Endogenous Political Institutions

Transcription

1 Endogenous Political Institutions Philie Aghion, Alberto Alesina 2 and Francesco Trebbi 3 This version: August 2002 Harvard University, University College London, and CIAR 2 Harvard University, NBER and CEPR 3 Harvard University Abstract Political institutions influence economic olicy, but they are themselves endogenous since they are chosen, in some way, by members of the olity. An imortant asect of institutional design is how much society chooses to delegate unchecked ower to its leaders. If, once elected, a leader cannot be restrained, society runs the risk of a tyranny of the majority, if not the tyranny of a dictator. If a leader faces too many ex ost checks and balances, legislative action is too often blocked. As our critical constitutional choice we focus uon the size of the minority needed to block legislation, or conversely the size of the (suer)majority needed to govern. We analyze both otimal constitutional design and ositive asects of this rocess. We derive several emirical imlications which we then discuss. We thank Daron Acemoglu, Marios Angeletos, Olivier Blanchard, Stefano Della Vigna, Jeffrey Frieden, Ed Glaeser, Elhanan Helman, Bryan Graham, Oliver Hart, Miklos Koren, Vardges Levonyan, Matthew Jackson, Andrei Shleifer, Adam Szeidl, Motohiro Yogo, and seminar articiants at CBRSS, CIAR, Columbia, Coenhagen, Harvard, MIT, NYU, Stockholm, University of Pennsylvania and the Mc Arthur Foundation grou on inequality for helful comments. Federico Etro rovided excellent research assistance. Alesina and Aghion gratefully acknowledge financial suort from the NSF through the NBER and from the Canadian Institute for Advanced Research resectively.

2 Introduction Political institutions contribute to determine the choice of economic olicies, which, in turn, determine economic success. However, institutions themselves are chosen by individuals and they evolve in resonse to changing oliticoeconomic conditions. Thus, in order to fully understand and roerly test emirically the economic effects of institutions on economic olicies and outcomes, one has to allow for the endogenous choice of the latter. The goal of this aer is to highlight how several characteristics of a society lead to different institutional choices. We focus on a very general feature of olitical institutions and we label it the degree of insulation of olicymakers. An insulated leader 2,onceaointed, can rule with little or no ex ost control. A non insulated one, instead, has to form large majorities to ass legislation and many grous may have veto ower ex ost. Alternatively, one can view this question as the determination of the (suer)majority needed to ass legislation. Classical olitical theorists were well aware of the imortance of this asect of constitutional design. For instance, in Democracy In America, Alexis de Tocqueville stressed that Our contemoraries are incessantly racked by two inimical assions; they feel the need to be led and the wish to remain free 3. The Founding Fathers well recognized this dilemma. For instance, in the Federalist aer n. 70 Hamilton writes that Taking for granted..that all men of sense will agree in the necessity of an energetic executive, it will only remain to inquire what are the ingredients which constitute this energy? How far can they be combined with those other ingredients which constitute safety in the Reublican sense? The theory of checks and balances, embodied in the work by Montesquieu (748) rovided the answer adoted by the framers of the American Constitution. More secifically, the question of suermajorities as a way to restrain the tyranny of the majority features rominently in the Constitutional theory by Hayek (960) and in Buchanan and Tullock (962). The latter, for instance, argue that the simle majority rule does not have any articular suerior standing and under certain conditions may lead to excessive costs imosed on individual liberties by collective action. We model insulation as the share of votes needed to block legislation to ass, or, to ut it in reverse, we look at the size of the (suer)majority that the leader has to command to ass legislation. The Constitution establishes For a broad emirical discussion on how institutions affect economic olicy see Persson and Tabellini (2002) and the references cited therein. For recent emirical work on the effect of institutions on fiscal olicy see Milesi-Ferretti, Perotti and Rostagno (2002). For the effect of democracy on growth see Barro (996). Gil, Mulligan and Sala-i-Martin (2002) argue that the effect of olitical institutions on social security is small, contrary to arguments by Persson and Tabellini (2002). 2 We use the terms leader, olicymaker and olitician interchangeably. 3 Volume 2, art 4, Chater 6, age 664 from the translation by Mansfield and Winthro (2000). 2

3 the degree of insulation, and we abstract from all other institutional details. We analyze first the case of a erfect democracy in which all citizens are truly behind a veil of ignorance at the time of the constitutional choice and all have a vote in this choice. We show how various features of the oliticoeconomic environment affect this choice, including: the distribution of voter references; the nature of uncertainty; the feasibility and costs of comensating the losers from olicy reforms; the ossibility for leaders to exroriate citizens; the average benefits and costs of otential olicy reforms; the degree of risk aversion. Second, we investigate the olitical economy of institutional design. That is, we discuss how the otimal choice of institutions would or would not be adoted in a system where the choice was not made comletely behind a veil of ignorance and/or only a fraction of the oulation has a voice in the choice of institutions. What we have in mind is a situation in which those who choose a Constitution are also those who know who will control olitical office after the Constitution is ratified. In this case, what is otimal for those who choose the Constitution may not be otimal for society as a whole. Our model delivers several results with clear emirical imlications. Although the resent aer is mainly theoretical, the goal of this research roject is emirically motivated and in the last section of the aer we discuss some emirical imlications and redictions of our model. More extensive emirical work will be the focus of future research. Related to the resent aer is a recent literature on choosing how to choose, i.e. voting on voting rules. Aghion and Bolton (997) introduce an incomlete contract methodology to analyze the normative choice of otimal majority rule. Maskin and Tirole (200) rovide a somewhat related discussion of government accountability. Barbera and Jackson (200) investigate the endogenous choice of a majority voting rule, investigating conditions of existence of a self-stable voting rule, an issue that we will revisit below. Koray (2000) exlores instead social choice functions and whether such functions are self-selecting. In an overlaing generations setting Polborn and Messner (2002) identify a trade off arising in the selection of voting mechanisms over a reform when only art of the oulation (the old) incurs the cost of the reform, but not the subsequent benefits. The aer is organized as follows. In section 2 we describe the model and its interretation. Section 3 solves the model and derives basic comarative statics results. Section 4 illustrates several extensions. Section 5 discusses the olitical economy of writing constitutions. Section 6 highlights several emirical imlications of our model and brings about emirical suort. The last section highlights lans for future research. 3

