Inefficient Lobbying, Populism and Oligarchy

Transcription

1 Public Disclosure Authorized Inefficient Lobbying, Poulism and Oligarchy Public Disclosure Authorized Public Disclosure Authorized Filie R. Camante and Francisco H. G. Ferreira February 18, 2004 Abstract This aer analyses the efficiency consequences of lobbying in a roduction economy with imerfect commitment. We first show that the Pareto efficiency result found for truthful equilibria of common agency games in static exchange economies no longer holds under these more general conditions. We construct a model of ressure grous where the set of efficient truthful common-agency equilibria has measure zero. Equilibria are generally inefficient as a direct result of the existence of grous with conflicting interests, which allocate real resources to lobbying. If lobbies reresenting "the oor " and "the rich " have identical organizational caacities, we show that these equilibria are biased towards the oor, who have a comarative advantage in olitics, rather than in roduction. If the ressure grous differ in their organizational caacity, both ro-rich (oligarchic) and ro-oor (oulist) equilibria may arise, all of which are inefficient with resect to the constrained otimum. 1 Introduction Public Disclosure Authorized Lobbies and ressure grous are everywhere, and everywhere they emloy roductive resources - time and money - in the ursuit of influence over government decisions. There is widesread historical and economic evidence that such rent-seeking activities have had (and continue to have) real imact on olicymaking. Farm lobbies have successfully sought to maintain high levels of rotection for domestic roduction in a number of laces, including the Euroean We are grateful to Alberto Alesina, Abhijit Banerjee, Francesco Caselli, Quoc-Anh Do, Elhanan Helman, Richard Holden, Humberto Moreira, and Andre Sant Anna for their helful comments, as well as to seminar articiants at the Harvard Macro lunch, the World Bank, the Catholic University of Rio de Janeiro (PUC-Rio), and the V Meeting of the Political Economy Grou of LACEA. The usual disclaimer on remaining errors alies. Financial suort for this research has been rovided by Faerj. Corresonding author. Deartment of Economics, Harvard University. Contact: camante@fas.harvard.edu The World Bank and Catholic University of Rio de Janeiro (PUC-Rio). Contact: fferreira@worldbank.org 1

2 Union, Jaan and, more recently, in the United States. In federal countries, like the United States, India or Brazil, regional and state reresentatives often lobby central governments to attract ublic sending to their states. Unions ressure governments to introduce rotection against imorts which comete against domestically-roduced goods, or to raise minimum wages. Richer folks lobby against inheritance tax increases. Poor folks lobby against benefit cuts. In this aer, we seek to investigate the economic consequences of lobbying, both in terms of economic efficiency and of distribution. Do the actions of ressure grous - exending real economic resources in order to influence olicy decisions - affect the efficiency roerties of economic equilibria? Do they affect the distribution of income or wealth intheeconomy? Ifso,howdotheyaffect them? Do the rich and owerful always gain from lobbying, at the exense of the oor and under-reresented? Then how do we exlain oulist regimes, such as the Peronista governments in Argentina ( ; ); Salvador Allende s rule in Chile ( ) or Alan García s in Peru ( ), during which unions and other organized ressure grous reresenting the oor aeared to have much greater influence over government olicies than reresentatives of the business elite? 1 The historical evidence suggests that strikes, street demonstrations and other time- and resource-consuming activities were successfully used by oular ressure grous to affect government decisions, in between elections, in a number of oulist regimes. 2 On the other hand, in other countries - or even in the same countries at other times - other equilibria have arisen in which the influence of lobbies reresenting the interests of the rich have seemed to dominate. Can economic theory shed any light on the mechanisms through which non-electoral, ressure-grou olitics affects economic efficiency and the distribution of wealth? The imortance of lobbying as an economic activity has of course long been recognized. 3 However, the conclusion that lobbying was inefficient, which characterized the original literature on rent-seeking, derived from its treatment of the contributions made by lobbies to government agents as deadweight loss. 4 Once the government is exlicitly recognized as an agent in its own right, with its own references and resources, economic analysis of ressure grou olitics changes. The most fruitful theoretical aroach to lobbies when the government 1 Some olitical scientists have in fact defined oulism as "... a set of economic olicies designed to achieve secific olitical goals, [namely] (1) mobilizing suort within organized labor and lower middle-class grous; (2) obtaining comlementary backing from domestically oriented business; and (3) olitically isolating the rural oligarchy... and large-scale domestic industrial elites. " (Kaufman and Stallings, 1991,.16.) This definition immediately brings to mind the idea of ressure grou cometition. 2 One examle was union ressure for adjustments to nominal wages in Argentina, which had been fixed by a government decree (the National Comromise Act ) in June In early 1974, however, "the government gave in to union ressures and decided to adjust wages before the scheduled date [...] The decision to modify the adjusment scheme would rove a major cause in the subsequent collase of the rogram. " (Sturzenegger, 1991,.99). 3 A seminal early treatment can be found in Krueger (1974). 4 See, for instance, Rodriguez (1999), which follows this rent-seeking literature. 2

