Federalism As A Facilitating Practice For Governme Essay, Research Paper

Federalism as a Facilitating Practice for Government CartelizationMany federal policies – including grants, mandates, and tax expenditures – closely resemble the instruments used by cartels to enforce anti-competitive agreements. Is it possible that many federal policies exist not to correct for interstate externalities or enhance equity, but to inhibit interstate competition? The current paper has two main aims: First, to develop an original model of federal-state interaction and political behavior which allows for the possibility of anti-competitive behavior in spite of political and economic constraints upon politicians’ decisions; second, to conduct some preliminary empirical tests of the model of government behavior in general, and the cartelization hypothesis in particular. The model reveals that typical federal policies are well-suited for anti-competitive purposes, and tend to exacerbate existing monopoly power of state governments. The empirical work shows that there is good evidence that federal policy hurts competition, and that the general theoretical model may be profitably applied to other issues of federal-state interaction. I would like to thank my advisor, Anne Case, as well as Igal Hendel, David Bradford, Robert Willig, Harvey Rosen, Alessandro Lizzeri, Gordon Dahl, Sam Peltzman, Tom Nechyba, Tyler Cowen, Bill Dickens, and seminar participants at Princeton for numerous helpful comments and suggestions. The standard disclaimer applies. 1. IntroductionThere exists a large literature on tax competition, but very little work has been done on governments efforts to curtail tax competition; in short, for governments to collude rather than compete. At the same time, most of the literature on tax competition continues to look at inter-governmental competition through the lens of perfect competition when imperfectly competitive models would actually be more appropriate. This paper investigates the possibility that state governments in a federal system, modeled as imperfect competitors, could restrict competition and form an effective cartel with the assistance of the federal government. While private sector collusion has received much more research attention than public sector collusion, public sector collusion may actually be a more serious problem because the new entry of political parties or governments in response to monopoly rents is extremely rare and often impossible. In the model, the players are: one federal government, two federal parties, states, and competing state parties. Each period, politicians compete by credibly committing to a certain size of government during their term in office. Electoral constraints are the only limitation upon the federal parties choices, but state governments face not merely electoral constraints but must also cope with citizens ability to move away to another state if they find conditions unsatisfactory. In contrast to many models of government behavior (see Inman [1988] for a general survey), both political/electoral and economic/locational competition work imperfectly. Rather than assuming that government’s objective is to maximize the utility of the representative agent, the model assumes self-interested motivations of politicians, then investigates the extent to which the economic and political constraints upon politicians behavior give them incentives to satisfy citizen preferences. The model is rich enough to let us formally distinguish competitive from monopolistic behavior of state governments, and to understand how it is possible for democratically-elected federal governments to act contrary to citizen interests by encouraging state-level collusion. The next section briefly discusses the theoretical literature on inter-governmental competition, imperfections in the electoral process, and collusion in the public sector. It also surveys some of the attempts to empirically determine how competitively the public sector behaves. The third section presents a formal model of a federal system with both political and economic imperfections, then shows how this model allows for the possibility of federally-enforced collusion on the state level. The fourth section discusses several additional applications of the model. The fifth section presents some preliminary empirical tests of both the general model of federalism as well as the presence of federally-enforced collusion. The sixth section concludes the paper. 2. Related LiteratureThe literature on tax, regulatory, and other forms of competition between states within federal systems is well-developed. Tiebout (1956) famously analogized sub-federal governments to perfectly competitive firms. Most of the Tiebout literature applies his insight to the supply of local rather than state-level public goods, but many articles on the economics of local government also shed light on state-level issues; see Zodrow (1983), and Mieskowski and Zodrow (1989) for a thorough survey. Epple and Zelenitz (1981) develop a variant of the Tiebout model in which economic forces are unable to perfectly constrain local governments even as the number of jurisdictions becomes large. Gordon (1983) develops a detailed model of federalism which allows for several different kinds of inefficient strategic interaction between decentralized states; the recommended remedies include most of the common federal policies, such as grants and tax deductibility. For some other recent approaches to competition between governments, see Bell (1989), Sinn (1990), Case (1993), Case, Rosen and Hines (1993), Besley and Case (1995), Bachetta and Espinosa (1995), Bucovetsky (1995), Hochman, Pines and Thisse (1995), Hwang and Choe (1995), and Wilson (1995). The model developed here depends critically upon the existence of an imperfection in the electoral process. The specific imperfection used in the model is similar to that introduced in Lindbeck and Weibull (1987), and more recently in Grossman and Helpman (1996). Both of these papers assume that parties may easily alter their positions on some issues (such as budgetary issues), while their positions on other issues (such as abortion) are relatively fixed. This structure somewhat weakens the competition between the parties, allowing deviations from the preferences of the median voter to occur in equilibrium. The present model does not make use of the assumption of the existence of “uninformed” and “informed” voters used by Grossman and Helpman, and well as Baron (1994), but it is possible to interpret the results in the light of this bifurcation. There is an extensive literature on private sector collusion; the work on collusion and other monopolistic behavior within the public sector is far less developed. Brennan and Buchanan (1980) informally discusses the benefits of competitive federalism, and the risk that federal policy could facilitate collusion between state governments. McLure (1986) examines the arguments against the desirability of interstate competition, and argues that in general they are unconvincing. Benson (1990) informally analyzes the obstacles to bilateral collusive contracts between state governments, arguing that voluntary cartel agreements between both state governments and firms break down in the absence of an outside enforcer. In Benson’s analysis, the federal government acts as the outside enforcer that the states need, using tax deductibility and federal grants as its main policy instruments. Nechyba (1996) develops a general equilibrium simulation in which local governments require the aid of an outside enforcer (such as a common state government) to jointly deviate from property taxes to income taxes. However, in Nechyba’s model is it necessary to appeal to citizens’ exogenous perception of the unfairness of property taxation to explain the motive for joint deviation, because collusion significantly increases neither voters’ nor politicians’ utility. The empirical work which