4 2 Political Insulation 2. The Model Consider an economy oulated by a continuum of individuals, assumed, for the moment, to be risk-neutral with resect to income. Members of this olity will differ ex ost on how much they benefit from olicy actions (labelled reforms ) which may be imlemented. If no reform is imlemented, all individuals obtain the same income, which we normalize at. Individual income from the olicy reform is given by: 4 where ½ eλi γ if reform occurs ey i = otherwise eλ i = λ i + a, with γ > 0 () with λ i uniformly distributed on [λ, λ], with λ < λ and λ m λ + λ 2 From now on, we label l = λ λ. ais a random variable with mean zero, uniformly distributed between [ A, A], where A>0. We use the uniform distribution to obtain simle closed-form solutions, but below and in Aendix we show how our results generalize. Note that if λ m γ > the olicy reform is ex ante efficient in the sense that it makes the average (and median) voter better off. This community selects a leader to romote and imlement reforms. With exogenously given robability the selected leader is good and romotes the reform; with robability ( ) the leader is bad and exroriates (in the broad sense of the term) the olity. Next section on interretation discusses some reasons behind our secification of the olitical trade off in such terms. Also for the moment we assume that all individuals are ex ante identical in terms of their wealth, so the costs of exroriation for each individual is the same and we label it bw. Sinceweassumethatw is for the moment identical for everyone, we normalized it to. Whether good or bad, a new reform can be blocked by a (suer) majority M of individuals, once the aggregate shock on references a is realized. We define M the degree of insulation : if M is high, only a large majority of voters can block the reform. On the contrary, a low M means that when in office the leader is checked by a large fraction of the electorate. Thus a leader asses a reform only if a fraction ( M) of the oulation favors it or can exroriate only if ( M) of the oulation does not object to this olicy. 4 The following secification builds uon Krusell and Rios-Rull (996) and Aghion and Howitt (998, Ch.9) on the olitical economy of vested interests. 4

5 Note that when M</2 then suermajorities are needed to ass legislation. Also, in order to exroriate the leader has to buy off a fraction ( M) of the oulation in order not to be blocked. Thus, ex ante each individual in the olity faces robability M of being subject to the exroriation, if the latter is not blocked. The model, then, identifies a trade off at the constitutional stage: a more insulated leader can be less easily blocked, so the robability that a good reform asses is higher, but individuals are also more likely to suffer losses from exroriation. The choice of M occurs ex ante, before the realization of a and the size of M cannot be made contingent uon the realization of a. Thus, we assume that the corresonding events cannot be described ex ante, and we rule out social contracts contingent uon messages that voters would exchange ex ost about the realization of these random variables. For the moment we assume that the λ e i are not observed by the olitician and that the olitician cannot comensate the losers. We analyze the case in which the constitutional choice on M is made behind a comlete veil of ignorance; that is, all individuals are identical and learn their λ i after the constitutional choice is made. In summary, the timing of events is as follows: i) M is chosen at the constitutional stage, by individuals behind a veil of ignorance, that is before the realization of the λ i in the interval λ, λ ; ii) λ i is realized; iii) the olitician rooses the reform or the exroriation; iv) the uncertainty on the distribution of ex ost references is realized; v) blocking of the reform may occur; the reform is imlemented if and only if it is not blocked by the voters; if the leader is bad he exroriates, u to the oint that avoids blocking. The motivation of this timing needs discussion. Stage i) reresents the constitutional level in which decisions are made behind a veil of ignorance. Stage iii) is rather trivial. The only role of the olitician is to romote a reform, that asses if not blocked, or to exroriate the citizens. Stes iv), and v) cature the ost electoral dynamics between leader and voters. The latter imlies that after the realization of the shock a the voters still retain a choice to block ex-ost undesirable reforms. If the threshold for blocking M is set low, then the voters insure themselves that they will have a voice ex ost. However, this makes reforms easily blocked. On the other hands, if M is high, reforms ass more easily, but a larger fraction of the oulation may be exroriated, thus, ex ante, the robability that each erson is taxed is higher. 5