3 is viewed as a real economic agent is that of common agency. 5 In the common agency framework, if rincials behave truthfully (in the sense of revealing their true references, and hence aying as contributions to the agent all they are ossibly wiling to give in exchange for the agent s decision), then it has been shown (by Bernheim and Whinston, 1986, and with greater generality by Dixit, Grossman and Helman, 1997, henceforth cited as DGH) that the equilibrium of the common agency game is Pareto efficient. Moreover, the adotion of truthful strategies is shown to be otimal from the rincials standoint, in a well-defined sense. This key result aears in most of the olitico-economic models that aly the common agency aroach 6, and even models that obtain inefficiency results do so by abandoning the notion of truthful behavior (e.g. Besley and Coate, 2001). 7 This aer argues that this efficiency result deends crucially on two features of the game considered by DGH: the (imlicit) ure exchange nature of the economy; and the existence of erfect commitment (i.e. the fact that all contracts can be costlessly enforced). We show that when a roductive activity is exlicitly modeled - and its consequences fully accounted for, esecially in terms of a distinction between the resources available before and after roduction takes lace - the efficiency of truthful equilibria requires some mechanism by which rincials could erfectly commit to announced contribution schedules, or a erfect credit market which allowed rincials to have access to resources that would otherwise only be available to them in the future. When this is not the case, the resulting allocation is generally inefficient, as olitical and roductive activities comete for resources. We aly this general result to a secific model, in which two ressure grous, defined by their ositions in the initial wealth distribution ( rich and oor ) make olitical contributions seeking to influence the comosition of government exenditures. Allocation decisions deend on three factors: grou sizes, their organizational ability, and their comarative advantages. The interlay between these three factors is shown to allow for a rich gamut of ossible outcomes. If organizational or coordination caacity within the two grous is 5 A common agency roblem is one in which several rincials, with different and ossibly conflicting interests, try to influence the decisions of a single agent. Unlike most rincialagent setus, common agency games resent a non-trivial efficiency issue even under erfect information, as they raise the question of whether an efficient allocation of resources can be achieved when the several rincials act in a noncooerative way. 6 For instance, DGH (1997) and Dixit (1996), on tax rate decisions, or Grossman and Helman (1994, 1995a, 1995b), on trade tariffs. 7 Bernheim and Whinston (1986), Grossman and Helman (1994), and DGH (1997) have all argued that we should exect equilibria with truthful contribution schedules to be layed. At the end of the day, however, they do not offer an account of the decision making rocess which guarantees convergence to these equilibria (Besley and Coate, 2001,. 79, our abbreviation). We feel that those authors arguments on why to exect truthful behavior are ersuasive: the fact that truthful contribution schedules are always a best resonse (which imlies the existence of truthful equilibria), the fact that they are the only coalition-roof equilibria, and the fact that they might be focal in the set of equilibria due to the very fact that they might lead to efficiency. 3

4 identical, the equilibrium turns out to be ro-oor, or oulist (in which the comosition of government sending is biased towards the oorer grou, relative to the efficient comosition). If organizational caacity is allowed to differ, say by grou size, then equilibria can be either oulist or, instead, they can be ro-rich (or oligarchic), in which the reverse inefficiency occurs. Both oulist regimes and oligarchies are inefficient, because of the distortionary effect of lobbying on ublic olicy. More secifically, the olitical equilibrium turns out to be biased recisely towards those who have a comarative advantage in the olitical activity, i.e. those who are relatively less efficient in roduction. 8 The comarative statics of the model are also quite rich. Increases in the incidence of overty, for instance, will make an oligarchic equilibrium even less efficient, but will have an ambiguous effect on a oulist equilibrium. If the marginal roduct of ublic caital is sufficiently high, and organizational ability among the oor is sensitive to grou size, the oulist equilibrium may become more efficient. These various outcomes are ossible because we allow for two effects which are usually ignored in the literature. The first of them, which we dub the efficiency effect, is due to the fact that the olicy variable has a direct roductive imact, which means that changes in the wealth distribution also change the efficient comosition of ublic exenditure, hence the otimal decision on the olicy variable. The second effect, which we label the coordination effect, is due to the imact of wealth distribution on the size of the grous and their ability to coordinate as such, as in the well-known Olsonian view of ressure grous (Olson, 1965), which affects the olitical equilibrium. Our results are also related to those in Loury (1981), Galor and Zeira (1993), Banerjee and Newman (1993) and Aghion and Bolton (1997), in the sense that the initial distribution of wealth affects the efficiency roerties of the long-run equilibrium of the economy. As in those aers, imerfect commitment (which may arise from credit market imerfections) lays a crucial role, albeit here in an entirely different context: a non-electoral olitical rocess. We therefore find an additional inefficiency, directly linked to the olitical rocess, beyond the one that results from the market imerfections themselves. In this sense, wealth distribution can have an even larger imact on efficiency, as it sets in motion the conflict between ressure grous. In an examle in the sirit of Galor and Zeira (1993) or Ferreira (2001), the inefficiency arises not only because a oor individual might not be able to afford a rivate education that he or she would otherwise ursue, but also because government exenditures on ublic education might themselves be distorted as a result of the lobbying activity. In common with the literature on wealth distribution and olitical economy (Bertola, 1993; Alesina and Rodrik, 1994; Persson and Tabellini, 1994; Bénabou, 2000), we find that wealth inequality can lead to inefficient equilibria, due to 8 This resembles the result from the classic model of cometition among ressure grous by Becker (1983), in which the equilibrium deends on relative olitical roductivity between grous. But that model does not include roduction exlicitly, so it does not address the issue of olitical roductivity relative to efficiency in roduction er se. It also contrasts with the model by Acemoglu, Aghion and Zilibotti (2002), in which lobbying under credit constraints benefits those who have more resources. 4