at least suggests the possibility of non-competitive behavior by governments is somewhat larger than the related theoretical literature. The most famous anomaly is the so-called “flypaper effect,” the tendency of federal grants to have a much greater effect on state expenditure than equivalent changes in personal income. Hines and Thaler (1995) provides a thorough summary of the literature and of the various attempts to reconcile the existence of the flypaper effect with competitive behavior by states and utility maximization by voters. Barnett (1993) shows that if there are matching grants and lump-sum grants, then there would actually be a “reverse flypaper effect”; i.e., lump-sum grants should produce a smaller increase in the size of government than an equivalent increase in personal income. Other recent work on the flypaper effect includes Ladd (1993) and Becker (1996). While the flypaper effect is probably the most empirically studied instance of alleged non-competitive behavior in the public sector, a number of papers attack the problem from very different angles. Filimon, Romer, and Rosenthal (1982) finds evidence that school districts have monopoly power which they use to increase spending on public schools. Oates (1985) tries to empirically test Brennan and Buchanan’s (1980) theory that government is too large; since Brennan and Buchanan claim that inter-governmental competition forces governments to more closely satisfy voter preferences, Oates reasons that countries with less concentrated political systems will have smaller public sectors. Oates finds that this prediction does not hold up, and concludes that Brennan and Buchanan’s “Leviathan” view of government is probably wrong; Nelson (1987) applies Oates’ empirical analysis to state and local government within the United States and finds that a positive relationship between concentration and the size of the public sector does in fact exist. Grossman and West (1994) tests the Brennan and Buchanan “Collusion Hypothesis” for Canada, and finds that collusion is a significant factor. Grossman and West’s paper also makes an effort to empirically distinguish collusion from correction for interstate externalities. 3. The Model3.1. The Structure of the GameThe players are state political parties, 2 federal parties, and a continuum of citizens whose number is normalized to 1. There are state governments and 1 federal government. Play is simultaneous, and in any Nash equilibrium:1. All citizens must reside in their most-preferred state, given state and federal policies. 2. All citizens must vote for their most-preferred party, both on the federal level and in the state in which they reside. 3. All political parties must offer platforms that maximize their expected utility. Sections 3.2 and 3.3 present and solve a general version of the model when the federal government does not care about the size of the state public sector. Section 3.4 adds to the general model the assumption that the federal parties have preferences about the state public sector and shows how this will generate cartel-enforcement behavior. 3.2. The General ModelA. Citizen Preferences and ConstraintsCitizens income must be divided between private consumption, , per-capita federal public goods consumption, , and per-capita consumption of the state public goods provided by the state in which the citizen resides, .(1)Total societal income is normalized to 1 and all citizens have the same income level. The price of private consumption is treated as a numeraire. The production bundle chosen must satisfy the per-capita budget constraint, so:(2)The functional form of the citizen of state k’s corresponding indirect utility function is assumed to be 3) indicates a given citizen s most preferred fraction of total income spent on federal public goods when ; indicates the most preferred level of spending on state public goods when . It is initially assumed that all citizens have the same values of and . is the utility level which a citizen would enjoy if actual policy exactly matched those he or she most preferred. is the utility that citizen c enjoys in virtue of residing in state k attributable to exogenous non-political amenities. is assumed to be independently distributed according to:(4) is c’s relative taste (which may equal zero) for federal party i versus federal party j; similarly, is c’s relative taste for party i versus party j on the state level. If the taste is positive, it indicates the amount of utility the individual would be willing to give up in order to be ruled by party i rather than party j; if the taste is negative, it indicates the amount of utility the individual would be willing to give up in order to be ruled by party j rather than party i. is uniformly distributed along the interval ; is uniformly distributed along the interval . is an indicator variable which is 1 if party i wins the election and zero otherwise; is an indicator variable which equals 1 if party i wins the election in state k, and 0 otherwise. Why would voters treat political parties as differentiated products? Several papers, including Lindbeck and Weibull (1987), Dixit and Londregan (1995a), Dixit and Londregan (1995b), and Grossman and Helpman (1996) offer interpretations of this sort of assumption. As Dixit and Londregan explain, “[V]oters are attached to parties for reasons other than their own receipts from tactical economic redistribution. For some, the reason is a strong attachment to a party’s issue positions, including such matters as international diplomacy and defense, or the balance between citizens’ rights and the needs of law and order; for others there are personal loyalties to the parties themselves.” (1995b, pp.6-7)According to the “ideological attachments” interpretation, parties take fixed but different stands on social (as opposed to fiscal) issues such as abortion, leading voters to treat parties as differentiated products. A voter may prefer party i’s position on abortion, but would consider voting for party j if its fiscal platform were sufficiently preferred over the fiscal platform of party i. Given this sort of preference, parties have some room to maneuver. In particular, one party may be able to win with certainty while offering a fiscal platform somewhat larger than the median voter would prefer. Perhaps the “fixedness” of the positions on social issues exists because it is impossible for a party to credibly commit to suddenly change its position on issues of strong moral principle. Alternately, it is possible to interpret the party preference as a pure taste, or an “historical attachment.” Such a preference is not essentially different from any other consumer brand preference for a good with no discernible “objective” advantage over its closest substitutes. This approach seems implicit in a number of articles in the political science literature, such as Gaddie (1995), Romero (1993), Geer (1991), King and Gelman (1991), and Brady and Grofman (1991). There seem to be many prima facie historical instances where voters tastes for parties as well as policies were substantial and had important political consequences.B. Federal Parties’ Preferences and ConstraintsIt is common to assume that the objective function of political actors is to maximize the utility of the representative agent. The next most common modeling strategy is to treat politicians as vote-maximizers; or, more generally, to assume that politicians maximize the probability of winning elections (see Wittman [1983]). In contrast, this model assumes politicians maximize their own power – that is, the size of government conditional upon their control of it – subject to political and economic constraints, and examines the conditions under which it is possible to derive the conclusion that politicians self-interest induces them to act in the interests of the representative agent. Two national parties {i,j} compete in an election held each period. The parties are assumed to have solved any