6 2.2 Interretation There are four critical elements in our model that need discussion. The first one is our notion of reform. If λ m γ > the reform embodies an element of efficiency, but also may involve winners and losers. Note that the arameter γ allows us to shift the overall benefit of the reform keeing the median voter and the distribution of voters unchanged. 5 Examles of olicy reforms that we have in mind include trade liberalization reforms, cometition or entryenhancing olicies, deregulation, labor market reforms, reforms of the social security system or fiscal adjustment ackages to eliminate deficits. These reforms may ex ante favor a majority, but create net costs for a minority. Ex ost, as a consequence of aggregate uncertainty, the distribution of costs and benefits may differ from the ex ante one, and, as a result, the distribution of those in favor and against the reforms may change over time. The reforms for which λ m γ < favor a minority but are harmful to the majority, at least ex ante, although they may be not harmful ex ost. The second element is the structure of uncertainty about the realization of voter references. The constitutional decision is taken behind a veil of ignorance, before the realization of the arameter λ i for all i s and with all individuals facing the same status-quo outcome in case reforms do not occur, an assumtion we shall relax below. The reference shock a has to be interreted as a change of the distribution of references occurring after the leader has taken office and while he is imlementing his olicy. This is meant to cature the idea that as a reform materializes through the effort of a leader new voters come in or the oulation at large matures definitive references about the reform for instance as they learn more recisely who will be a winner or loser form the reform. The role of the olitical leader in the basic model is highly stylized. A leader is needed to romote the reform and to ass it (unless it is blocked). However, the leader can take advantage of his osition to exroriate. Obviously, if the citizens could romt reforms without a leader, exroriation would not occur, but we emhasize the realistic idea that a centralized entity is needed to coordinate the reform olicy. Also the Constitution could rohibit exroriation, but not reforms that would be a Pareto imrovement. In reality it is difficult to fully restrain the authority of the government in this resect to exroriation without restricting its ability to govern in other areas. We will return to this issue below. The third imortant element is the degree of insulation, catured by the arameter M which we view as a summary statistic for a wide variety of institutional rules that limit the ower of aointed leaders. The most direct interretation of M refers to the question of what majority a government has to command to ass legislation. In general terms the issue of the otimal 5 We could achieve similar effects by setting γ = and varying λ m. However we refer this notation, which also makes it easier to analyze infinite horizon extensions (see Section 4.4). 6

7 suermajority rule is a widely debated question by Constitutional theorists. Those who favor suermajority rules (low insulation) worry about limiting the ower of aointed leader and about the tyranny of the majority. Those who oose them view the (simle) majority rule as the essence of democracy. The real world examle closest to the letter of the model would be a oular referendum on olicy, an institution that is however, reasonably seldom used. In this case the most extreme form of non insulation would be a referendum that requires a majority of 00 er cent to ass legislation, so that any individual voter can block olicy. This institutional arrangement would set exroriation to zero, but would make it imossible to ass any legislation which is not a Pareto imrovement. Given that referendum is rarely used, in the majority of institutional settings blocking takes lace indirectly, within the institutional structure of delegation. In the case of Presidential regimes like the US, one can view the Presidential- Congressional relationshi as a key element of the system of checks and balances 6. In arliamentary democracies the question of insulation refers to the control over the ower of the Prime minister and the relationshi between majority and minority in arliament. For given size of the arliamentary majority the ower of the executive, the agenda setter, is also determined by the voting rules within the arliament, an issue that has received much discussion in the literature 7. Various voting rules governing rocedures within legislatures can be interreted as giving more or less insulation to the executive, i.e. in most cases the leader whoholdsamajority 8. For examle, an imortant distinction is one between oen rules and closed rules in arliamentary voting. With oen rules the legislature has a vast latitude in amending olicy roosals of the agenda setter (the government); with closed rules the government can revent amendments to its roosals and, as a result, it has a larger strategic ower. One may a riori associate oen rules with low insulation (low M) and closed rules with high insulation (high M), since they imly different degrees of strategic ower for the executive. A vast literature on fiscal institutions, emhasizes the different effects on fiscal olicy of hierarchical versus horizontal systems. Key elements defining the former are rules that make it easier for the executive to overcome Parliamentary oosition in assing the budget. On the contrary horizontal institutions are those for which at every stage of the rocess large consensus is required to ass fiscal olicy decisions. The evidence shows that hierarchical institutions are associated with faster and more efficient fiscal reforms and fiscal adjustments when needed 9. Similar arguments aly to fast 6 This is a oint already made by Hayek (960). See Alesina and Rosenthal (995) for an extensive formal discussion of this issue. 7 See for instance Baron and Ferejohn (989) and Baron (99). On bicameralism see Diermeier and Myerson (995). 8 In some cases we can have minority governments, in which the executive does not command a simle majority in the legislature. See Persson and Tabellini (2000) 9 See the volume edited by Poterba and Von Hagen (999) for an extensive discussion of 7