5 the existence of conflicting references over ublic olicy. In contrast with most of it, however, in our model inefficiency does not arise from distortions inherent to the nature of redistribution (such as a tax on caital holdings). Instead, it arises from the very nature of the olitical rocess. Whereas those authors emhasize the inefficiencies caused by tax choices arising from electoral ouctomes, we consider inefficient sending decisions, as a result of lobbying from interested ressure grous. 9 Other aers that are related to ours are Esteban and Ray (2000) and Do (2002), which also look at wealth distribution and olitical cometition over government decisions. The former uses a signalling game aroach that is not suitable to model lobbying by ressure grous, which is what we focus on here. In addition, the results emerge from the imerfect information setu, whereas our results hold even under erfect information. The latter aer also analyzes ressure grou interaction under credit market imerfections, but unlike ours it focuses on the endogenous emergence of barriers to entry in regulated industries, rather than on the allocation of government sending. Moreover, neither of them is interested in the analysis of oulist as oosed to oligarchic outcomes. Finally, our aer also relates to the literature studying oligarchies versus oulism, of which Bourguignon and Verdier (2000) and Acemoglu (2003) areexamles. Theseaers,however,donotstudytheeffects of a olitical rocess based on lobbying, which are our main focus. The aer is structured as follows. Section 2 discusses how the efficiency result in common agency games changes when a roductive activity is considered exlicitly. Section 3 resents a model of interaction of ressure grous, defined along the wealth distribution, trying to influence the comosition of government exenditures. Section 4 concludes. 2 Efficiency roerties of truthful equilibria of common agency games in a roduction economy In this section we will briefly discuss the efficiency roerties of truthful equilibria of common agency games, once a roductive activity is exlicitly modeled, and its consequences fully accounted for. A formal discussion can be found in Aendix A. 9 The need to move beyond voting rocesses in order to understand the economic effects of olitics was emhasized by Atkinson (1997,. 316), [it seems] imortant to see how far the findings deend on whether the outcome is governed by the references of the median voter, or by the ideology or references of olitical arties, or by olitical ressure from different interest grous (...). There has been relatively little research by economists which has set side by side different ossible exlanations of income redistribution. 5

6 The common agency framework - as laid out by Bernheim and Whinston (1986) and DGH (1997), for instance - is one in which many rincials (say, ressure grous) try to influence the choice of some vector of actions chosen by the rincial (which we take to be a government, or olicy-maker), which have an imact on their utilities. They do so by making ayments to the rincial, conditional on the action chosen. The key result in this literature, established by the aforementioned authors, states that, if rincials offer a truthful ayment schedule, the resulting subgame erfect equilibrium (named truthful Nash equilibrium)isparetoefficient. 10 The intuition for this result is quite clear (DGH, 1997): cometition between the rincials enables the agent to extract all of the surlus in the game, and it will therefore be in her best interest to maximize that surlus. This is accomlished by an efficient allocation efficiency. Moreover, it can also be shown that such truthful behavior is otimal for agents, in the sense that the set of best resonses for any choice of strategies by other layers contains a truthful contribution schedule. In fact, although agents end u with the same utility they would achieve if no ayments were made at all, they are traed in a sort of risoners dilemma: every rincial would be better off if no rincial contributed, but if no one else contributes, then it is individually otimal to do so (DGH, 1997,. 767). These results, however, do rely on commitment between the rincials - who commit to the contribution schedules - and the agent - who commits to the action vector she chooses. This reliance is not articularly troublesome if we think of both actions being taken simultaneously, in the sense that the agent imlements the action with one hand, while collecting contributions from the rincials with the other. Things change, however, if we consider a roduction economy. The key feature of a roduction economy, in the resent context, lies in the distinction between available resources before and after the roductive activity is undertaken. If caital markets do not work erfectly, so that resources which will become available after roduction cannot be made available in advance, then rincials may not be able to rely on those resources to ay contributions. More secifically, suose the olicy choice by the agent concerns a vector of inuts, each one to be used in roduction by one rincial. Then resources made available by roduction cannot be used to ay for contributions simultaneously to the imlementation of the olicy vector, which by definition comes before roduction. In this case, if there is no other means by which commitment of the rincials - with resect to the announced contribution schedule after the agent s action is imlemented - can be achieved, then the efficiency result discussed above need not aly, as the agent will not accet contributions romised for delivery after roduction in any subgame erfect equilibrium. In other words, the set of feasible contributions in equilibrium reduces to a strict 10 A truthful ayment schedule, also known as comensating ayment schedule (see Grossman & Helman, 2001) is one in which rincials reveal their true references over the agent s actions and ay contributions accordingly, that is ay their comensating variation with resect to a given utility level. 6