internal principal-agent problems or other conflicts, so the behavior of the party can be satisfactorily summarized by a single utility function. The utility of party i is given by:(5)while party j s utility is:(6) is the per-capita spending of the federal government, and is the total output of the economy. As previously stated is normalized to 1, so on the federal level . is an indicator variable which is 1 if party i wins the election and zero otherwise, and . Electoral victory goes to whichever party gets a majority of votes. While some models assume that parties have different objective functions (typically, one wants a larger public sector than the other; see e.g. Alesina and Rosenthal [1995]), the current model gives both parties parallel goals: both parties would like the size of government to be as large as possible so long as they are running it; i.e., and . Parties campaign for smaller government not because they value smaller government as such, but because it increases their probability of winning sufficiently to offset the lesser value of a smaller prize. The standard assumption of decreasing marginal utility holds, so and . It is further assumed that : Controlling a government with no resources gives the same utility as being out of power. C. State Parties’ Preferences and ConstraintsWithin the federal system, there are states indexed by s={1,2,…,k,… }. In each of these states, there are two parties that compete in each election. Political parties {i,j} in state s maximize:(7)As with the federal level, voters have tastes over parties as well as policies, and a party wins the election if it receives a majority of votes from the citizens in its state. In addition to this electoral constraint, state officials face a “resource mobility constraint.” If the tax-and-services package becomes unattractive relative to other states, population or business leaves the state. For the federal government, the resource mobility constraint is assumed to be essentially irrelevant (because there are no close substitute countries to which the citizenry can relocate), so is constant on the federal level, but on the state level it is a function of policy. When governments are modeled as perfect competitors, it is natural to think of governments as “utility-takers” which must offer inhabitants a given level of utility in order to have any inhabitants at all (Stiglitz [1983]). That is, with perfect competition, governments operate with the constraint that:(8)Similarly, when inter-governmental competition is imperfect, it is natural to think of governments as “utility-setters,” so the number of inhabitants in a state is an increasing function of the utility offered by the state and a decreasing function of the offers of all of the alternative states. Thus, states act subject to the constraint that the number of inhabitants is a function of utility offered:(9)The heterogeneity of the citizen’s “state tastes” implies that state governments face a constraint of precisely this form, as the next section shows. 3.3. Solving the General ModelA. Citizens’ Optimal Voting and Location ChoiceSince citizens vote to maximize their own utility, a citizen votes for federal party i if:(10)and for party j otherwise. Similarly, a citizen votes for state party i if:(11)and for party j otherwise. A citizen’s locational choice is more complicated because multiple equilibria are possible. However, the equilibrium in which party i wins in all states, supported by the belief that , is particularly interesting to examine because it is the most electorally competitive outcome of the compete set. Other equilibria are possible, but this equilibrium can be proven to set an upper bound upon how effective state-level elections can be. Intuitively, if voters with opposite party tastes divide up into their own “safe districts,” then effective electoral competition may be absent in every state. , Faced with the choice between state 1 and state 2, a citizen then chooses state 1 iff:(12)It is convenient to define ; then the previous equation may be written:(13)In order to attract a given citizen to a state, it is necessary that the citizen s utility in that state exceed the utility available in all other states. A citizen selects state k iff:(14)Using the independence of each citizen s , we can derive the distribution of by multiplying their cumulative distribution functions. (15)Since the total population is normalized to 1, and all citizens income is equal, the fraction of the country s population in state k will be exactly equal to , the total income of the population of state k. Using (15), it is possible to derive the fraction of the population (and, by implication, the fraction of national income) that most prefers to reside in state k:(16)This can be re-written as:(17) Note that it is possible to take a first-order approximation of this function:(18)Substituting in this first-order approximation yields:(19)Evaluating the definite integral:(20)which gives the demand curve for residence in state k as (a) a decreasing function of the deviations from voters preferences for state government services in state k, and (b) an increasing function of the deviations from voters preferences for state government services in the remaining states. is just a sensitivity parameter which captures how “mobile” citizens are; notice that appears in the citizen s indirect utility function in (3). While the preference assumptions underlying the model may seem restrictive, what is crucial for the model to work is that states lose more population than they gain when they deviate from typical preferences. The citizens who are pushed out by immoderate policies must be greater in number than the immoderate citizens from other states that those policies pull in. B. Federal Parties’ Optimal PlatformsIn the standard median voter model of elections, much like in the Bertrand oligopoly model, the existence of two competitors is enough to ensure ideally competitive behavior. In the Bertrand model, duopolists find that economic competition forces them to price at marginal cost; in the median voter model, two competing parties find that electoral competition forces them to execute the preferences of the median voter. Most theorists find the result of the Bertrand duopoly game to be counter-intuitive, and have shown that it ceases to hold if the assumptions are altered. If one of the Bertrand duopolists has a cost advantage, or if the duopolists sell differentiated goods (Simon [1987]; Bester [1992]) then a two-firm market structure no longer produces a first-best outcome. This paper follows the second approach to model a similar imperfection in the electoral process. While it still uses the basic framework of the median voter model, political parties are modeled as “selling” differentiated products. The greater the extent of parties’ perceived differentiation, the more the electoral results can deviate from the preferences of the median voter.Suppose initially that the distribution of can be observed without error by the political parties, and that the median of ; moreover, assume initially that federal parties cannot affect state policies. Given a system of majority rule, it will not be an equilibrium for both political parties to offer the median preference (with identical preferences, the most-preferred point of the representative agent). , If , then party i wins with certainty if it plays ; due to its advantaged position it can still win with certainty (and increase its utility) even if it plays more than . Party i will want to keep increasing the offered level of government until:(21)where is the median value of . If party i were to set , then party j would be able to win by playing , reducing party i s utility to 0; if party i were to set , it could increase its utility because it could increase and still win the election with certainty. Given that party i is playing its optimal strategy, it will be impossible for party j to win, leaving it indifferent