8 track legislation in trade. This rocedure is viewed in the US as critical for the imlementation of free trade agreements, which otherwise might be blocked by various secial interests. 0 Another imortant element of insulation refers to the role of the judiciary. A well functioning and truly indeendent judiciary system can block reforms when they deart from roer constitutional grounds. La Porta et al. (2002) distinguish between a British style and American style judiciary as a guarantee of freedom, as in Hayek (960). The former restricts the ower of the ruler to interfere with the administration of justice, the latter gives more ower to the judiciary by allowing it to interfere more in the legislative rocess by checking its adherence to the will of the eole sanctioned by the constitution. In fact, the role of the Courts in American history has been extensive. In a famous case, in 893 the Sureme Court blocked the introduction of a federal income tax and it took the Sixteenth amendment of the Constitution, almost 20 years later, to overcome this block. Skocol (992) discusses how the role of the Courts in US history influenced and shaed the evolution of its welfare state through a series of blocks of welfare olicies in defense of roerty. Finally, with regard to the role of legislative institutions, a broad interretation of M could include a comarison of different electoral rules. Proortional rules tend to roduce olitical systems in which governing by coalition is the norm, rather than the excetion. In majoritarian systems, the majority arty can govern with fewer constraints. 2 Even more broadly, one could also use M to comare dictatorshi or oligarchy versus fuller democracy. In a sense, one can think of a dictatorshi as a system in which a ruler, when in office (no matter how he gets there), is uncontrolled, while an essential element of democracy is some sort of checks and balances on the oliticians, above and beyond the fact that the latter are elected. Finally some readers may find that we simlify too much and that we ignore too many details of institutional design. There are two answer to this criticism. A more aologetic one is to say that one has to start with a simle model and further research will add comlications and institutional details. We discus some of these issues in the Conclusion. The more aggressive resonse is that, in fact erhas the details of institutional design do not matter that much and the fundamental issues of suermajorities, insulation and veto ower are critical these issues. The choice of status-quo oint in case the arliamentary negotiation rocess fails, for examle when discussing the budget, also affect the extent to which the olitical system insulates agenda-setters; see Aghion and Bolton (997) for a detailed discussion on this oint, with reference to the constitutional change in France in See Grossman and Helman (200). La Porta et al. (200) classify 7 constitution along the British- American dimension and find that the American system is a better redictor of olitical freedom, while the British system is a better redictor of economic freedom. 2 Persson and Tabellini (200) and Milesi-Ferretti, Perotti and Rostagno (2002) resent recent studies which comare roortional versus majoritarian systems concerning fiscal olicy choices. 8

9 regardless of the details of different systems. We believe that there is an element oftruthinbotharguments. 3 Solution of the Model We roceed by backward induction. In stage vi) the voters with low λ e i oose the reform; those with high λ e i favor it. A cutoff oint divides these voters: bλ = γ (2) The realization of a, for given M, determines whether or not the reform asses or not. The reform will ass if and only if: bλ (λ + a) <M l or a> λ b λ lm. Therefore, ex ante the exected utility of the generic voter, who is behind a veil of ignorance, is given by: bλ λ lm Z Z max A M 2A da + (λ m + a)γ 2A da ( )bm (3) A bλ λ lm Obviously for λ λ lm b > A the reform would never ass, while for λ λ lm b < A the reform would always ass. The first two terms in (3) reresent the exected benefits of the socially efficient reform (multilied by the exogenouslygiven robability of such an event, ), the second the exroriation. Note that ex ante behind a veil of ignorance and with risk neutrality the generic voter acts as the average voter. Looking first for an interior solution to maximizing (3) relative to M, and remembering that λ b =/γ, we obtain, after straightforward maximization: M = 2Ab 2 l 2 (4) γ whenever the RHS of (4) is ositive. If the RHS of (4) is negative, then the otimal insulation level will be M =0. One can then immediately establish: Proosition The references of voters are single eaked on M and the otimal degree of insulation is either zero or it is interior to the interval (0,) and equal to the 9

10 exression above if ositive. In the latter case, the following comarative static roerties hold: dm dm db < 0; dγ > 0; dm d > 0; dm dm dl > 0; da < 0. Proof: By insection. Several comments are in order.. First, note that in the absence of exroriation (b = 0) (or with no bad leaders, =)wehave: M =/2. This result follows from risk-neutrality and the utilitarian nature of the maximization roblem in (3) but it extends to more general distributions of idiosyncratic and aggregate shocks on references, as we show in Aendix. The intuition is as follows 3. Suose there were only two alternatives, x and y to be chosen between ex ost. Ex ante the individuals in the constituency (of size normalized to ) are under the veil of ignorance and do not know whether they will refer x or y. Suose k is the number of individuals that refer x to y. If an individual has ex ost income α if her referred alternative is selected and income ( β) if the other alternative is chosen, then under risk-neutrality the ex ante total utility of choosing alternative x, is equal to: U(x) =αk β( k); similarly: U(y) = βk + α( k) if alternative y is chosen. Choosing M =/2 will then guarantee that the alternative that maximizes total ex ante utility is always chosen, namely x whenever k>/2, and y otherwise. This reasoning extends to collective decision roblems like ours that boils down to a utilitarian maximization roblem: if M > /2 (res. M < /2) then reforms would take lace too (res. not sufficiently) often from the standoint where voters exect their references for reform to lie ex ost. 2. Insulation is decreasing in the robability ( ) ofexroriationandin the loss b from it. Thus, low rotection of roerty rights (i.e. higher scoe for exroriation) would require lower insulation. 3. Insulation is increasing in γ, the average benefit of the reform. With more exected benefit from the reform, the voter behind a veil of ignorance is willing to accet a higher risk of exroriation in order to increase the robability that the reform asses. 3 This was kindly suggested to us by Matthew Jackson and it builds on Rae (969). 0