7 subset in comarison to the case in which commitment is erfect, and the resulting allocation needs no longer be efficient with resect to the set of allocations that could ossibly be achieved in this roduction economy. 11 What is really crucial to the argument above is the imossibility of having access to roduced resources before engaging in the roductive activity. In this sense, the existence of erfect credit markets could lay the role of erfect commitment, in that it would allow for the anticiation of future resources. The inefficiency of truthful equilibria of the common agency game could thus be associated with some kind of credit-market imerfection. It should be noted, however, that erfect credit markets require erfect commitment between borrowers and lenders, in the sense of erfect enforcement of contracts. This reveals theverynatureoftheinefficiency under analysis: it is linked to some institutional roblem which gives rise to a contract enforcement failure, either within credit markets or between rincials and agent. In other words, this inefficiency essentially results from a roblem of incomlete contracts. 12 Finally, we should stress that this discussion on efficiency alies to the allocation of resources among the layers taking art in the game, as stressed for instance by Grossman and Helman (1994a). If there are other individuals in the economy who are not art of the common agency game, there can (and, in general, will) be inefficient outcomes for the economy as a whole, where these individuals end u being "exloited" by those layers in detriment of efficiency. In this section we discussed why truthful equilibria in generalized common agency games without erfect commitment need not be efficient. A formal argument is made in Aendix A. In the next section, we use an examle of such a game, where truthful equilibria are indeed generally Pareto inefficient, to shed light on the mechanisms through which ressure grou olitics affects economic outcomes. In articular, we are concerned with how lobbying and ressure grous may lead to either oligarchic or oulist equilibria. 3 Rich and oor ressure grous; oulist and oligarchic equilibria Having discussed the ossibility of inefficiency in a common agency game that is generalized to encomass the existence of roduction, let us now make use of this framework to investigate the imact of distribution on efficiency when the olitical rocess is modeled as an interaction of ressure grous trying to affect the comosition of government exenditures. 11 The maintained assumtion here is that the relevant set for efficiency analysis is the one which embodies the technological roduction ossibilities of the economy. The imlication is that an equilibrium reached subject to the feasibility constraints imlicit in the constrained set of ossible equilibrium contributions need not be Pareto efficient in that larger set. This efficiency assessment is analogous to stating that a general equilibrium with a missing market is no longer Pareto efficient (as comared to one with comlete markets). 12 Itisossiblethatcommitmentdevicesmightemergeinacontextinwhichthegameis reeated. As with the standard Folk theorem, their existence would require some uer bound on discount rates. 7

8 3.1 The model Individuals and Production We model an economy that exists for a single eriod, and consists of a continuum of rivate individuals forming a oulation of size one, and of a searate agent, called the government, whose attributes are discussed below. The individuals in the continuum are identical, excet for their initial wealth, which is distributed as follows: a roortion of the oulation has initial wealth w, whilethe remaining 1 is endowed with w, where0 <w< w, as in Bourguignon and Verdier (2000) or Acemoglu and Robinson (2000). There is a single good, which can be either consumed or invested, in either of three tyes of caital. k denotes rivate caital, which can be accumulated by individual agents. g and s denote two different kinds of ublicly-rovided caital, which can only be roduced by the government. The objective function of the government agent, which we will resent below, imlies that it is amenable to rivate contributions, that can influence its allocation choice between g and s. It will thus be rational for rivate agents to make contributions, contingent on the government s actions: C(g, s). Private roduction occurs by means of atomistic rojects with inelastic and unit labor suly, according to the following roduction function: ½ A(g + αs) Ψ(k, g, s) = a k 1 a,k >k Bs a k 1 a,otherwise (1) where g and s denote the government er caita exenditures on the two different kinds of ublicly-rovided rivate goods 13, 0 <α<1, 0 <a<1, Aα a >B(which means that, given the otion, individuals will refer to use the first technology). Caital markets are assumed to be non-existent. The resence of the exogenous threshold k reresents a nonconvexity of the roduction set, and as a result gives rise to the ossibility of two classes, which will be called rich and oor, defined by initial wealth distribution: agents who have the ossibility of investing at least k will have access to a more roductive technology, while those who have not will have to settle for a less efficient one. Since there are no caital markets, investment is limited by initial wealth. Since each individual lives for a single eriod and derives utility only from his own consumtion, his objective will be to maximize disosable income, which will be totally consumed. Therefore the utility function of a rich (oor) individual can be written simly as u R (k R,g,s)=Ψ R (k R,g,s) (u P (k P,g,s)= Ψ P (k P,g,s)), as given by (1). The secification in (1) imlies that the ublicly-rovided goods lay a fundamental role in rivate roduction (as in Barro, 1990). g and s can thus be seen as two kinds of ublic caital, with differentiated imacts on roduction: while g is useful only to the rich, s is more beneficial to the oor (given 13 The analysis would not be qualitatively changed if we consider ublic goods instead of ublicly-rovided rivate goods (in which case g and s would stand for total government exenditure in each tye of good). The analytical exressions that we derive below would be slightly different. 8