between all of its possible strategies. However, if party j offers any , then (21) will no longer be an optimal response for party i. In equilibrium, then, party j will have to set , and party i will set , leaving neither of them with an incentive to deviate.If , then the analysis reverses: party j wins with certainty by playing , and party i sets . This result has several attractive properties. First, the equilibrium size of the federal government depends positively on the preferences of the representative voter. Secondly, the equilibrium deviation from median preferences is positively related to two factors: , the parameter weighting the sensitivity of utility to deviations from the most-desired government size, and , which captures the magnitude of the political advantage of the dominant party. As there is no uncertainty in this simple model, the advantaged party invariably wins. It simply deviates as far away from median preferences as it can get away with without losing office. Adding the assumption of “product differentiation” to democratic elections with perfect information yields an outcome similar to the Bertrand duopoly game with cost advantages. The party with the greater political advantage always wins, but is constrained in its choices by the presence of the alternative, less popular party.The theoretical results of this formal model are quite consistent with Peltzman’s informal analysis of how government can continue to grow in spite of voters’ fiscal conservatism:The larger question is how political agents can persistently dissipate voter wealth; that is, why has government grown so much (and why is it fiscally progressive)?… One [possibility] is that the political costs of growing budgets are too weak to compel much restraint… These exercises suggest that incumbents can indulge in nontrivial spending growth before they risk a close call next election day. (1992, pp.358-359)C. State Parties Optimal PlatformsState parties maximize their utility subject to two constraints rather than one. The electoral victory constraint, however, is merely an inequality constraint which will not necessarily bind:(22) is the median value of in state k; is the median value of for the national population. It will be informative to first solve for the equilibrium when the voting constraint fails to bind in all states. One can think of this as the equilibrium if electoral competition in each of the states were completely ineffective or if there were only one political party. In this case, equation (7) would reduce to:(23)By inserting (20) into (23) and maximizing with as our control variable, we can derive the reaction function of state k for the no-voting equilibrium. (24)We can then solve for the symmetric equilibrium of the -state game:(25)Equation (25) has the “nice” properties that: and . In general, however, in the no-voting symmetric equilibrium, interstate competition is not capable of preventing the size of the public sector from exceeding the level most desired by voters. Even though citizens are mobile, the state governments are “selling” differentiated products, so the state governments possess the public sector equivalent of market power. This naturally leads us to ask whether voting can bring state governments behavior in line with citizen preferences. Intuitively, the advantaged party picks the value that optimizes , subject to the condition that not be so large that the party will be voted out of office. Thus, we can set up the following Kuhn-Tucker problem:(26)Note that and . The problem is complicated by the fact that while is exogenous, is clearly endogenous. If party i is in power in state k, people with high (positive) values of will move to state k, and drive out people with large negative values of . A technical appendix however shows that so long as , an equilibrium in which party i wins in all states and will exist. The candidate equilibrium supported by citizens’ belief that will be a true equilibrium. In consequence, can be substituted into the above equation. In addition to being tractable, this equilibrium will generally exhibit more effective overall political competition than any other possible equilibria; the results derived from this assumption about political beliefs set an upper bound on how well political constraints work. We can also set up corresponding problems for the remaining states. If all the election constraints fail to bind ( ) then there exists a symmetric equilibrium defined by equation (25). If all the election constraints bind, then we may again solve the model analytically for a symmetric equilibrium given by:(27) 3.4. Applying the Model to Federal Cartel EnforcementNothing in the model so far implies that the federal government will try to reduce the competitiveness of the state game; rather, the model is general enough to allow for this possibility. The imperfections on the federal level enable the federal government to restrict interstate competition contrary to the interests of the citizenry. The ability of citizens to move to other states gives state governments a motive to restrict competition. Salop (1986) explains that oligopolists frequently find it difficult to collude, but:The likelihood of successful co-ordination may be increased by the adoption of industry practices that increase oligopolists’ incentives to co-operate and reduce their incentives to compete, despite their divergent interests. Contractual provisions can add credibility to such tacit agreements, because they will be enforced by courts. Anti-trust commentators refer to such practices as ‘facilitating devices.’ (p.265)The same logic also applies to state governments, the main difference being that instead of avoiding the hostility of federal anti-trust enforcement, states may call upon the federal government to make collusion work. Until now we have assumed that federal politicians care only about the size of the federal government, and state politicians care only about the size of their respective state governments. (For a somewhat different treatment of federal-state interaction, see Dixit and Londregan [1996]). Yet in many federal systems, there are important links between the federal and state party systems. The party systems could even be viewed as vertically integrated firms with important economies of scope. The state parties share the burden of marketing and advertising not only their local party chapter, but also their affiliated federal party; they also invest considerable resources in the human capital of state-level politicians who eventually move on to federal politics. Given the benefits the federal parties receive from their related state parties, it seems plausible that federal parties would care about their welfare. Thus, the utility of the federal parties could instead given by:(28)Suppose further that the federal parties also make offers for , where is the symmetric equilibrium of the non-cooperative state game. They make expansions of the state public sectors incentive-compatible by using different policy instruments: The deductibility of state taxes is one way to subsidize state spending (as suggested in Feenberg and Rosen [1986], Feldstein and Metcalf [1987], and Benson [1990]). Federal grant programs are another, more flexible and quantitatively significant way to get the same results. State politicians often complain about spending mandated by federal rules; but possibly this is analogous to a member of a cartel bargaining for a larger quota. All of these policy instruments would be interesting to look at empirically, but theoretically what matters is that when becomes a control variable for federal parties, their constraints change. The advantaged federal party i faces a slightly altered electoral constraint:

