Abstract We show that federalism will lead to higher economic growth. We present a model of endogenous growth where government services, funded by income and capital taxes, are a component of production. In this model a decentralized government will choose tax policy to maximize economic growth, while a centralized government will not do so. Furthermore, these conclusions hold regardless of whether the government is beholden to a median voter or is a rent-maximizing Leviathan. However, a decentralized government will underprovide a consumptive public good.

Finally, we show our results are robust to imperfect capital mobility between districts and in such a model that districts with a lower total factor productivity will choose a more growth-enhancing tax policy. The author may be contacted at hat…[email protected] edu. We are particularly obliged to Paul Milgrom and Romain Wacziarg for their guidance during the course of this research. We are also grateful to Syed Nageeb Mustafa Ali, B. Douglas Bernheim, William Hauk, Jr. , Patricia MacriLassus, Ben Malin, Hui Li, Yuan-Chuan Lien, Daniel Quint, Antonio Rangel, Azeem Shaikh, and John Shoven for their helpful comments.

0 1 1 Introduction Determining the policies that increase economic growth is one of the foremost challenges for today’ economists. These questions are of enormous practical impors tance, as the welfare consequences of even small increases in economic growth can be tremendous. 1 And yet, …nding these optimal economic policies is only half the job. Economic policy is not decided by benevolent social planners, but by government o? cials, usually with at least one eye to their reelection prospects.

We consider the question of how di¤erent political institutions will result in di¤erent economic policies, and hence in di¤erent growth outcomes. In particular, we consider how a centralized or decentralized governance structure a¤ects equilibrium economic policy, and hence economic growth. decisions will enhance growth. We show that decentralizing government policy It is well understood in the literature that di¤erent political institutions may lead to di¤erent policy outcomes.

Persson and Tabellini (2000, 2003) consider many of these questions, such as the e¤ects of electoral regime, existence of a president, etc. from both theoretical and empirical perspectives. The question of how decentralized governance a¤ects economic outcomes, however, is far from settled. The change in economic outcomes from introducing interdistrict interaction was …rst discussed by Tiebout (1956), but we are concerned here with the narrower question of how tax competition between districts a¤ects these outcomes. There are two strands of literature in this area. The …rst, more theoretically based, largely espouses “race to the bottom” conclusions.

They argue that tax competition will drive tax rates on mobile factors, such as capital, to be too low, and hence district governments will underprovide consumable public goods (Zodrow and Mieszkowski 1986, Keen and Marchand 1996) and social insurance (Rom et al. 1998). The second strand of literature argues that tax competition drives the districts to choose better economic policies in their quest to acquire and keep capital.

Brennan and Buchanan (1980), the foremost proponents of this view, argue that federalism may restrain a Leviathanlike government;competition for capital may place a constraint on how much in rents states can acquire. 2 We strive in this paper to unify these views into a comprehensive whole. The key For further discussion of this point see Lucas (1988). There is also a large, more recent literature on the possible bene…ts of decentralization due to a reduction in political ine? ciences by devolving power away from the center: see Lockwood (2002) and Rubinchik-Passak (2005), for instance.

However, our focus is on how interdistrict competition for capital (by setting tax rates) leads to policy di¤erences. 2 1 2 di¤erence between models of the …rst type and those of the second is the motivations of the government; federalism acts as a constraint, which bodes ill if there is a welfaremaximizing government, but is useful in the presence of a rapacious one. Hence the welfare implications (in these static models) depend on the form of government incentives. In our dynamic model, however, federalism will drive district governments to choose the growth-maximizing policy, in a sort of “race to the top. ” A central government will not choose to maximize growth, regardless of whether it is Leviathanlike or responds to the wishes of the voters.

In the former case, it will collect positive rents, and in the latter, it will use tax policy as a lever for redistribution to the median voter. For these dynamic models, however, the question of welfare is more subtle and will be discussed in section 3. 3. The main positive conclusion, however, is that a decentralized government will result in higher growth outcomes, which is not a question that, to our knowledge, has been discussed in the theory of federalism literature heretofore. However, Qian and Weingast (1997) and Montinola et al. (1995) argue that decentralization has been key to China’ recent development.

s While this paper will consider the di¤erences of how a federal versus a centralized government a¤ects tax policy and economic growth, the work here could also easily be seen as an essay on some of the political economy e¤ects of international capital mobility. Both DeLong (2004) and Obstfeld (1998) consider issues of how the possibility of such capital ‡ ows may help to reduce incentives for corruption. The political economy e¤ects of international capital mobility, however, are more general than simply stemming corruption.