9 the assumtion on α). This gives rise to a conflict of interests between classes within the model, and is meant to stand for the fact that there are many tyes of exenditure which are aroriated exclusively (or redominantly) by the richest strata in society, while other tyes are more useful to the oor, even though they can also be used by the rich (Ferreira, 1995). Public healthcare exenditures may exemlify the latter, while subsidies to tertiary education in develoing countries could illustrate the former. 14 What is imortant is that the existence of these two tyes of exenditures means that the decision on the comosition of total government exenditures has distributional consequences. The government finances the roduction of g and s through taxation. It is assumed that the government s taxation technology is such that it can only raise funds through a linear wealth tax at the beginning of the eriod. The government is subject to a balanced budget constraint: τ [(1 )w + w] =(1 )g + s where τ is an exogenously given tax rate on initial wealth. 15 This restriction imlies that s may be exressed as a function of g, Political rocess First we assume that the two classes actually exist - those individuals with initial wealth w are the oor, and those endowed with w are the rich. 16 Moreover, they are articulated as ressure grous, each trying to influence the government s olicy decision concerning the choice between the two aforementioned tyes of exenditure, by means of olitical contributions. Each grou romises to ay some amount to the government, deending on the olicy choice (a = {g,s}). These contributions actually stand for a lethora of real-life ractices, such as money (or time) devoted to camaign contributions, or ure bribery, among many others, as in Grossman and Helman (1994). We also assume that individuals can only influence government behavior through this channel if they are art of an organized ressure grou: each individual erceives himself as too small to influence olicy decisions on their own (Grossman and Helman, 1994). As before, we assume that there cannot be erfect commitment to the announced contribution schedules: there is no way by which either the rich or the oor can credibly comit to meet their announced olitical contributions after the government has imlemented its decision. As we have seen, this imlies that contributions must be aid before roduction, hence resources available for roductive investment must be net of such ayments. To summarize the time structure of the model, we can reresent life in this one eriod economy in the timeline deicted in Figure Wewilllaterconsidertheolarcaseinwhichα =0, ossibly reresenting exenditures targeted exclusively to the oor. 15 Note that olitical contributions do not enter the budget constraint. One should think of them as going to a arty fund, or simly being shifted towards rivate consumtion of the government agent. 16 This amounts to assuming w <k and w>k +C R (g 0 ),wherec R (g 0 ) is the equilibrium contribution of a rich individual, as will soon be defined. 9

10 [FIGURE1HERE] Following Grossman and Helman (1994) and DGH (1997), we assume that the government s objective function is a convex combination (with weight x) of the contributions it receives and of a social welfare function. This can either be interreted as allowing for some benevolence on the art of the government agent, or simly to cature the fact that actual olitical rocesses are not limited to the interaction of ressure grous, and may also include more democratic channels, which make the government care about the welfare of the eole. To allow for different organization (or coordination) caacities across the two grous of agents, without exlicitly modeling their formation, we attach weights to each grou s contribution in the government s utility function, reresenting a given grou s relative ease of organization - and thus greater lobbying effectiveness - by a greater weight. In fact, the usual argument in that discussion states that smaller and less diserse grous have a higher robability of actually being formed, due to transaction costs and to the roblem of free-riding, and this could be reresented by letting these weights deend on each grou s size. Within the resent context, where the grou of the oor (rich) has size (1 ), we can define these weights as λ P () and λ R (), whereλ0 P () < 0 and λ0 R () > 0, in order to cature this idea. 17 All these features may be exressed, drawing uon Grossman and Helman (1994), by modeling a government that maximizes the following objective function: G = x [λ R ()(1 )C R (g)+λ P ()C P (g)] + (1 x)[(1 )Ψ R (g, C R (g)) + Ψ P (g, C P (g))] (2) where C j (g) is the olitical contribution from an individual member of grou j as a function of the comosition of government exenditures, 18 which we shall assume to be continuously differentiable, and x [0, 1] is the weight attached to contributions vis-à-vis social welfare (considering for simlicity a Benthamite welfare function in which every individual has the same weight 19 ). The individual s utility - which is identical to its ost-roduction disosable income - is written in (2), with a slight abuse of notation, as Ψ j (g, C j (g)), since k j =(1 τ)w j C j (g). Assuming erfect information, the roblem is therefore written exactly as a generalized common agency game - where ressure grous are the rincials, and the government is the agent - and its solution may be obtained as such. 17 This an admittedly very reduced-form attemt to allow for differentiated coordination caacities across different coalitions, which may imact on their effectiveness as ressure grous. The classic reference is Olson (1965), and Becker (1983) is another instance of alication of this insight. 18 To write C j (g), we use the fact that the government s budget constraint allows us to write s as a function of g. 19 It should be ointed out here that such assumtion imlies that there is no social inequality-aversion. 10