(29)The advantaged party must still ensure that the utility loss of its own policies relative to optimal policies not exceed . The difference is that now the advantaged party has two policy instruments, the relative usefulness of which depends upon the pre-subsidy state equilibrium, so then the advantaged party would solve:(30)If the mobility constraint binds in all states as in equation (25), then for we may substitute ; if it is the electoral constraint that binds in all states as in equation (27), then we would set . In general this will have a standard solution with both control variables greater than zero. The winning federal party still uses part of its advantage to expand the size of the federal government, but uses the rest to reduce the competitiveness of state government behavior. Such behavior is particularly likely when (a) the constraint upon the federal party is weak (i.e., is large) and (b) the constraints upon the state parties are strong (i.e., is small, is large, and/or is large. Even though it has been assumed that the ruling federal party cannot discriminate in favor of states controlled by its affiliates, from (30) it can be seen that federal parties will be more inclined to push for expansion of the state public sector as (a) the number of states they control increases and (b) the federal parties’ concern for the welfare of their state affiliates increases. Note further that just as a federal government might use its position to reduce the competitiveness of the state game, so too might a state use its position to alter the competitive relationship of localities within that state. In fact, if localities within a state are very close substitutes for each other, then the net gain of cartelization for local governments would actually be markedly greater than the net gain for state governments. 4. Applications4.1. Citizen Welfare Under Alternative RegimesWith identical tastes for government services, evaluation of the welfare properties of different regimes from the perspective of the citizenry is fairly straightforward. The worst possible regime would be the one-party, one-state dictatorship. Unconstrained, the federal party could take everything. The resulting expression for utility would be:(31)This outcome is somewhat paradoxical; realistically, some fraction of income for private consumption and state public goods would be necessary to keep the citizenry alive. Still, the qualitative result that totally unconstrained government makes the representative citizen maximally unhappy is not unreasonable.The next regime to consider would have federal elections, but no state elections. The federal equilibrium would be given by equation (21), and the state equilibrium given by equation (25). Plugging into the indirect utility function of the representative citizen: (32)The outcome is preferable to the dictatorship along both dimensions, as might be expected. The voting constraint forces the size of the federal government closer to median preferences, and at the same time the threat of exit does the same for state governments. In the symmetric equilibrium, the threat of exit is a costless constraint on state behavior, because in equilibrium the right to exit is never exercised. In the more realistic asymmetric case, we would have to take this into account. Finally, we turn to the case in which both federal and state governments are run democratically. If the election constraint fails to bind, then citizens utility is no different than in (32). However, in the symmetric case in which all of the voting constraints bind, the representative voter s utility is necessarily weakly greater:(33)Would it be better to have mobility without voting, or voting without mobility? The answer to that question is indeterminate in this model. What does come out of this model, however, is the intuitive result that two constraints are at least weakly better than one. This gives a somewhat novel perspective on the economic benefits of federalism. The most frequently cited benefit of federalism is that it allows the satisfaction of diverse tastes for public goods. To take one example, Rose-Ackerman (1983) states:A multiple-government system has little normative appeal if everyone has the same tastes and incomes and if the government apparatus is controlled by voters. Then there are no gains from having more than one government unless a single government cannot operate as efficiently as a decentralized system. (pp.25-26)Yet in the present model, despite the assumption of homogeneous tastes, and without any explicit assumption of diseconomies of scale, choice in government still serves an important purpose. Citizens’ option to relocate helps to mitigate the welfare losses due to imperfections in the voting process. The Bertrand-like implications of the simple median voter model lead us to overlook the potentially important corrective role of inter-jurisdictional competition. 4.2. State Governments with Heterogeneous Citizen PreferencesThe cartel-enforcement model used the simplifying assumption of homogeneous tastes for state government services. This section reconsiders the results of the state game without voting when there are heterogeneous tastes for state government services. Generalizing the model in this way shows that this assumption is not driving the results. Let there be T distinct types, each with its own most-desired level of state government services, . Each of these types constitutes a fraction of the population, so:(34) . The “state tastes,” are assumed to be independent of taste for government services. Thus, parallel to equation (20), we have:(35)This simply gives the residence of type t in state k as a function of the squared deviation from type t’s most-desired level of public services in both k and in all alternative states. Recall that in the model with homogeneous tastes, ; by analogy, . Summing over all of the types, total residence in state k as a function of its policies is given by:(36)As before, it is useful to derive the equilibrium without voting, when mobility is the only constraint upon state government behavior. To derive the reaction function of state k for the no-voting equilibrium, we merely substitute (36) into (23) and maximize with as our control variable. This yields:(37)Combining terms, (37) may be re-written as:(38)It is convenient to define , the mean most-desired level of government services. Then (38) implies:(39)which is the reaction function for state k. It is now possible to solve for the symmetric equilibrium of the N-state game. Setting , (39) implies:(40)This equation looks remarkably like the original result for the homogeneous-taste model given in equation (25). The only difference is that with heterogeneous preferences, the mean most-desired level of government services becomes the deciding factor. The homogeneous taste model is just a special case of the heterogeneous taste model, because with identical preferences . It is particularly interesting that when elections constrain government behavior, the decisive factor is the median preference, but when mobility constrains government behavior, the crucial factor turns out to be the mean preference. Equation (40) has the “nice” properties that: and . Thus, as the number of states increases, and as the sensitivity of utility loss to deviations from most-preferred public service levels increases, competition drives states to satisfy the mean preference. While it might initially appear that different tastes will sort themselves into different states, this is not correct. The independence of the “state tastes” and the public service tastes essentially makes the model behave continuously. In order to get a Tiebout-like result, with people with different tastes clustering together, it would be necessary to eliminate the state tastes, or else alter the independence assumption. Caplin and Nalebuff (1991) and Ma and Weiss (1995) also discuss “mean voter theorems,” although strictly speaking the result here is a “mean mover theorem” since it applies to dictatorships with free migration rather than any sort of electoral system. Moreover, policy satisfies the mean preference with equality only in the limiting cases in