With international capital mobility, agents (and hence politicians) within a country are no longer directly concerned about internal rates of return on capital, but they must now worry about such issues as capital ‡ ight and foreign investment, and its e¤ects on wages. These e¤ects, again, can be seen as either bene…cial or harmful— witness the current debate in Europe over tax harmonization. A world with capital controls would be analogous to our model with a centralized government, while a world with international capital mobility would be analogous to our model with a federal government structure.

Hence, our paper could be seen as an argument that, with capital mobility, tax competition between nation-states will lead to higher economic growth. In order to examine the question of how centralized or decentralized provision of public goods a¤ects economic growth outcomes, we use a model of endogenous growth, …rst developed by Uzawa (1965), and later expanded by Lucas (1988) and Barro (1990), among others. Barro’ work is of particular interest to us, as it includes s a role for government, which provides a public good necessary for production, and 3 this public good must be provided for by taxation.

Alesina and Rodrik (1994) show that in the Barro model, a centralized government with access to one tax instrument, a capital tax, will choose a tax rate that is too high to maximize economic growth. In our work, we consider a model where governments have access to two tax instruments, a capital tax and a tax on labor income that is used to …nance a productive public good. We show that for a centralized government, even though the policy space is two-dimensional, a Condorcet winner exists and furthermore, in equilibrium, this policy does not maximize economic growth.

In contrast, the political equilibrium for district governments is to choose tax rates to maximize economic growth, and hence limiting the available tax instruments can never enhance economic growth. Our main contribution is to show that a decentralized government will implement a policy that results in higher economic growth. The next section of this paper establishes the economic model we shall use throughout the paper. Section 3 considers the equilibrium economic policy of centralized and decentralized governments beholden to a median voter, and consider the e¤ects of restricting the set of tax instruments available to the government.

Section 4 considers the provision of a consumptive public goods, and show that it is provided only within a centralized regime. Section 5 shows that the proposition that tax policy will be growth-maximizing if and only if government is decentralized holds for Leviathan governments as well. Section 6 considers externalities between districts. Section 7 considers the e¤ects of imperfect capital mobility and nonidentical districts. Imperfect capital mobility allows equilibrium tax policy to drift from the growth-maximizing policy, but a decentralized government will always induce higher growth than a centralized one.

Furthermore, districts with lower total factor productivity are likely to have a tax policy better for growth. The last section concludes. All proofs are in the Appendix. 2 2. 1 Economic Model Setup of the Economy We consider a growth model with three factors of production: capital (k), labor (l), and productive government services (g). Capital should be thought of in an expansive sense to include human, physical, and other types of capital. Government services can be thought of as funding for police, roads, protection of property rights, etc.

Our production function is y (t) = Ag (t)1 k (t) l (t)1 (1) 4 The single produced good is perfectly fungible: it can either be consumed or used as capital. We shall …x its price at time 0 at 1. We choose this model as it provides a role for government; the government must provide productive public services in order for production to occur. Government expenditures are …nanced through both a capital tax and a labor income tax that do not change with time. We shall assume that at each moment in time, the government’ budget is balanced, so that s g (t) = w (t) l (t) + k (t) (2) where w (t) is the wage.

While g is primarily a productive good, the government can also use g as a vehicle for redistribution, as its choice of tax policy can in‡ uence the after-tax returns to capital and labor. For instance, increasing the capital tax will increase the marginal productivity of labor by increasing the amount of the productive public good, but may decrease the after-tax return to capital. Since each agent has the same disutility function for labor, but di¤ering amounts of capital, the government can change the tax makeup and size of government spending to redistribute between agents.

Due to this form of government expenditures, we impose that 1 2 0 (18) Hence, using Topkis’ theorem, we can show that the labor supply for agent i is s increasing in !. Hence the total labor supply, l (! ), is increasing in !. Furthermore, @2U = @ i @li ( i ! 0 in the direction to improve the rate of return on capital, can ensure it has a higher rate of return than any other district. Hence it obtains all of the capital, causing a discrete positive increase in the wage of a worker in that district; it also allows the agents to obtain higher rates of return on their capital.

Since the e¤ect on the wage of the change in tax holding capital …xed is …rst-order in “, the …rst e¤ect dominates if ” is small enough, and so district n will wish to make this change in its taxes. Proof of Proposition 5: Suppose not. Then some district is not setting tax policy to (^ ; ^). Hence, unless no other district is setting this policy, this district is losing all of its capital, and all the agents in this district would be strictly better o¤ with the policy (^ ; ^), as this ensures them a positive wage.

However, following the same lines of proof as in the preceding proposition, some district must be o¤ering the policy (^ ; ^), and so all districts must o¤er this policy. Proof of Proposition 6: For the parameter values A = 10; for the unconstrainted, centralized government is ( m 1 = 2; = 7, 2 (l) = l2 2 and where all agents have the same starting capital, the growth rate in equilibrium ; m ) 2:73, while the con- strained centralized government equilibrium has a growth rate of (~ m ; 0) 3:38.

