j. Hence, an rium tax rates, we are nevertheless in a position to study the effect j increase in the Leviathan parameter, all else equal, always increases of a change in state size on the equilibrium state tax rate through the equilibrium tax rate. a simple exercise in comparative statics. State size is our (inverse) The relationship between state size and equilibrium tax rates is measure of fragmentation.15 We abstract from the impact of a small interesting in itself and can be measured empirically. However, we change in the size of one state dpj on the relative size of the other ∼ 0).16 ultimately strive for statements about welfare effects of tax com- states (dpi=j = petition. This requires that we establish a link between, on the one Let H denote the system of first-order conditions characterized ∂H hand, the observable relationship between state size and the equi- by Eq. (5), Ht the Jacobian matrix with element i, j equal to ∂ t i and j librium tax rate (the “tax rate effect of size”), and, on the other hand, ∂H Hpj the vector with i-th element ∂ p i. The expression for the vector the unobservable relationship between state size, the tax rate and j ∂t ∇tpj with elements ∂ pi is then given by ∇tpj = −Ht−1 ∗ Hpj. welfare (the “welfare effect of size”). Since relative state size serves j as an inverse measure for the intensity of tax competition, the wel- For our welfare analysis we need to impose some additional fare effect of size can be interpreted as an inverse measure of the structure. We assume, as in Devereux et al. (2008), that the marginal desirability of tax competition.18 rate of substitution (MRS) between S and g is constant in the neigh- The utility function Eq. (3) implies that welfare is not affected by borhood of equilibrium, which implies that tax rates are strategic pj directly but indirectly via the effect of pj on tj. The welfare effect of complements.17 Strategic complementary, and hence supermodular- a change in state j s size, conditional on T, can then be written as ity, allows us to use the monotonicity results of comparative statics in supermodular games, i.e. Theorem 4 (Monotonicity Theorem) of   Milgrom and Shannon (1994) and Theorem 2.3 of Vives (2000). In the ∂ Wj ∂ Wj  ∂ ti context of our model, these results imply that sign(∇tpj ) = sign(Hpj ), =. (11) ∂ pj ∂t ∂ pj or in other words, all elements of Ht−1 are negative. Recall also that i our results hold conditional on T. ∂t ∂t Concerning the sign of ∂ pj (and ∂ pi ), we can therefore concentrate Thus, the welfare effect of size is the product of (a) the derivative j j on of state welfare relative to the vector of state tax rates and (b) the tax rate effect of size, summed across all states of the federation.    It can be shown that, with purely benevolent governments, inten- ∂ Hj   S = 1 − lj − Kj q + Cg tj K  q + CG TK  (q + 1) + lj tj K  q. sified tax competition via smaller state size will reduce welfare, ∂ pj (1 + q) irrespective of the dominant tax externality.19 However, even the (9) parsimonious model studied here does not allow for general results on the welfare implications of state size for positive values of the This implies that the state tax rate may increase or decrease with Leviathan parameter.20 We can, however, identify a mapping from state size. The net effect depends on  14 ∂ Wj  The corresponding expression for pure Leviathansis: ∂ t l =1 =   j S   18 As we are focusing on the change in welfare induced by a change in state size, − Kj q − Kj + 1 − pj Cg K  tj q + CG TK  (q + 1) ≷ 0. ( 1 + q)       our results are consistent with the finding that, with horizontal tax competition and    >0 0. and the tax rate switches sign as l j increases. How is this possible? j j The intuition is as follows. The more Leviathan a state government, ∂ tj the less it cares about federally financed public goods, since federal Proof. If ∂ pj < 0, then strategic complementarity  ∂ ti funds are assumed to be distributed equally to all citizens without implies i ∂ pj < 0. Further, Eqs. (8) and (9) imply that transiting through state budgets. As a result, the vertical external-    S      ity loses force as l j increases. In the limit, for a pure Leviathan state, (1+q) − Kj q + Cg tj K q + CG TK (q + 1) < 0, and the the existence of the upper-level government is irrelevant to the rela- proposition follows.  