which the term goes to zero. 4.3. Elections with Uncetainty About Political AdvantageAdding uncertainty to the election model makes the predictions considerably more interesting, especially for empirical testing. Assume that instead of observing with certainty, both parties observe only , where , and is the true value of the median party preference. Parties maximize their expected utility. As previously stated, and , where is an indicator variable equal to 1 if party i is in power and equal to 0 otherwise. Thus, the utility of controlling a powerless government is equal to the utility of being out of office. The expected utility of party i is therefore given by:(41)while party j’s utility is:(42)Party i wins so long as:(43) . Given and , and the assumption of normally distributed error terms, the probability that (43) will be satisfied is:(44) ,where the operator is the cumulative distribution function of the standard normal density. (41) and (42) may therefore be re-written as:(45)(46)In equilibrium, both parties must maximize their utility conditional upon the strategy of the opposing party. To find the optimal response conditional on its opponent’s behavior, party i differentiates (45) with respect to , while party j differentiates (46) with respect to . In consequence:(47)(48)Combining (47) and (48) yields the following equilibrium condition:(49)It is easier to interpret these equations with the aid of Diagrams 1-3. Diagram 1 illustrates the choice problem of party i given the behavior of party j. Party i’s platform choices lie along the x-axis; its probability of victory choices lie along the y-axis. The U-shaped curve centered at is party i’s “budget constraint”; points on the frontier are feasible, while points outside of the frontier are infeasible. Note that while a typical budget constraint has a negative slope for its entire range, this budget constraint is negatively sloped only to the right of . It is possible to draw a class of indifference curves on this diagram, indicating the “bundles” of platforms and victory probabilities which give equal utility levels; north-east is the direction of increasing utility. As is usually the case, the party’s optimum is shown by the tangency of an indifference curve to the budget constraint. Note that due to the fact that the budget constraint is only negatively sloped to the right of , the optimal point will definitely be tangent to the right of that point. What factors shift the budget constraint? There are two: the strategy of party j, and the value of . When party j increases the deviation of its platform from voter preferences, party i’s victory prospects improve for every offered platform. Similarly, if the value of increases, the entire budget frontier shifts upwards. Thus, “political advantage” turns out to be precisely analogous to wealth. Larger political advantage, like greater wealth in a standard choice problem, shifts the budget constraint out and thereby makes it possible to buy both a larger probability of victory and a larger offered size of government. Diagram 2 shows the same choice problem from the perspective of party j. Party j’s budget constraint shows the set of feasible platforms and victory probabilities given party i’s platform. It is rotated because the labels on the axes remain unchanged: the y-axis still gives the probability of party i winning, and the x-axis gives party j’s platform, with movement away from the origin indicating a platform with a larger public sector. South-east is party j’s direction of increasing utility. When j’s political advantage increases ( becomes more negative), its budget constraint shifts out; the same happens if party i offers a platform which deviates more from voter preferences. Diagram 3 combines the two diagrams in a manner somewhat analogous to an Edgeworth box. The two parties’ combined probability of winning must equal 1, and both parties must maximize utility subject to the behavior of their competitor. However, while the probabilities must sum to 1, the proffered platforms and face no such restriction. In equilibrium, both parties will have their indifference curves tangent to their respective “budget constraint” of available victory probabilities and levels of government. Both equilibrium points will lie along a horizontal line, since the joint probability of victory is unity, but in contrast to an Edgeworth box, there is no need for the equilibrium points to be identical. The resulting equilibrium will not in general be Pareto optimal from the point of view of the two parties. Electoral collusion could be quite profitable for them. It would allow them to both move further to the right on the horizontal line along which their respective chances of winning lie. Even without collusion, however, Nash behavior gives no assurance that both parties will offer ; to the contrary, with standard preferences both offers will exceed . 4.4. Changing Assumptions About Political Parties’ Objective FunctionsThe maintained assumption of the preceding model of government behavior is that both parties’ utility increases as the size of the public sector increases. For empirical purposes, it will be useful to briefly examine how the results of the model would change if the assumption about the parties’ objective function were altered. That way, it will be possible to contrast the empirical performance of the model with some clear alternative hypotheses. A. Parties With No Policy PreferencesSuppose that the advantaged federal party had no policy preferences whatever, but simply desired to win:(50)Similarly, suppose the advantaged state parties’ objective function were given by:(51)The equilibrium would then be extremely simple. Regardless of imperfections in the electoral and mobility constraints, the state-level result would be . So long as it observed without error, the advantaged federal party i would be indifferent between all such that . Adding uncertainty eliminates this indeterminacy, so the advantaged federal party sets . Without policy preferences, there would also be no reason for the federal parties to try to increase the size of the state public sector. B. Parties that Maximize the Size of the Private SectorAlternately, suppose that the objective of the advantaged party were to maximize the size of the private sector. Then for the federal party:(52)For the advantaged state parties:(53)The result in equilibrium would be that , while the size of the state public sector would be:(54)if the electoral constraint does not bind, and if it does. Note that the limiting tendency of as persists regardless of parties’ objective functions. The equilibrium result in the game with uncertainty is predictable: just as a party that gains utility from more government makes a tradeoff between the probability of electoral victory and increasing the size of government beyond , a party that gains utility from a larger private sector trades off between the probability of electoral victory and reducing the size of government below . By considering the implications of changing the party’s objective functions, it can be seen that there is a bias towards excessive government only if parties value a larger public sector. If parties value a larger private sector, there is a bias towards insufficiently large government. If parties have no policy preferences, then there is no bias at all. Similarly, a monopolist only has a bias towards excessive prices if it maximizes profits; if it maximizes consumers’ surplus, prices tend to be suboptimally low; and if it weights producers’ and consumers’ surplus equally, prices will be exactly optimal. In both cases, monopoly power can yield optimal outcomes given the right objective function; the main difference is that private firms’ desire to maximize profits is uncontroversial, whereas there is extensive dispute about political parties’ goals. The next section looks at which objective functions best fit the data on political parties’ behavior. 