11 3.2 Results Efficiency of truthful equilibria As with the standard common agency game discussed in DGH (1997), this game comorts a multilicity of subgame erfect Nash equilibria. We follow Bernheim and Whinston (1986), Grossman and Helman (1994) and DGH in restricting our attention to truthful symmetric Nash equilibria only, since truthful contribution schedules are always a best-resonse strategy and are the only coalition-roof equilibria of these games. In this subsection we therefore turn to the efficiency roerties of truthful Nash equilibria within this model. First let us characterize in Proosition 1 the constrained efficient allocation in this economy 20, which will serve as a benchmark for comarisons with the olitical equilibrium allocation. Proosition 1 APareto-efficient allocation {kr,k P,g,s }musthave h i kr 1 1 a B k g +αs = P 1 α(1 ) A s. (3) Proof: See Aendix B. The efficiency condition established in Proosition 1 follows directly from the first-order condition of the roblem faced by a hyothetical social lanner whowishedtomaximizeabenthamitesocial welfare function (corresonding to total outut) in this economy, by choice of ublic exenditure, subject to kr the government budget constraint. The exressions g +αs and k P s reresent what may be called the rivate-ublic caital ratios of the rich and the oor, resectively. In other words: how many units of rivate caital are invested er unit of ublic caital obtained by a given individual. Proosition 1 says that those ratios must be related in a recise manner in order to obtain an efficient (1 ) allocation. The term 1 α(1 ) is equal to (1 )[1 α(1 )], which is exactly the ratio between the marginal cost to the grou of the oor of an increase in the rich-secific tye of exenditure g (i.e. a change of the comosition of government exenditures) and its marginal benefit to the grou of rich. The term B A gives a measure of the roductive efficiency of the oor relative to that of the rich. Therefore an efficient allocation must equate one grou s marginal cost to the other s marginal benefit, taking into account their relative efficiency on roduction. On the other hand, a truthful olitical equilibrium may be characterized as follows: Proosition 2 (i) A feasible allocation {k 0 R,k0 P,g0,s 0 } is a truthful equilibrium only if 20 What we are calling an efficient allocation takes as given the fact that there are some individuals that are restrained in their roductive ossibilities, to a worse technology, given the absence of credit markets in which they could ossibly have access to k. We could otherwise consider the outcome with erfect credit markets as being the efficient one, and our resent notion of efficiency would be a second best. More on this will come later. 11

12 k 0 R g 0 +αs 0 λp () λ R() 1 α(1 ) k 0 P s 0 (4) = (ii) Such allocation is almost always Pareto-inefficient. Proof: See Aendix B. The intuition behind equation (4) is analogous to the one behind equation (3): a olitical equilibrium equates marginal costs and benefits, only now the grous relative efficiency in lobbying is taken into account. The first art of Proosition 2 is obtained by imosing the requirements that the equilibrium allocation be otimal for the government, given the rincials contribution schedules, and otimal for each of the grous (rincials), given the government s feasibility and rationality constraints. The second art, where the inefficiency of the olitical equilibrium is established (excet by coincidence) relies on the fact that the ratio of rivate-ublic caital ratios within each grou in (4) will only be equal to that in (3) for secific arbitrary values of the exogenous arameters λ P () and λ R (). See the roof. Additionally, this inefficiency result does not deend on the introduction of the articulation ower functions λ P () and λ R () in the model. Quite the contrary, it is only by allowing for the ossibility that they differ across the two grous that it becomes ossible (with robability measure zero in the arameter sace) that the olitical equilibrium attains constrained efficiency. This oint is made formally in the following corollary: Corollary 3 If both grous have the same articulation ower (λ P () =λ R ()), then a truthful equilibrium allocation is always inefficient. Proof: See Aendix. This is an examle of the inefficiency of generalized common agency games without erfect commitment, discussed in Section 2 and formally established in Aendix A. 21 It shows that the restriction on the Pareto-efficiency of truthful equilibria that is imosed by the absence of erfect commitment can actually be binding. It is worth emhasizing that the inefficiency under analysis is not the one related to the absence of credit markets and the roductive nonconvexity, as is usual in the literature. Indeed, what we call an efficient allocation in Proosition 1 already embodies the fact that some agents are constrained to a less roductive technology: it is a constrained otimum, or second-best. The equilibrium allocation described in Proosition 2 is therefore not even the constrained otimum: there is an additional inefficiency linked to the olitical rocess. We therefore have two levels of inefficiency: the first one generated by the existence of individuals who are constrained to a worse technology, the second one deriving from the fact that not even the constrained otimum is attained, 21 It is easy to check that introducing erfect commitment in our model actually leads to an efficient allocation if λ P () = λ R (), which is a mere alication of the result due to DGH (1997), but can also be verified by an argument identical to the one used in the roof of Proosition 2. The result with erfect commitment may be inefficient if we consider λ P () 6= λ R (), but that would be trivial in that such inefficient would be generated simly by corrut government behavior. 12