11 4. Insulation is decreasing in A and increasing in l. In order to gain intuition about these last two results it is useful to study the ex ante robability that the olicy reform asses (i.e. the robability that the fraction of individuals below the threshold is less than M). This robability φ(m) is given by: Ã! λ b λ a φ(m) = Pr M λ λ = 2 + µlm γ 2A + λ. (5) Note, first, that φ(m)/ M = l/2a is increasing in l. This is because M reresents the required fraction of individuals necessary to block the reform, so that Ml is the required number of blocking individuals. Thus, the higher l, the more an increase in the fraction M will increase the robability that the reform is not blocked, and therefore the higher the exected gain from increasing insulation. Second, φ(m)/ M = l/2a is decreasing in A. This can be interreted as a status-quo bias effect of uncertainty. In order to gain intuition, consider first the secial case where l = λ =andm = γ =/2; in this case, the reform will be not blocked whenever a>0, that is with robability /2, for any admissible value of A. Next, suose that M =/2 > γ ; then, the reform will ass for all a s such that a +/2 > γ, that is for all ositive realizations of a and also for a ( γ /2, 0); the higher A the smaller the set ( γ /2,A] relative to the overall suort [ A, A]; in other words, higher aggregate uncertainty will increase the relative weight of blocking losers among the whole set of voters; similarly, when M>/2 > γ, then the reform will ass for all a s such that a+ M> γ, that is for all ositive realizations of a and also for a ( γ M,0); once again, the higher A, the smaller the set ( γ M,A] relative to the overall suort [ A, A]. So, to the extent that with no aggregate uncertainty (A = 0) and for given M the reform would not be blocked, then more uncertainty decreases the effect of increasing M; that is it increases a status quo bias 4. The statusquo bias effect in turn imlies that an increase in aggregate uncertainty of the reform outcome, measured by A, should reduce insulation: a higher A reduces the extent to which increasing insulation hels increasing the robability of reform while the exroriation cost of increasing insulation remains unaffected by A. This last comarative statics result may exlain why decisions such as the signing of new international treaties or changes in the Constitution often entail 4 For a different and insightful model of status quo bias in olicy reforms see Fernandez and Rodrik (990)

12 lower olitical insulation than more routine decisions. For examle, while enlargement decisions require unanimity vote within the EU council of ministers, ordinary decisions only require a qualified majority vote. 4 Extensions 4. Polarization of Preferences Let us now investigate the case of a non linear distribution of λ i. In order to kee things simle we introduce a very stylized form of non linearity in the distribution, namely, we assume that a oint mass ( /2) is now added to the two extremes of the distribution [λ, λ]. Obviously the median and average oint of the distribution, λ m, does not change. The reform will not be blocked if and only if: bλ λ a ( 2 )+ M l At the constitutional stage the otimal choice of insulation will solve: ( max M = max à Z bλ λ (M )l ( 2 ) A Z A 2A da + bλ λ (M )l ( 2 ) (λ m + a)γ 2A da ( λ b λ (M )l ( 2 ) + A)+(A λ b + λ + (M )l à ( 2 ) )λ mγ A bλ λ (M )l 2! ( 2 ) 2 γ 2A bm The first order condition for this maximization imlies:! ) ( )bm M = 2Ab 2 l 2 ( 2 ) 2. γ Proosition 2 The otimal degree of insulation within a system deends ositively on the olarization arameter : dm d > 0. Proof: By insection. The imlication of this result is that more olarization of references lead to more insulation. To better understand this result, consider for a moment the alternative case where olarization only occurs at the bottom of the reference distribution, i.e. where there is a robability mass of at λ i = λ only. In this case the reform would not be blocked ex ost whenever: bλ λ a ( )+ M, l 2.

13 The analogous maximization as above leads to the first order condition: µ µ ((λ m ( )+λ ) γ )l + lγ bλ λ ( ) (M ) l ( ) 2A b =0. In this case an increase in the degree of olarization would have an ambiguous effect on insulation. On the one hand, if we abstracted from its negative effect on the exected return from the reform (λ m ( )+λ ), an increase in olarization would increase insulation as formally roven in Aendix. This choice would be justified to avoid oosition to reform by the lower tail of the reference distribution. On the other hand, the fact that more olarization at the bottom also reduces the exected reform outcome leads to an ambiguous overall effect of downward olarization on insulation; this latter effect is eliminated in the case analyzed above where olarization occurs symmetrically at both ends of the reference interval and individuals are risk-neutral. In the next section we shall see how olarization interacts with risk aversion. 4.2 Risk Aversion Let us now return to the basic model, with no olarization of references, that is =0, and a uniform distribution of λ i, but let us add a very simle form of risk aversion, with ex ost individual utilities being given by: ½ yi if y u(y i )= i θ, u otherwise, where y i is ex ost income and where u>0. Thus, only if income is above a threshold θ wehavethesameutilityasinthebasicmodel. Weassumethatthe status quo outcome is always above such threshold (i.e. θ < ) linking more tidily risk aversion and reform. Moreover, for simlicity we take the robability of a bad reform to be zero, i.e. =. Incidentally, this also shows that with risk aversion one obtains a well defined interior solution for M even without exroriation. max M In this case the choice of M will solve the following roblem: ( Z bλ λ lm Z Ã A Z θ 2A da + γ a Z λ u l dλ i + A bλ λ lm λ (λ i + a) γ θ γ a l dλ i! ) 2A da (6) given that, in case of reform, the bad outcome occurs to every i for whom (λ i + a) γ < θ. To further simlify the algebra we assume the threshold θ to be 0; in this case sufficient conditions to have risk aversion oerative are that λ 0 and that the reform is ex ante efficient, i.e. λ m γ >. Such restriction on the arameter sace are meant to make the analysis under risk aversion meaningful. Tedious, but straightforward maximization leads to the following: M = +lγ + u (2uγl + u 2 +2γl) lγ 3