5. Empirical Examination5.1. Empirical ImplementationThe general model of federal-state behavior allows for the possibility of cartel-like behavior, but does not necessitate it. In consequence there are two interesting areas to look at empirically: (i) how well the cartelization view explains the purpose and effects of federal policy, and (ii) how well the general model explains government behavior overall. The general model could still be empirically interesting even if it turned out that federal policy did not reduce interstate competition. The tests requires two points of explanation. First, we are testing the model with uncertainty as explained in section 4.3. Political advantage is not directly observable; but with political advantage of uncertain magnitude, parties treat their advantage as wealth which they may “spend” on both increasing their probability of ruling and increasing the size of government conditional upon ruling. So long as both of these are normal goods and the degree of risk-aversion is constant across parties and over time, there will be a positive relationship between the probability of winning and the size of government. Second, the probability of electoral victory is also not directly observable, but there are several plausible proxies (e.g., fraction of seats held or fraction of votes won) that could be used in state elections. These preliminary tests use the fraction of seats held by a party in a given year. For reasonably large political bodies (such as both lower and upper houses in state legislatures), the fraction of seats should be a good proxy for the probability of victory selected by individual candidates.This approach gives rise to a number of recurring political variables. #Dem and #Rep are simply the raw numbers of Democratic and Republican legislators in a given legislative body. Dem and Rep are dummy variables: Dem= 1 if # Dem/(#Dem+#Rep)|.5, and otherwise =0, while Rep=1-Dem. This variable allows for the possibility of a discontinuous effect of a change in the party controlling a given legislative body. Second, there is the Distance variable, which is defined as Distance . Distance is the absolute value of the difference between 50% and the percentage of seats held by the ruling party in a legislative body. (A few earlier studies of federal spending, including Wallis [1996], Grossman [1994], Anderson and Tollison [1991], and Wright [1974] use somewhat similar variables). Finally, Distance is interacted with Dem and Rep to yield Dem*Distance and Rep*Distance to allow the effect of Distance to differ between parties. The sequel uses data for the contiguous United States over the 1950-1989 period to test both the viability of the general model and the effects of federal policy on interstate competition; unless otherwise stated, fiscal data is expressed in per-capita real terms, and specifications include state and year dummies and control for personal income. 5.2. Null and Alternative Hypotheses TestedThe model’s primary predictions are that:+ For the general model: Greater political advantage increases the size of the public sector by equal amounts for both political parties. While the model uses a balanced-budget assumption where spending and taxation are always equal, for empirical purposes it is interesting to look separately at the effect of political advantage on both measurements of the size of the public sector. + For the cartel enforcement application: Federal grants have a large positive effect on state spending. Since there are many other explanations for this frequently-observed comovement (see Hines and Thaler [1995]), we also look at the effect of grants on tax collections, which is difficult to reconcile with most of the alternative explanations. The bulk of the model builds upon the assumption that both parties share the desire to make the public sector as large as possible (conditional upon electoral victory); this will serve as our null hypothesis for testing purposes. However, section 4.4 worked out the implications of alternative assumptions about parties’ motivation. If the data do not support the null hypothesis, section 4.4 suggests some alternative hypotheses to consider. The table below summarizes the key features of the null hypotheses and two plausible alternatives. HypothesisDem Taste forGovernmentRep Taste forGovernmentComovement Between G and P(Victory)Null: “Leviathan”+++,+Alternative I: “Ideologue”+-+,-Alternative II: “Technocrat”000,0″Leviathan” is the null hypotheses that both political parties share the same propensity to increase the size of the public sector when feasible. Then there are at least two interesting alternatives with different predictions. The “Ideologue” hypothesis states that competing parties have opposite preferences – one wishing to make the public sector as large as possible, the other the private sector as large as possible. The “Technocrat” hypothesis is that neither party has any tastes about policies. Insofar as the data are inconsistent with the null hypothesis, we can consider the performance of the alternatives. 5.3. Baseline ResultsThe first block of regressions looks at the effect of the political variables and federal grants on sales, income, corporate, and total taxation, and total state spending. To check the sensitivity of the results to specification, we look at the results using both the lower and the upper houses of the state legislature to measure the Dem, Rep, and Distance variables. Table 1 shows the coefficients and standard errors of the interesting variables. Some of the notable results include:+ Sales taxes are almost completely insensitive to political variables.+ Income taxes and total taxes appear to be highly sensitive to political variables; increases in Dem*Distance increase income and total taxes in a statistically and economically significant way: if the Democratic majority in the lower legislative house were to increase from .51 to .61, income taxes increase by about $15 real per-capita (about a 13% increase in the average observation) , while total taxes increase by about $20 real per-capita (about a 4% increase in the average observation). + A larger Republican majority usually (insignificantly) increases the tax burden. The effect of Rep*Distance on sales, income, and total taxes is almost always positive but insignificant. + Democratic predominance always tends to increase corporate tax collections, while Republican control tends to decrease such collections most of the time.+ The effect of grants on spending is extraordinarily large, even greater than most previous research on the so-called “flypaper effect” typically finds. The coefficient of 1.227 on grants in the first regression of spending on grants indicates that every $1.00 of federal grants stimulates about $1.23 worth of additional spending. Most previous estimates of this effect are large, but they rarely exceed unity by such a large amount. In spite of this big effect, this estimate may actually be biased downwards. The data on grants combines the quantity of federal grants to states with federal grants to localities within a state. Thus, the level of grants directly to the state is overstated, so the effect of a given level of grants is understated. It is also interesting that a large effect of federal grants on state spending is accompanied by a large and positive association between federal grants and state tax collections (see e.g. Inman [1989] for some representative earlier results). The conventional analysis is that states pass along at least part of grants to citizens, but in fact grants seem to increase rather than cut tax collections. (Since a $1 increase in grants tends to increase spending by $1.23 and taxes by $.28, grants tend to slightly reduce the budget deficit, as one would expect). The null hypothesis postulates identical objective functions for both political parties. The results in Table 1 suggest that parties are less polarized about their preferred size of government than many