13 because of the olitical inefficiency. This second level is the distinctive feature of this model: the oint is that not only is there an inefficiency due to the fact that the oor cannot afford to ay for the level of rivate education that would make them more roductive, for instance, but there is also another inefficiency due to the fact that the government will not rovide them with the otimal level of ublic education. Let us consider the nature of this inefficiency further. h be measured by the absolute value of θ 1 α(1 ) B A i 1 1 a Note that it can λ P () λ R() 1 α(1 ), which is exactly the difference between the rivate-ublic caital ratio of the rich (relative to that of the oor) in the efficient allocation and in the olitical equilibrium. Let us also define the olitical equilibrium as ro-oor (or "oulist") if θ<0, and as ro-rich (or "oligarchic") if θ>0. These definitions refer to the fact that in the former case, ublic sending deviates from the constrained otimum by allocating more units of ublic caital er unit of rivate caital to the oor than would be efficient, while such advantage belongs to the rich in the latter case. They are resented grahically in Figure 2, where O stands for oligarchic and P for oulist. [FIGURE2HERE] It is interesting to note that if λ P () =λ R (), i.e. both grous have the same articulation ower, then we have a oulist equilibrium. To ut it another way, if the oor can organize themselves as effectively as the rich, the allocation of ublic exenditures generated by the olitical equilibrium will be more beneficial to them than the efficient one. This result may at first aear surrising, as it means that a olitical system in which the government s decision-making is influenced by channeling economic resources to the government turns out to be relatively beneficial to the oor. Let us consider it more closely. The result is driven by the comarative advantages of each grou: λ P () and λ R () reresent each grou s olitical roductivity, i.e. their effectiveness in lobbying the government. Since the rich have an absolute advantage in roduction (by assumtion in (1)), identical olitical roductivities imly that the rich have a comarative advantage in roduction, while the oor have a comarative advantage in olitics. Each grou will tend to secialize (artially, rather than comletely, due to decreasing returns to both tyes of caital) in the activity in which it has comarative advantage. 22 The oor thus secialize in lobbying, shifting the olitical equilibrium towards them. Note that the Inada condistions satisfied by the Cobb-Douglas roduction function, by ensuring that the marginal roduct of ublic caital tends to infinity as the availability of rivate caital aroaches zero, ensure that the oor do value access to ublic caital, even when they are severely constrained (i.e. extremely oor). In those circumstances, even though they may have little to contribute in absolute terms, 22 This can be seen in (4) by checking that λ j () is inversely related to the caital that is rivately invested by members of grou j, and is therefore directly related to their olitical contribution. 13

14 their contribution will be strictly ositive and will be higher in relative terms than that of the rich. With identical organizational caacities (lambdas), the comarative advantage of the oor in olitics imlies that the equilibrium ratio will be lower than the Pareto-otimal ratio, no matter how small k may be, as long as it is strictly ositive. The key idea is that the oulist or oligarchic nature of the equilibrium here is defined in terms of ublic exenditure er unit of rivate exenditure: for instance, it is ossible that in a oulist equilibrium therichobtainmoreublicexenditureinabsolutetermsthantheoor. This result is consistent with the historical evidence from eisodes such as Peronism in Argentina, Getúlio Vargas s government in Brazil ( ), and Alan García s government in Peru ( ), when unions and other oular organizations exercised their democratic right to demonstrate and ressure the government for the adotion of secific olicies, even long before elections. Some such olicies, such as fixing real wages above market-clearing levels, were unlikely to lead to greater economic efficiency. They were, in addition, contrary to the interests of caitalists. They were adoted as the result of olitical ressure, which often required considerable investment of time and resources on the art of lobbyists. Consider once again the examle of Argentina s National Comromise Act of 1973, which was mentioned in the introduction. After President Peron s government resonded to union ressures by granting increases in administered wages in March 1974, while uholding the freeze on other rices, "illegal rice increases and black markets began to roliferate." (Sturzenegger,1991,.99). As art of its suort for the government s olicy, the Argentine Trades Union Congress (Confederación General del Trabajo) formed "sectorial commissions in order to control the adherence to the rice and suly olicy and therefore contribute to avoiding seculation and abuse... " (La Nación, 28 March 1974, in Sturzenegger, o. cit.). The oint is that organized (and time-consuming) olitical action by grous drawn from the lower art of the income distribution (such as urban factory workers) have, on a number of occasions, succeeded in determining economic olicy outcomes. 23 Such outcomes were generally biased in favor of those oorer grous, and were often economically inefficient. While our model is thus consistent with instances in which the oor dominate the olitical rocess, leading to government olicies that are biased against the rich, it is also erfectly consistent with outcomes in which the absolute advantage of the rich in lobbying is so large that - even though the oor still secialize in lobbying - the olitical equilibrium is ro-rich. This requires that λ P () <λ R (), 24 whichmayariseinracticewhentherichareasmallerand less disersed grou than the oor. As a general oint, the olitical equilibrium 23 In Latin America, it was often the case that the oorest grous in society (say, rural workers) were not actually reresented in the ressure grous underinning oulist regimes (say, mostly urban workers). As long as these latter ressure grous found themselves in olitical cometition with a richer grou (say, traditional elites and factory owners), our results remain relevant. Our model could be trivially modified to accomodate an underclass which was oorer than w, rovided it did not articiate in the game. 24 This is a necessary, but not sufficient, condition. The stronger, sufficient condition is that θ>0. 14

