A Theory of Competitive Authoritarian Elections

A Theory of Competitive Authoritarian Elections Mario L. Chacón April, 2012 Abstract This paper develops a model to study the effects of electoral competition in nondemocratic regimes. In this model, an authoritarian government can introduce a system of fraudulent elections anticipating a democratic transition. If this government allows this sort of competition, it will undertake some investments, a portion of which are sunk costs, in setting up the organizational capacity to mobilize new voters. This organization is complementary to the institutions of democracy, hence, once a reform has occurred, the authoritarian elite is more willing to tolerate democracy in the future. Comparative statics suggests that when elections under a dictatorship are more competitive, democracy is more likely. This result is investigated empirically using a panel of countries from 1972 to 2002. The evidence shows that, controlling for time invariant differences as well as global and regional trends, an increase in the competitiveness of the electoral system in a dictatorship increases the probability of a transition to democracy in the following period. This effect is robust and consistent with the motivating theory. I would like to thank Elizabeth Dickinson, Thad Dunning, James Robinson, John Roemer, Frances Rosenbluth, Nicholas Sambanis, Jonah Schulhofer-Wohl, Seok-ju Cho, Ragnar Torvik and Alex Debs for their advice and helpful comments. I also thank Pablo Beramendi, Bruce Bueno de Mesquita and other workshop participants at Yale, MIT, Duke and New York University for comments. New York University Abu Dhabi and New York University, 19 w 4th St, New York NY 10012. E-mail: mario.chacon@nyu.edu. 1

1 Introduction The last quarter of the twentieth century was a period of political liberalization across the world. This transformation included not only an important number of political systems that transitioned from nondemocratic to democratic governments, but also, minor liberalizations in countries that did not become fully democratic (Huntington 1991: 14-15). For example, the proportion of regimes having an elected legislature but widely considered as nondemocratic went from 58 percent in 1975 to more than 80 percent in 2002 (Keefer 2007). Similarly, the proportion of nondemocratic regimes with more than one autonomous party in the legislature rose from 20 percent in 1972 to 63 percent in 1996 (Gandhi and Przeworski 2006). Hence, a salient empirical regularity resulting from this global trend is the proliferation of regimes that combine seemingly democratic institutions with elements of authoritarianism. 1 An important question is how these reforms, which have taken place in nondemocratic societies, affect the likelihood of future regime change and democratization. Recent research on comparative politics offers different interpretations for seemingly democratic institutions and different conclusions about how these reforms affect the future of democracy. On one hand, institutions promoting the participation and representation of opposition sectors in authoritarian regimes are interpreted as commitments to policy concessions (Magaloni 2006; Gandhi and Przeworski 2007; Boix and Svolik 2008). Since these policy and material concessions are sufficient to maintain or co-opt a ruling coalition, a liberalization should enhance the political survival of the regime, making democracy less likely. 2 Yet on the other hand, some works have defined liberalizations under authoritarian rule as the new mode of democratic transition. These studies highlight how elections empower the opposition and may precipitate a popular movement that can actually overthrow the existing autocratic regime (see e.g., Lindberg 2009 and Schedler 2006). Hence, according to this view, political liberalizations should have a negative effect on the stability of such regimes. The empirical literature addressing this question is also is inconclusive. On one hand, 1 The emergence and proliferation of this type of regimes is documented by Carothers (2002), Diamond (2002), Gandhi (2008:Ch 1), Hadenious and Teorell (2007), Huntington (1991), Levitsky and Way (2002), Ottaway (2003) and Schedler (2002,2006a,2006b) among others. 2 The idea that some political institutions during dictatorship can be used to co-opt the opposition is also suggested in the early literature on authoritarian regimes (e.g., Linz 1973, 1975, O Donnell 1973 and Hermet, Rose, and Rouquié 1978). 2