14 Proosition 3 The otimal degree of insulation within a system deends negatively on the risk aversion arameter u : dm du < 0. Proof: See Aendix. Thus, more risk aversion leads to lower insulation: in choosing insulation, ex ante the voter takes into account the risk of falling below 0 ex ost. Thus, more risk aversion leads to choosing a system were ex ost olicy reforms can be more easily blocked. Now let us reintroduce olarization of references with a ositive mass of individuals at both ends of the interval [λ, λ], as in the revious subsection. Also let us set for simlicity such an interval to be symmetric around zero, i.e. λ = λ. Proosition 4 For sufficiently large degree of risk-aversion as measured by u, more olarization reduces insulation: dm,u d < 0. where M,u is the otimal degree of insulation for this case which is exlicitly defined in Aendix. Proof: See Aendix. The intuition for this result is straightforward: more olarization increases the risk of ending u at the bottom of the reference distribution, which in turn leads to a low income when the reform is imlemented; reducing insulation will limit that risk. 4.3 Comensation In general, those who are net losers from a olicy reform can be comensated by transfers, even though the latter will generally induce welfare costs, such as the costs of distortionary taxation. Suose that after M is chosen, a fixed amount of resources ω can be raised from all individuals through taxes, and assume for simlicity that taxes are raised before the idiosyncratic and aggregate shocks on references are realized. Note that this imlicitly assumes that any increase in income obtained through the reform cannot be used to comensate, since the amount available for comensation is fixed ex ante. The maximum amount of ω is w which reresents initial individual wealth. The only urose of taxation is to comensate losers for the reform, which means that if the available resources for comensation exceed the needs they are returned lum sum at no cost and the leader does not retain any revenue for rivate benefits. The oeration of this transfer scheme involves a ositive deadweight cost k er unit of taxed funds, and the net tax revenues are used by the olitician to comensate the required number of losers in order to avoid blocking. Ex ost, for 4

15 given realization of the aggregate shock a, either more than ( M) individuals are willing to suort the reform even without any comensation (this will be the case whenever bλ a λ <M), in which case no comensation will take lace; λ λ or assing the reform requires comensations to be made (this will be the case when bλ a λ λ λ bλ a λ >M: the olitician needs to comensate the fraction M λ λ of individuals for otential loss of utility due to the reform). Comensations are aid to enough individuals who would, ex ost, vote against the olicy reform, in order to kee them in. Obviously, the cheaer individuals are comensated, i.e., those closer to the cut oint of indifference between having or not having the reform. This, however, requires that the references of individual are ex ost observable, since comensations are made deendent on them. If references are not observable, either transfers scheme are not oerational or they require some revelation mechanism. We do not exlore this question here. Thus the total amount of comensation needed to ass a reform is given by: c(a) = bλ a Z λ+lm ( b λ λ i a) l dλ i Obviously if λ + lm > b λ a, no comensation is needed. Straightforward maniulation leads to: c(a) = (b λ λ lm a) 2 2l A olicy reform will ass with comensation aid if and only if: ( + k)c(a) ω Two cases must be considered: () λ b λ lm + A<³ 2lω 2, which will be true whenever ω is sufficiently +k large (i.e. enough funds have been raised); in this case reform will always take lace (as it will be always affordable) and, at the constitutional stage behind a veil of ignorance, the generic individual will choose M in order to maximize: max { (λ m γ ke a c(a)) ( )bm} where E a c(a) the exected comensation costs are given by: ³ Z bλ λ lm bλ λ lm a 2 E a c(a) = 2l 2A da A = (b λ λ lm + A) 3 2Al 5

16 The solution of this roblem leads to (2) λ b λ lm + A> becomes 5 : max M à R bλ λ lm ( 2lω +k) 2 A λ b λ + A 2 M() = l q Ab k ³ 2lω 2 +k, in which case the roblem for the voter 2A da + R A bλ λ lm ( +k) 2lω (λ 2 m + a)γ 2A da k 2Al ( ) bm and, similarly as before, the maximization roblem imlies: ³ (7) 2lω +k 3 2! M (2) = 2 2Ab l 2 γ µ 2ω 2 (8) l ( + k) Therefore, from the results in cases () and (2), the following result holds 6 : 5 The third term in the arentheses multilied by is the exected deadweight loss from comensation, calculated as: E a c(a) = 2l = Z bλ λ lm 2 bλ λ lm 2lω bλ λ lm a 2 2A da +k ³ 2lω 3 2, 2Al +k while the first two terms reresent the usual exected value of income. 6 In case () a sufficient condition for individuals to ot for comensations at the constitutional state instead of the no-comensation solution analyzed in the revious section, is: à r λ mγ b 4 b λ λ + A l 3! Ab > ψ λ, λ,a,γ,,b k where ψ λ, λ,a,γ,,b = Ãb λ λ l à + A Ab l 2 γ µb λ λ l µ! + A! 2 2Ab l 2 γ 2 + ÃA b λ + λ + l à 2 γ 2Ab 2 2Ab l 2 γ Ã!! 2 2Ab l 2 γ λ m γ! < Let us note that the LHS is not defined in k = 0, but it is continuous in k for any k>0. If we consider the limit k 0, it is ossible to show that LHS diverges to +. Hence, by 6