15 will favor the grou that has comarative advantage in the olitical activity, in other words, those who are relatively less efficient in roduction. 25 It should also be stressed that what is being meant by an efficient allocation does take into account the role of the government as a layer: resources used as olitical contributions are not being considered a deadweight loss, as is usual in the literature on rent-seeking. 26 The inefficiency that arises in our model is due to the fact that lobbying distorts two key decisions: both the rivate investment decision of the rincials and the comosition of government exenditures. If there were erfect commitment, it would be ossible to searate roduction from olitics and what would revail would be erfectly analogous to the outcome of the common agency game without roduction. Without commitment, the two cannot be disentangled, and resource allocation ends u being distorted. Moreover, such inefficiency is not a mere consequence of corrut government behavior: erfect commitment allows for an efficient outcome desite the fact that the government still receives contributions and derives utility from them. It may also be noted that this olitical inefficiency does not deend on the weight that the government attaches to olitical contributions vis-à-vis social welfare - as long as this weight remains strictly ositive - which can be seen from the fact that x does not aear in either (3) or (4). This remark reinforces the observation that the inefficiency stems from the mere existence of lobbying, in the absence of erfect commitment or credit markets. In this sense, it does not deend on how democratic the olitical rocess turns out to be. 27 As a final note on our results, let us oint out that they remain valid when government sending on the kind of ublic caital which is referred by the oor (s), is erfectly targeted to them (α =0). One could interret α as a "leakage rate" of the sending on s to the non-oor. It it is easy to show that the in the olitical equilibrium with zero leakage, the following must hold: k 0 R g 0 = λ P () kp 0 λ R () s 0 (40) while the efficient allocation would have required that: k R g = 1 B 1 a A k P s. (30) 25 This is in contrast with a different exlanation for inefficient lobbying under credit constraints: it could be the case that the olitical equilibrium is biased towards those who have more resources to ay olitical contributions, which might not be those with the highest-return rojects. This is what haens, for instance, in the aer by Acemoglu, Aghion and Zilibotti (2002), where "richer agents can ay greater bribes and [thus] have a greater influence on olicy" (. 37). 26 To use the terminology of Esteban and Ray (2000), our model features allocational losses. But if one is willing to think of olitical contributions as being socially wasteful, our model also features the conventional conflictual losses that are tyical of rent-seeking models, for equilibrium contributions must be ositive - if they were zero the government would be maximizing social welfare. 27 Formally, this result stems from the enveloe theorem: when the government considers the imact of a change in the comosition of exenditures, the effect on the agents welfare - the democratic comonent - vanishes because of the first-order condition for agents otimization. This is clearly a consequence of the concet of truthful equilibrium. 15

16 This shows that our result does not deend on the additive comonent of the roduction function (1), but solely on the existence of a olitical decision on the comosition of government exenditures between two kinds of ublic caital: one which is referred by the rich, and another which is referred by the oor Comarative Statics: inequality, overty and efficiency Having characterized the inefficiency of the olitical equilibrium, let us now examine the imact of changes in the wealth distribution on the equilibrium. The distribution of wealth in this economy is fully described by three arameters: one of the wealth levels (say, w);, the roortion of the oulation which is oor (or equivalently the relative size of the two grous), and d w w, which can be thought of as a measure of inequality. In this subsection, we discuss some comarative statics, concerning the effects of changes in d or on θ, which measures the extent of oligarchic inefficiency of the equilibrium. Starting with the former, it may seem at first that inequality has no imact on the magnitude of the inefficiency, as θ is not functionally deendent on d. This conclusion, however, deends crucially on the exact nature of the change in d, given the nonconvexity of the roduction set. A decrease in inequality that gives the oor access to the more roductive technology, without leading the rich to become oor, takes the economy automatically to the efficient allocation, for any conflict of interests vanishes. All agents would then refer that the government roduce only the g - caital good. Conversely, if the tranfer between rich and oor makes everyone oor, so that for examle k > w>w,then conflict also disaears, with everyone referring the government to roduce s. The effect of changes in inequality on inefficiency is discontinuous: a marginal change can have a large imact if it haens to fit one of the above cases. This discontinuity, which arises from the nonconvexity in the roduction set, is a feature that our model shares with many imerfect-caital-market models which also rely on such a nonconvexity (Galor and Zeira, 1993; Banerjee and Newman, 1993). On the other hand, a significant change in inequality could have no effect whatsoever, rovided that the resulting wealth distribution still consisted of two grous, each using a different technology, and each thus referring the government to roduce a different kind of ublic good. Our model, like Banerjee and Duflo (2003), is one in which inequality will only lead to inefficiency insofar as it leads to the formation of grous with conflicting interests, for it is the olitical interaction of such grous that generates a distorted allocation. This link does not stem from any inherently inefficient roerty of the redistribution activity er se - as it did in Alesina and Rodrik (1994) or Persson and Tabellini (1994). Inefficiency arises from the fact that lobbying ower is not necessarily roortional to roductive caacity (because rincials can not erfectly commit to use their outut to ay contributions). The lobbying rocess thus gives rise to an allocational decision by the (self-interested) government which is not necessarily socially otimal. Since the government s outut is an inut 16