tionship between smallness and chosen state tax rates. Hence, if the vertical externality is strong for a relatively benevolent state govern- To put this simply: if intensified tax competition implied by ment, leading taxes to fall in state size pj , the relative force of this smaller state size leads to higher equilibrium tax rates, then tax com- externality will be reduced if that state government becomes more petition is unambiguously welfare reducing. The logic of this result Leviathan, leading equilibrium state taxes to rise in pj. is as follows. If equilibrium tax rates rise as states get smaller, this Finally, we can derive Proposition 3, which maps Proposition 2 must mean that vertical tax externalities dominate the horizontal tax into welfare. This will be the main result informing our empir- externalities, as they are the only force pushing towards higher taxes ics, as it allows us to make welfare statements based on observed as states get smaller. Combined with the tendency of Leviathans to relationships between tax rates, state sizes and government types. overtax irrespective of state size, this implies suboptimally high state tax rates. Proposition 3. Suppose Proposition 1 holds for lj = l j < 1. Proposition 1 is the starting point for a unique mapping from the   ∂t  > 0 implies that ∂ p j  ∂W tax rate effect of size to the welfare effect of size for the specific case Then, ∂ pj  l j j lj =l j where we compare jurisdictions of which some have higher l j  s than others, and where the tax rate effect of size for the lower-l j juris- dictions is negative (i.e. Proposition 1 holds for the more benevolent Proof.  A sufficient  condition for this Proposition to hold is that  states). tj∗  > tj∗  ∀lj , j, which holds from Eq. (10).  lj =l j lj =l j The second step is offered by Proposition 2: Equilibrium tax rates are always higher for less benevolent Proposition 2. Suppose Proposition 1 holds for an interior value of governments, ceteris paribus. This implies that equilibrium tax rates ∂t l j. Then there exists a pivotal value l̃j above which ∂ pj > 0.lj is are the lowest, for any pj , when l j = 0, i.e. the purely benevolent j characterized by: case marks the lower bound. Hence, if the vertical tax externality dominates in a state under   relatively benevolent government, then if a decrease in this state’s S      size under a less benevolent government will lower equilibrium tax (1+q) − Kj q + Cg tj K q + CG TK (q + 1) l̃j =   , rates this decrease in state size increases welfare: tax competition S        (1+q) − Kj q + Cg tj K q + CG TK (q + 1) − tj K q will be welfare improving. This is illustrated in Fig. 1, which plots the deviation  of opt evaluated at equilibrium. equilibrium state tax rates from their optimum tj∗ − tj against Fig. 1. Smallness, Leviathan and equilibrium tax rates.  M. Brülhart and M. Jametti / Journal of Public Economics 177 (2019) 104037 7 different levels of smallness. When governments are purely benev- where j again indexes states, k denotes different federations, njk olent (l j = 0) and there is only one sub-federal state, the state represents smallness, and Xjk represents idiosyncrasies in revenue opt tax rate is optimal tj∗ = tj. Negative correlations between tax needs and tax-base elasticities. rates and smallness have sometimes collectively been interpreted A negative relationship between state tax rates and smallness as evidence of Leviathan taming. It turns out that taxes fall in could reflect (a) the dominance of horizontal externalities and rela- smallness irrespective of government preferences in all cases where tively benevolent governments, or (b) the presence of Leviathan state horizontal externalities dominate. In those configurations, increasing governments. According to the theory, a positive relationship would smallness (i.e. tax competition) can be a good or a bad thing, depend- in turn point unambiguously to dominant vertical externalities. opt ing on whether tj∗ is above or below tj. Traditionally, regressions If underlying state government objectives (l jk ) are measurable, of government size on decentralization were (at least implicitly) the natural cross-section empirical specification becomes: opt predicated on the assumption that tj∗ is above tj , but this is not something that can be ascertained empirically. Hence, the usefulness   tjk = b0 +b1 njk +b2 ljk +b3 njk ∗ ljk +b4 Tk +ct̄−j,k + Xjk d + ek +ujk , of the case where equilibrium tax rates rise in smallness for l j up to a pivotal level l̃j , but fall in smallness for l j above l̃j. In that case, (13) Proposition 3 states that smallness (i.e. tax competition) is an unam- biguous force for good for all lj > l̃j , as tj∗ is monotonically lowered where t̄−j,k is a weighted average of neighboring state tax rates, ek is opt opt by increasing smallness towards tj , but tj never falls below tj , a fixed effect that absorbs all unobservable time-invariant variation i.e. the equilibrium tax rate under dominant vertical externalities is at the level of federations, and ujk is a stochastic disturbance. never lower than the first-best tax rate tj. The estimated coefficient  b1 represents the (inverse of the) tax In terms of Fig. 1, Proposition 3 is equivalent to stating that, given effect of size for relatively benevolent governments (lj = l). For our either dominant externality, the lines representing equilibrium tax empirical purposes, we shall treat l as a dummy variable, setting rates over the support of smallness never cross. This is what we show the lower-bound value l equal to zero. If  b1 > 0, vertical externali- to hold in the proof to the Proposition, based on expression (10). ties dominate at l, and Proposition 1 applies. The coefficient  b3 then quantifies the differential effect of smallness on state tax rates for 2.4. Fragmentation “relatively Leviathan” governments (lj = l̄ > l). This will be our main coefficient of interest. The key variable driving the intensity of tax externalities in our According to Proposition 3, if b1 > 0,  b3 < 0, and b1 + b3 < 0, model is the relative smallness of states, whereas the related empir- we can infer that tax competition tames the Leviathan and increases ical literature uses two different exogenous variables, fragmentation social welfare. We call this “strong Leviathan taming”: stiffer tax com- and centralism. petition from increased smallness improves welfare in Leviathan Our definition of smallness can be taken as a measure of frag- states. Another possible parameter configuration is  b1 > 0,  b3 < 0, mentation, because, from the point of view of a representative state, but b1 + b3 ≥ 0. We refer to this as “weak Leviathan taming”. In this a fragmented federation implies relatively small states.21 The model case, stiffer tax competition from increased smallness is less harmful clearly shows that observed inverse relationships between tax rates in Leviathan states than in relatively benevolent states. and fragmentation are not sufficient to infer Leviathan governments. Some additional issues need to be considered in taking the theory However, it also offers an analytically rigorous version of the popu- to data. First, estimation of Eq. (13) requires variation in l jk and lar view that intensified competition from increased fragmentation in njk. This is most likely to be found in a comparison of multiple can “tame the Leviathan” (without constraining it excessively), pro- federations, which ideally should be similar to each other in all other vided that vertical externalities dominate when state governments relevant respects. Note that for the identification of strong Leviathan are relatively benevolent. taming, we do not need to observe the full range of l jk , but only What about the exogeneity of smallness? Jurisdictional some instances of ljk > l̃jk and some instances of ljk < l̃jk. Second, definitions may be endogenous with respect to taxation in certain the theoretical model assumes states to be identical except for their settings (Perroni and Scharf, 2001), especially in the context of size. Empirical estimation needs to control for relevant asymmetries single-purpose districts (Hoxby, 2000). Our analysis is based on a across states and federations, such as revenue needs, preferences for sample of general-purpose jurisdictions with historically predeter- public goods, locational advantages and federation-level tax rates. mined boundaries, such that jurisdictions’ size in geographic terms Hence, federation-level fixed effects as well as time-varying state- can reasonably be taken as exogenous. Smallness in population level control variables Xjk are included in Eq. (13). Third, we express terms, however, may in reality be to some extent influenced by tax all non-dichotomous variables in