models of party motivation assume (e.g. Dixit and Londregan [1996]). Still, several differences stand out: e.g., Republicans cut corporate taxes when they have a supermajority, even though they do not significantly cut overall taxes. Are there other important compositional differences between the parties? Table 2 examines this question by splitting up spending into five categories: education, health and hospitals, highways, public welfare, and other spending. Several unexpected patterns in the disaggregated spending data appear:+ Neither party shows any tendency to increase education spending when its political dominance increases. + Republicans actually seem more inclined than Democrats to increase spending on health and hospitals; the former consistently and significantly increase spending on health and hospitals, while the latter have no clear spending pattern in this category. + A third – and completely unexpected – result is that Democrats drastically cut highway spending when their political dominance becomes greater, while Republicans slightly increase highway spending. + Public welfare spending fits conventional ideological stereotypes: Democrats greatly increase public welfare spending, while Republicans slightly reduce it. + “Other” spending shows a marked tendency to increase when Democratic power increases, but no clear tendency to change one way or the other when Republican power increases. The parties’ divergent spending patterns merit more intensive study, especially when the data show that ideological stereotypes seriously oversimplify the facts. The Leviathan hypothesis wrongly assumes that parties share identical goals, but the alternative Ideologue hypothesis seems to place too much emphasize on spending levels rather than spending composition. 5.4. Additional ResultsA. Controlling for Governor’s PartyThe empirical results in Tables 1 and 2 only look at the effects of the party composition of lower and upper legislative houses. This is quite unsatisfactory, because there are numerous other political variables with which legislative party composition might be correlated, leading to biased results. This section looks at how the baseline results change after controlling for the party affiliation of the governor. Governor=1 if the governor of a state in a given year is a Democrat; Governor=0 otherwise. Tables 3-4 show what happens when Governor is added to the baseline regressions. The main result is that the party of a state’s governor appears to make very little difference for budgetary policy; the coefficients on Governor are rarely statistically significant, and in any case never exceed $10.21 in real per-capita terms (about 1% of the mean value of real per capita spending). There are other margins along which the party of the governor matters. Democratic governors appear to reduce corporate taxes in an economically small but statistically significant way in every specification. Democratic governors also seem to increase total spending in every specification, although this increase is not usually significant at conventional test levels. The presence of a Democratic governor slightly increases spending on education and “other” spending in every specification, and slightly decreases public welfare and highway spending in every specification. These effects are small even if statistically significant: $7 real per-capita dollars on education, $9 on “other” spending, $3 on highways, and $5 on public welfare. B. Expressing Fiscal Variables as a Percentage of Personal IncomeThe baseline specification sometimes implies implausible restrictions upon the magnitudes of the coefficients. This specification requires, for example, that the absolute effect of various political variables be constant over time, even though the absolute size of the economy is constantly increasing. To alleviate this concern, the baseline regressions were re-run, but fiscal variables were expressed as a percentage of personal income rather than in absolute levels. The interpretation of these results, shown in Tables 5 and 6, may also be more intuitive than the interpretation of the coefficients in the baseline regressions. Once again, the overall results are resilient to the specification change. Since both grants and the taxation and spending variables are expressed as a percentage of personal income, the numerical value of the coefficients on Grant should not change much compared to the baseline model; they do not, except that the influence of grants upon state sales taxes increases sharply. The main change in the other variables is that the coefficient on Rep*Distance becomes slightly more positive. + The sign on Dem*Distance in Table 5 is invariably positive and significant. The coefficient of 1.901 on lower Dem*Distance in the first regression Spending column, for example, indicates that an increase in Democratic seats from 51% to 61% would raise state government spending as a fraction of personal income by .19%. The most notable change is that Dem*Distance now appears to also increase sales taxes as a fraction of personal income; when sales taxes were expressed in levels rather than as a fraction of personal income, there was no discernible effect. + The sign on Rep*Distance in Table 5 typically remains insignificant, though frequently positive. + The compositional effects of political variables change slightly more. When education spending was expressed in levels, both parties appeared to make cuts as Distance increased; yet if education spending is expressed as a fraction of personal income, Dem*Distance has a slight positive effect on education, and Rep*Distance has a slight negative effect. The tendency of Democrats to cut health and highway spending, and to increase public welfare and “other” spending, remains. Republicans’ tendency to increase spending on health and highways also persists despite the specification change. C. Other Sensitivity TestsThe preceding specifications have all implicitly restricted the effect of grants on budgetary behavior to be the same for both parties. Does this introduce any important bias into the results? The regressions shown in Tables 1 and 2 were re-run, with the grant variable replaced with grant*Dem and grant*Rep to allow the effects of grants to vary by dominant party. Relaxing the implicit restriction alters a few conclusions, but does not change the overall picture. Grants induce Democrats to increase total taxes by more than Republicans; yet at the same time, grants seem to increase total spending more for Republicans than for Democrats. The income tax is the main margin on which this difference manifests itself: Democrats consistently increase income taxes by more in response to federal grants than Republicans would. Measuring political advantage by summing the seats in the lower and upper legislative houses makes very little difference for the results. Jointly estimating the effects of lower and upper houses does tend to increase the standard errors and reduce the absolute values of political coefficients, but makes little difference for the estimates of other variables. Another potential doubt about the baseline specification is that it implicitly assumes that political or economic variables function contemporaneously – if the governing forces or economic factors change, government policy changes in the same year. Replacing the explanatory variables in the baseline specification with their first lags revealed few changes from the baseline results. 5.5. Analysis of ResultsA. Results for the General ModelThe Leviathan hypothesis assumes that all political parties share the desire to expand the size of the public sector; the Ideologue hypothesis assumes that one party wants to expand the public sector and the other to contract it; the Technocrat hypothesis assumes that neither party has preferences one way or the other. All three hypotheses sometimes fail to match the empirical evidence. Contrary to the Leviathan hypothesis, Democrats show

36c