For the second part, note that district governments choose policy to maximize growth (where the proof follows as in Proposition 3), so restricting their domain can never enhance growth. Proof of Proposition 7: Exactly corresponding arguments to those in the previous section show that the among the set of points most favored by some voter (the e? cient frontier), m ; m ; 1+ is a Condorcet winner. For any ine? cient point, as before, m an e? cient point that all agents prefer can be found, so by individual transitivity of preferences the point of voters. ; m ; 1+ = is favored over the ine? cient point by a majority 1+ It is shown that is the policy most favored by all voters (and hence the median voter) in the text.

The tax policy in ( ; )-space is the same as in the previous section, and so will not maximize growth, just like in Proposition 2. Proof of Proposition 8: Suppose not. Then no district can be setting tax policy to (^ ; ^; 0), as otherwise all the capital would be in this district, since it is o¤ering 28 the highest rate of return on capital, since the rate of return on capital is constant with respect to the amount of capital, as the private rate of return is A ! ( ; ) l (! ( ; )) + (1 ) 1 l (! ( ; ))1 Now consider a district which is not receiving any capital.

By choosing a tax policy (^ ; ^; 0) the district becomes the one with the highest rate of return on capital, and so obtains all of its capital, making all of the citizens of that district better o¤, as their wages are now positive instead of 0, and they can get a higher rate of return on their capital. (They receive no consumptive public good in any event. ) If no such district exists, then all districts are o¤ering the same rate of return on capital. A district n, then, by changing its tax policy a small amount ” > 0 in the direction to improve the rate of return on capital, can ensure it has a higher rate of return than any other district.

Hence it obtains all of the capital, causing a discrete positive increase in the wage of a worker in that district; it also allows the agents to obtain higher rates of return on their capital. Since the e¤ect on the wage and the consumptive public good, holding capital …xed, of the change in tax is …rst-order in “, the increase in utility due to obtaining all the capital dominates if ” is small enough, and so district n will wish to make this change in its taxes. Proof of Proposition 9: For the …rst statement, if the government chooses it receives no rents, while > 0 ensures positive rents.

= 0, = 0 as the government does not receive any bene…t from providing the consumptive public good. Now, …xing and , consider the e¤ect the initial level of government consumption by increasing . @ ! l 1 + @ = l @! ! @l (! ) @! + +1 1 @ 1 @! @ @! @ However, from the proof of Proposition 2, we know that is positive. Hence, the initial level of government size is increasing in . If were then nondecreasing in , the government would be unambiguously better o¤ by raising . Since this violates optimality, ! l must be decreasing in , and hence lowering would increase growth.

Now consider the e¤ect of increasing + . This can only be decreasing in in !. Then, however, 1 on the initial level of government size if ! is decreasing in , as l (! ) is increasing ^ = A (1 )1 !( ; ) l (! ( ; )) + (1 ) 1 l (! ( ; ))1 29 is also decreasing in . Since the Leviathan will always choose a tax policy that is e? cient in terms of maximizing and 1 ! l + , given , it can not be that government size is weakly decreasing in . Hence, for the policy the Leviathan chooses to be e? cient, it must be that increasing decreases the growth rate.

Proof of Proposition 10: Suppose not. Then no district can be setting policy to (0; ^ ; ^; 0), as otherwise all the capital would be in this district, since it is o¤ering the highest rate of return on capital. Now consider a district n which is not receiving any capital. By choosing a policy such that ” > 0, and ( n, n; n ; 0) n to the policy that maximizes the private rate of return given = minn0 6=n f n0 g ” for some small the district becomes the one with the highest rate of return on capital, and so obtains all of the capital, and so the government now obtains positive rents.

If no such district exists, then all districts are o¤ering the same rate of return on capital. A district n, then, by decreasing n a small amount ” > 0, can ensure it has a higher rate of return than any other district. Hence it obtains all of the capital, causing a discrete increase in the g in that district. Since the direct e¤ect on government rents is …rst-order in “, the …rst e¤ect dominates if ” is small enough, and so district n will wish to make this change in its policy. Proof of Proposition 11: Proof given in text. Proof of Proposition.

12: For ” small enough, the median voter would like to change the tax policy from is converging to (^ ; ^). ( n; n ), as from the previous proposition, we know this It it maintains the Now consider a small change in an . same growth rate, it will keep the same amount of capital that it had before. For a small enough change in an , the other districts can change their policies so district n can not set its tax policy to take their capital. Hence, the district can choose a policy with the same growth rate and a higher ! , and since this makes the median voter better o¤, the district will do so. V>? m (? ) V.