during the post-war period, legislatures and party competition do not seem to affect the tenure of dictators (Gandhi 2008:175-177), or if they do, they have a negative effect on the lifespan of the regime (Hadenious and Teorell 2007:151). Second, cross-sectional analyses for the period 1972-2002 indicate that autocratic governments without a single dominant party are more likely than other nondemocratic regimes to become democracies (Hadenious and Teorell 2007; Brownlee 2009). Lastly, recent research focusing solely on the effect of elections on regime change finds evidence consistent with the theory that elections during dictatorship may even promote democratization (Lindberg 2006, 2007; Howard and Roessler 2006). In this paper, I develop a framework that reconciles the role of seemingly democratic institutions in authoritarian regimes and the evidence suggesting that different types of dictatorships have different propensities for democratization. In this framework I emphasize three key ideas. First, in an eventual democratization, any incumbent elite in a nondemocratic regime would want to preserve their power by successfully contesting elections. Second, given that this elite would try to participate in the new electoral system, their existing party organization will influence their incentives to accept a real democratization. And lastly, I argue that the establishment of a party organization with strong ties to the electorate is associated with irreversible investments that are specific to electoral politics. Hence, the political impact of competitive authoritarian institutions is mediated by the effectiveness and complementarity of previous party investments, and at the same time, these are shaped by the electoral competitiveness of the system. To make these ideas precise, I develop a model of authoritarian politics that combines elements of the probabilistic voting model (Lindbeck and Weibull, 1987) with elements of the democratization model pioneered by Acemoglu and Robinson (2000, 2001). In this setting, an authoritarian regime can engineer a system of uneven elections anticipating a democratization in the future. If this form of electoral competition is allowed, the incumbent elite in this regime will undertake some investments, a portion of which are sunk costs, in setting up the organizational capacity to mobilize new voters. A liberalized regime is not stable however since opponents can organize and demand democracy. In the case in which democracy is demanded, any authoritarian government can repress the opposition or it can democratize. And in the latter case, the ancien régime elite will try to compete in the system of free and fair elections. 3

The model shows how, under certain circumstances, the introduction of electoral competition during dictatorship induces history dependence in the sense that the decision to allow or block democracy is influenced by past actions. Specifically, in this cases, a democratic transition occurs only if and only if the authoritarian regime was liberalized in the past. The intuition for this result is straightforward; since some of the organizations created during a liberalization are complementary to the institutions of democracy, a ruling elite in a liberalized dictatorship is more willing to support a democratic system. In addition to showing exactly how different institutions under dictatorship influence the incentives of the regime, the model offers a number of comparative static results. One that is particularly noteworthy is the relationships between the degree of electoral competitiveness in a liberalized dictatorship and the emergence of democracy. The model suggests that liberalizations that induce greater levels of electoral competition are more likely to influence the decision to democratize in the future. This is the case because the differential value between a closed and a liberalized dictatorship increases the more competitive elections are. Thus, the lock-in effect of having a liberalized dictatorship, as opposed to a fully authoritarian regime, is greater for cases in which elections under dictatorship are more fair and competitive. Consequently, the competitiveness of the electoral system conditions the effect of liberalizations on regime change. Lastly, the paper investigates empirically these predictions. This analysis focuses on the conditions under which electoral competition under dictatorship influences the likelihood of democratization. I explore these conditions using panel data from a cross-section of nondemocratic regimes across the world during the post-cold War period. My empirical strategy is to exploit the within-country variation in the data, controlling for timeconstant unobserved effects. This strategy is particularly useful to eliminate a potential source of omitted-variable bias in an econometric model of transitions and institutional characteristics. 3 The econometric results shows that, controlling for these country-specific effects, an 3 Even if within-country comparisons are an important improvement over the pure cross-sectional analyses prevalent in the literature, the estimates reported may not represent the causal effect of electoral competitiveness on democratic transitions. The presence of time-varying omitted variables affecting the likelihood of transition and correlated with the main explanatory variable cannot be ruled out. Nevertheless, the conditional correlations uncovered are robust to global and region-specific time trends in democracy and to some of the time-varying covariates explored and tested in the literature. 4