natural logs, so that the estimated rates. We address this issue in the empirics by considering smallness coefficients can be interpreted as elasticities. in terms of both population and area. Finally, our data are in panel format and therefore allow us to exploit time variation in the data. Indexing years by t, we can 3. Empirical setting estimate the following state-level fixed-effects (FE) model: 3.1. The regression model   tjkt = b0FE +b1FE njkt +b2FE ljkt +b3FE njkt ∗ ljkt +b4FE Tkt +cFE t̄−j,kt +fjk +uFE jkt , Eqs. (5) and (9) imply: (14)   tjk = f njk , ljk , Tk , −j,k |Xjk , (12) where the state-level fixed effects are written as fjk. Identification of the parameters of interest is therefore driven solely by the timing of 21 The empirical Leviathan literature has paid considerable attention to centralism, changes in smallness njkt and government preferences l jkt. i.e. the allocation of fiscal powers between the federal and state government levels. Unlike fragmentation, however, the degree of centralism cannot be considered as an 3.2. Switzerland: a laboratory for research on tax competition exogenous determinant of the intensity of tax competition. See e.g. Wilson and Janeba (2005) who study how the choice of t/T may be used strategically by the central government to minimize the distortions arising from the interplay of horizontal and Although the reduced-form predictions we seek to put to the vertical tax externalities. test could conceivably also be estimated on data for other federal 8 M. Brülhart and M. Jametti / Journal of Public Economics 177 (2019) 104037 systems, Switzerland presents a particularly propitious empirical Substantial evidence exists also of the heterogeneity in direct- setting. The Swiss fiscal constitution distinguishes three largely democratic institutions within Switzerland. Ladner and Fiechter autonomous jurisdictional layers (national, cantonal and municipal). (2012) provide a descriptive account of municipality-level institu- Each jurisdictional layer collects a roughly equal share of total tax tional variety, documenting how some municipalities “dispose of revenues.22 We will concentrate on the cantonal and municipal lev- a long tradition of direct-democratic involvement of citizens (... ), els. Direct taxation at both these levels of government encompasses whereas others rely on more representative forms of local democ- four conventional tax bases: personal income and wealth, and corpo- racy” (p. 438). This institutional diversity has been shown to affect rate income and capital. Personal income is by far the most important policy outcomes significantly. Based on canton-level data, Feld and tax base, accounting for over 70% of municipal and over 60% of can- Matsusaka (2003) and Funk and Gathmann (2011, 2013) find that tonal tax revenues. In contrast to many other countries, property direct democracy acts as a brake on public expenditure. The effects taxation is small even at the local level.23 Hence, local tax bases they find are economically substantial: Funk and Gathmann (2011), are for the most part highly mobile. Summary statistics are given in for instance, estimate that the existence of a mandatory budget ref- Table 1. erendum reduces canton-level expenditure by 12%. Frey and Stutzer Two institutional features make Switzerland particularly well (2000) report that, ceteris paribus, residents of more direct-democratic suited to our study. The three-tier fiscal constitution implies that cantons are happier than those of cantons with more strongly Switzerland can be considered as a federation of federations. We will delegated government. take cantons to represent the federations (k) of our empirical model, In our model, the choice of the median voter represents the while municipalities represent the states (i, j). Switzerland is divided socially optimal state-level response. If one were to allow for hetero- into 26 cantons, which in turn contain between 3 (Basel Stadt) and geneous voter preferences or incomplete information, however, the 404 (Bern) municipalities.24 welfare implications of direct democracy would no longer be so clear The second institutional feature we exploit is variation across cut. When voter preferences differ in nature and intensity, referenda municipalities and cantons in the intensity of direct-democratic can be inferior to cost-benefit analysis (Osborne and Turner, 2010) involvement in the tax setting process. Measures of this intensity or to logrolling in an elected legislature (Matsusaka, 1995). Informa- serve as our proxy for l j. We distinguish three categories: “assem- tional advantages on the part of the government could offer another bly” municipalities that set tax rates and budgets via show of hands rationale for delegated decision making (Maskin and Tirole, 2004), as at town hall meetings of the entire citizenry and are thus associ- could the presence of a large share of noise voters (Besley and Coate, ated with the lowest values of l j ; “referendum” municipalities, whose 2008). Hence, a perfectly informed and benevolent state government constitutions feature compulsory referenda on fiscal decisions above might opt for a different tax rate from that chosen through direct- opt certain thresholds and are associated with intermediate values of democratic voting. In such a setting, tj could still be interpreted as l j ; and a residual “Leviathan” category of municipalities where fiscal the median-voter choice in the absence of tax externalities, but not matters are largely under the control of elected executives. as the social optimum. Furthermore, b2 could conceivably be positive Our strategy is built on the assumption that decisions in munici- for reasons other than Leviathan government preferences (although palities with greater scope for direct-democratic participation in the we do not find that particularly plausible). These factors will have no tax setting process are more likely to correspond to the policy pre- bearing on our parameter of central interest, b3. A perfectly benev- ferred by the median voter, whereas less direct-democratic control olent and well informed state government might choose a different offers greater leeway to Leviathan governments. This assumption tax level from that preferred by its median voter, but there is no rea- has considerable theoretical and empirical support. Gerber (1996) son to expect it to react differently from its median voter to changes and Besley and Coate (2008) model how the availability of direct- in the intensity of tax competition.26 democratic instruments will push policy outcomes towards the pref- Several additional aspects of fiscal policy making in Switzerland erences of the median voter. Empirical work by Gerber (1999), Lutz correspond closely to the features of our theoretical model. The the- (2010), Matsusaka (1995, 2004) and Matsusaka and McCarty (2001), ory assumes perfectly overlapping tax bases. This is precisely true based on extensive analyses of US data, confirms that proposition.25 within cantons, where tax bases are determined by the cantonal tax law and thus identical. Even across cantons, tax bases are very sim- ilar, as the information to determine the national tax base is drawn from tax forms used to report to the cantonal authorities. 22 The theory furthermore implies full fiscal autonomy of sub-federal Over our sample period 1990–2009, revenue shares have remained fairly constant at some 30%, 40% and 30% for the national, cantonal and municipal government levels, states. This largely applies to Swiss cantons and municipalities. In respectively (Feld et al., 2010). spite of considerable harmonization of tax bases across cantons, 23 See Brülhart et al. (2015). cantonal authorities enjoy full autonomy in choosing tax rates. Most 24 These numbers refer to 1995. The total number of municipalities is in slow decline, cantons use the following procedure to set taxes. The cantonal tax law as micro-municipalities (some with populations below 100) are encouraged to merge. Since our sample includes 362 relatively large municipalities, such changes do not determines a tax schedule on the main tax bases. This schedule deter- affect our data. mines the level and progressivity of each tax instrument. The cantonal 25 Cases have been documented, however, where local-level direct democracy did authorities annually decide on a multiplier that shifts the base tax not work as modeled in our paper. Asatryan (2016), for example, finds that the schedule, determining the effectively applied cantonal tax. Most can- introduction of voter initiatives in German municipalities was associated with an tons fix a single multiplier across the major tax bases. Similarly, most expansion of local government