17 Proosition 5 When the comensation scheme is available and k is not too large, the otimal degree of insulation M (either M() or M (2) ) increases with the taxation cost k; otherwise it satisfiesthesamecomarativestaticsroertiesasinproosition. Proof: By insection. The basic message of this roosition is that a less efficient system of transfers should lead to a choice of more insulation (higher M) inordertoreduce the need for comensation ex ost. 4.4 A sequence of reforms Consider a simle infinite horizon extension of our model in which the economy is oulated by a continuum of non-overlaing dynasties of one-eriod lived individuals. Each individual maximizes his current utility, and the game analyzed above is layed reeatedly over time; in articular elections are held every eriod and there is a countably infinite set of reforms of vintage v, v +, v +2,... that can be imlemented sequentially. So, income of individual i is as follows: ½ λ e ey i = i γ ν γ v if v reform occurs otherwise with γ > 0 (9) Since in each eriod all layers face the same choices and the same objective functions, excet for a multilicative constant, the otimal choice of M in this dynamic environment will be the same M as above in each eriod, as long as it is taken before the realization of the shock. The average innovation rate of our infinite horizon economy is then simly equal to the average frequency of innovations φ(m ), where M is given by (4), namely: g = 2 + 2A (λ m γ ) b lγ. In articular, we see that the equilibrium rate of reform g decreases with the robability and the cost of exroriation, with the degree of aggregate uncertainty, A, (in the case of ex ante efficient reforms) and it increases with the average value added of a good reform as measured by γ. It is also ossible to continuity and the algebraic result that LHS is decreasing in k, it follows that there exists a k such that comensation will be a viable alternative k (0, k]. In case (2) a sufficient condition for individuals to ot for comensations at the constitutional state instead of the no-comensation solution analyzed in the revious section, is: ( λ mγ) l + l2 γ 4 r 2ω +2Ab > 0 l obtained by substituting into (3) and (7) the otimal level degree of insulation in both cases, resectively (4) and (8). 7

18 comute the reform rate for the cases of olarization, risk aversion, and comensation, and to derive the comarative statics roerties. One interesting result is that in the resence of a sufficiently high degree of risk aversion (u) the effect of olarization is negative on the degree of olicy reform. 5 Remarks on the Political Economy of Constitutions Thus far we have examined the case of a erfect veil of ignorance, behind which everybody is identical. This, in a sense, is equivalent to a normative model of constitutional writing. In reality, Constitutions are not written by social lanners, and veils of ignorance have holes in them. In fact, in virtually every instance of Constitutional reform, a large amount of bargaining and conflict occurs at the Constitutional table. One simle way of caturing the comlexity of the olitical economy of writing Constitutions is to generalize our model by assuming that not everybody derives the same (known) utility from the reform. For examle, assume that individuals differ ex ante with regard to their exroriable wealth, namely a b i -individual exects to be exroriated by an amount equal to ( )b i M on average. We have already seen in Proosition that the otimal degree of insulation is negatively correlated with the scoe for exroriation, b i. It then immediately follows that an individual i with higher b i at the Constitutional stage will otimally choose a lower level of insulation, with: M (b i )= 2 2Ab ( ) i, and that individual i s reference for insulation is single eaked around this maximum. This in turn has interesting imlications for the olitical economy of Constitutional writing. Consider a symmetric distribution of exroriation losses b i between b and b and suose that the Constitution is decided by majority rule. In this case, the median voter, b m = b+b 2, will revail and imose her most referred level of insulation, namely: M (b m )= 2 2Ab m ( ). Alternatively, if M had to be chosen by unanimity, any M>M (b) wouldbe vetoed by high-b individuals. But individuals with low b, thosewhobenefit more from the reform (since they exect lower exroriation) may want to comensate the high b-tyes to imose their referred level of insulation. More generally, the outcome of the bargaining rocess at the Constitutional table will deend on the bargaining rules, the suort of the b i distribution, and the voting rule on M. 8

19 Another examle of interest concerns the writing of Constitutions by a minority of wealthy individuals who might worry about the ossibility that new redistributive olicies be introduced by future majorities as the extension of voting rights rogresses. In the language of our model this can be interreted as follows. Suose that individual wealth is heterogeneous across individuals and distributed between w and w with density f(w). Suose also that the exroriation rate is the same b for all individuals, and that those who decide on the Constitution lie in the uer art of the wealth distribution, say between some w m and w, withw m >w. Assuming that all individuals with wealth w [w,w] vote ex ost, clearly the Constitution designers will choose a lower degree of insulation than if the franchise was not to be extended, because they exect more exroriation as a result of the franchise being extended. In other words, Constitutions written with an eye on defending roerty rights against future redistribution of exroriation will include a number of checks and balances and require suermajorities, i.e. low insulation to ass legislation. How should the voting rule on M be chosen when Constitution designers are ex ante heterogeneous? One could also think of some sort of fixed oint argument in voting rules, that is, one may want to argue that a choice of M has to be aroved itself with a blocking rule M. That is, a Constitutional choice of M can be vetoed only by a M (suer) majority. This is exactly the aroach taken in models by Barbera and Jackson (200) and Polborn and Messner (2002). While this self-stable solution is very elegant, its realism may be called into question. In fact, voting rules and rocedures to select or change the Constitution are generally different from the rules regulating the assage of normal legislation. In general, the blocking coalitions needed to revent changes in the Constitution are lower than those required to block normal legislation. In fact, our model, and more secifically Section 4.2 on risk-aversion, suggest one ossible reason why changing the Constitution would require smaller blocking coalition (larger majorities): Constitutional change may bring about a more uncertain distribution of winners and losers and voters may be esecially risk averse concerning radical changes of the rules of the game. Another reason might be the need to revent an elected leader from easily changing the rules of the game restricting entry of cometitors. 6 Emirical Imlications and Discussion In what follows we discuss several emirical imlications of our theoretical analysis. Rather than a formal test of the model we highlight several of its imlications that seem to shed light on some asects of institutional choices and economic develoment. 9