size. This might be due to the fact that – unlike in the Swiss case – these initiatives were allowed only on non-budgetary issues, such that municipalities annually set a single multiplier, which, applied to the the fiscal implications often were not salient at the time of voting or not even borne cantonal tax schedule, determines the effectively applied municipal by the municipalities themselves. Asatryan’s (2016) result also points to the distinc- tax. This particular procedure implies that municipalities are heavily tion between initiatives, which can aim at both higher or lower public expenditure, constrained in their choice of the “tax mix”, and that tax setting and referenda, which generally serve only to block tax and spending proposals (see e.g. Matsusaka, 2004). Another diverging case has been documented by Hinnerich and authorities concentrate their decisions on tax bases with the highest Petterson-Lidbom (2014), who show that the introduction of town-hall decision mak- impact on tax revenue (i.e. personal income taxes). Reflecting the ing in Swedish municipalities after 1919 often led to capture by local elites and lower spending on public welfare than in municipalities with representative governments. The authors mainly attribute this outcome to the hierarchical nature of largely agrar- 26 ian communities in early 20th-century Sweden, allowing for “de facto” political power This might not be true if governments understood the implications of tax exter- of elites to have greater weight in town hall meetings than in (anonymous) elections. nalities for their decision while voters did not. Given the intensity of the public debate This mechanism is much less likely to be relevant in late 20th-century Switzerland. about tax competition within Switzerland, such a scenario does not appear realistic.  M. Brülhart and M. Jametti / Journal of Public Economics 177 (2019) 104037 9 Table 1 Descriptive statistics. Obs. Mean Std. dev. Min Mun or cant. with min. Max Mun or cant. with max. Municipal personal income tax rate (%) Married, median inc. 6519 3.74 1.26 0.37 Baar (ZG) 8.66 Menznau (LU) Married, median inc. (l = 0) 5549 3.79 1.28 0.37 Baar (ZG) 8.66 Menznau (LU) Married, median inc. (l = 1) 970 3.45 1.12 0.37 Zug (ZG) 6.83 Sierre (VS) Single, median inc. 6519 5.10 1.34 1.27 Freienbach (SZ) 8.84 Menznau (LU) Married, high inc. 6519 10.72 2.25 2.37 Freienbach (SZ) 16.01 Balsthal, Solothurn (SO) Smallness (%) 6519 97.07 4.65 62.82 Appenzell (AI) 99.96 Kilchberg (BL) Area based smallness (%) 6380 98.58 2.09 80.11 Urnaesch (AR) 99.97 Nidau (BE) Government objective (l) 6519 0.15 0.36 0 (Several) 1 (Several) Municipal controls Population (in thsd) 6519 9.85 23.43 0.10 Kilchberg (BL) 386.60 Zürich (ZH) Dependency ratio 6519 32.07 3.51 20.51 St. Moritz (GR) 45.77 Sumvitg (GR) Share of protestants 6519 40.71 21.08 2.14 Sumvitg (GR) 85.05 Kilchberg (BL) Share of foreigners 6519 16.20 8.49 1.37 Sumvitg (GR) 51.85 Renens (VD) Area (km2 ) 6519 19.56 28.87 1.30 Schönenbuch (BL) 203.90 Zernez (GR) Urban center dummy 6519 0.15 0.35 0 (Several) 1 (Several) Aggregate municipal fiscal equalization 6378 94,509 104,081 0 (Several) 413,348 ZH flows (in thsd CHF per canton) Cantonal personal income tax rate Married, median inc. 6519 3.43 1.15 0.46 Zug 8.51 Basel-Land Single, median inc. 6519 4.70 1.18 1.79 Schwyz 9.51 Basel-Land Married, high inc. 6519 10.34 2.36 3.35 Schwyz 16.39 Basel-Land Notes: Government objective (l) according to referendum definition. high degree of cantonal and municipal tax setting autonomy, tax sample of 365 municipalities.29 In 256 of those municipalities, taxes rates and schedules vary substantially across cantons and munici- are set by citizen assemblies, in 93 municipalities taxes are set by palities. For example, the highest municipal income tax rate on a the municipal executive, and 16 municipalities have changed system median-income single household exceeds the lowest one by a factor within our sample period. Of the 93 municipalities with tax setting of almost seven (see Table 1). by the executive, 43 have a compulsory referendum, 25 a voluntary The theory abstracts from interjurisdictional fiscal transfers. referendum and 25 have neither. Although vertical and horizontal fiscal transfers exist within cantons, We construct two alternative dummy variables, denoted by  l, they on average represent