20 6. Economic Develoment and Institutions A well known feature of develoing countries is that they have not well functioning fiscal systems. The share of transfers of GDP is larger in OECD countries than in develoing countries, and more generally, the role of government in transferring resources across individuals, the welfare state, is much more widesread in richer countries. In the nineties the average for subsidies and other current transfers as ercentage of the current exenditure in the highincome countries samle (World Develoment Reort , Word Bank) was about 60 ercent. In the lower middle income countries it was 8 ercent cent in 990 and 26 ercent in 997. Part of the reason is that it is easier to collect taxes in more advanced industrial countries and also targeting towards the truly deserving is esecially difficult. These considerations suggest that develoing countries should adot more insulated systems of government, since, in the language of our model they have a high k. On the other hand, roerty rights tend to be less rotected in develoing countries, and insulated leaders may have more latitude to ursue olicies which favor the leader himself and its close allies. The otential for exroriation, broadly defined, is larger in develoing countries. This is catured by a higher b in our model. One may argue that the technology of exroriation and of taxing for comensation go hand in hand, to the extent that they both involve collecting fiscal revenues. However, a comensation scheme involves a fairly sohisticated system of targeting, while exroriation, esecially in its more brutal form, can be rather easy to accomlish to the extent that the government has the monooly of coercion, a monooly which will itself increase with more insulation. These features ut develoing countries between an institutional rock and a hard lace. High insulation leads to high exroriation. Low insulation means that olicy reform are not imlemented. In fact, we believe that this stee trade-off between the ossibility to imlement reforms with winners and losers and the likelihood that insulated leaders turn into dictators, may be one of the key reasons for institutional failures in develoing countries. 6.2 Polarization, fragmentation and Institutional Choice Our analysis of the effects of olarization on olitical insulation in Section 4 shows that, when constitutional designers are not too risk averse, an increase in olarization would lead them to increase insulation. This would haen in order to limit the scoe for ex ost blocking by the lower tail of the distribution of references over the reform. In addition, our discussion of the olitical economy of Constitutional choice with an imerfect veil of ignorance in Section 5, suggests that if a grou which is not behind a veil of ignorance, knows that the chosen olicies will be in its favor (for instance because it is olitically dominant), then 20

21 again it will choose more insulated systems. 7 This subsection briefly exlores these two oints emirically. Measurement issues are difficult. With regard to the main exlanatory variable, a articularly relevant, and reasonably exogenous, measure of olitical olarization is an index of racial fragmentation, widely used in the literature to exlain economic erformance. 8 We use three indices. Two are the traditional ones used originally by Easterly and Levine (997) and by many others after them. One is an index based on an historical Russian classification of languages; thesecondone(whichwelabelavelf)isanindexroosedbyeasterlyand Levine which is a average of five related linguistic indices. The third index is new and recently constructed by Alesina et al. (2002) by combining linguistic measures with other ethnic variables, like racial origin. In many ways we feel that this new index imroves over the linguistic ones, although, in any case, our results are robust to the use of all three indexes. In all cases the index of fractionalization is comuted (as it is standard in the literature) as the robability that two randomly drawn individuals form the oulation of the country belong to two different grous. It is almost imossible to come u with an undisutable measure of insulation for a samle of more than a hundred countries in which institutional arrangements vary on many dimensions. In Table 9 we choose two aroximations of insulation. The first one is a simle dichotomy democracy vs. autocracy as the most general aroximation for the searation between not insulated and insulated countries. The second one refers to democratic forms of government and we assign to residential systems the role of most insulated form of government, Semi-residential the middle level, and arliamentary systems the least insulated. This choice is justified by the substantial reduction in the number of veto layers within residential systems and their intrinsic winnertake-all nature, which distinguishes them from arliamentary systems 20. In fact we checked that the unconditional correlation between the form of government from Presidential to Semi-residential to arliamentary and the Freedom House index of olitical freedom is -.5, significant at the er cent level, for the samle of countries used in Table and 3 below. That is more residential regimes are associated with lower olitical rights, which we interret as a roxy for more insulation. 2 Results by Persson and Tabellini (200) and Fatas and Mihov (2002) suggest that fiscal olicy in residential regimes has a more 7 This effect may be moderated by risk aversion as discussed in section See for instance, Easterly and Levine (997), La Porta et al. (998), Alesina et al. (2002). 9 A full descrition of data sources is given in the Aendix. 20 For examle, there is a relevant, negative (-0.50), and significant (at ercent level) correlation between residentialism and a measure of olitical constraints within the olitical arena (as resented in Henisz, 2002). For more discussion on insulation and residentialism, see Shugart and Carey (992). 2 Further analysis available uon request shows that this correlation holds u even when controlling for a battery of other variables, including log of er caita GDP, school enrollment ratios, regional dummies, oenness. 2