increase in the competitiveness of the electoral system in a dictatorship has a positive and significant effect on the probability of a transition to democracy in the following period. This means that reforms allowing electoral competition and representation in nondemocratic regimes have a positive effect on the likelihood of democratization. This result is consistent with the motivating theory and robust across samples, regime classifications and model specifications. The framework developed contributes to recent formal works on the role of political institutions under dictatorship (see e.g., Gandhi and Przeworski 2006, Myerson 2008, and Boix and Svolik 2010). These models focus on the stabilizing effect of power sharing institutions without considering the dynamics of electoral competition and regime change. Following earlier models on the creation and consolidation of democracy (e.g., North and Weingast 1989, Acemoglu and Robinson 2000), the common element in these formalizations is the capacity of democratic institutions to mitigate commitment problems between autocratic leaders and their supporters (or their opponents). 4 Then, issues such as the conditions under which a political reform under dictatorship influences the incentives to democratize, or how the level of electoral competition affects the organization of the party system, are not discussed in these works. Moreover, these analyses neglect the role of authoritarian elites in new democracies, a key element in the model advanced in this paper. In addition to the works cited above, this paper is related to a number of formal models of democratization such as Feng and Zak (1999), Rosendorff (2001), Conley and Timimi (2001) and Lizzeri and Persico (2004). These works focus on the economic conditions under which a rich elite will find optimal to extend voting rights to poor individuals. Yet none of these studies look at how different political institutions under dictatorship influence party organizations and democratization. Somewhat closer to the theory developed is that of Llavador and Oxoby (2005), who emphasize the role of elite competition over economic interests as a major factor explaining the extension of voting rights in nineteenth century Europe. Although competition between elites is an important element in my analysis, the main contribution of the model developed is to emphasize that the organizations result- 4 As I show in section 4, the framework advanced in this chapter can easily incorporate (in a reducedform way) the role of political reforms as a commitment devise used to make concessions under dictatorship credible. This role of democratic institutions can lead to some form of co-optation equilibrium in a competitive authoritarian regime, in which opponents are less likely to rebel. This equilibrium is derived in an extended version of the baseline model presented in sections 2 and 3. 5

ing from this competition may influence the intertemporal incentives of a nondemocratic regime. To the best of my knowledge, no previous study has formalized the role of party organizations and irreversible investments in a context of democratization. This paper also relates to a large literature in political science studying the dynamics of authoritarian politics and democratization (see, e.g., Huntington and Moore 1970; Rustow 1970; Dahl 1971; O Donnell 1973; Linz and Stepan 1996). Some of the arguments proposed in this literature are consistent with the idea that some type of electoral competition may precede a full democratization. For instance, inspired by the political histories of England and Sweden, Dahl (1971) argues that the path most likely to produce a stable transition toward democracy is one in which:...the rules, the practices, and the culture of competitive politics developed first among a small elite, and the critical transition from nonparty politics to party competition also occurred initially within the restricted group. (pp. 36) Similarly, case study works and cross national analyses about political transitions in Latin America and Africa suggest that democratization was in fact less resisted in countries with legacies of political competition (see, e.g., Collier and Collier 1991; Remmer 1989; Mainwaring 1999; Mahoney 2001). For example, explaining the domestic conditions that conditioned the democratization in a sample of African countries in the early 1990s, Bratton and van de Walle (1997) argue that Getting democracy is easier from a regime in which competition is encourage and the main challenge is to broaden participation; getting to democracy is much more difficult from a regime that has no tradition of political competition...(pp. 273) Since these studies seldom explain explicitly how competitive politics alters the incentives in a way that is conducive or detrimental to democracy, this paper complements these studies by providing explicit microfoundations and a mechanism explaining how some of these institutional features may affect the prospects of democratization or popular protest under nondemocracy. Lastly, the empirical analysis presented is related to an important number of empirical works investigating the determinants of democracy and democratic stability (see e.g., Londregan and Keith T. Poole 1996, Barro 1999, Przeworski et al. 2000). These works focus 6

mainly in the relationship between income and democracy employing pooled regressions and maximum likelihood methods. Thus, the empirical strategy used is closer to recent works such as Acemoglu et al. (2008, 2009) and Bruckner and Ciccone (2010) which employ fixed-effects models. Following the earlier empirical literature, these recent papers focus also on the causal effect of income per capita on democracy and not on the institutional determinants highlighted in this paper. The paper proceeds as follows. Section 2 presents a dynamic model formalizing the main theoretical contribution and defines the equilibrium concept used. Section 3 presents the analysis and the characterization of the equilibria and Section 4 extends the baseline model to explore the robustness of the theory. Section 5 derives an estimable model and presents some empirical evidence on the key implication of the theory. Section 6 offers some conclusions and Section 7 (the Appendix) presents all the proofs omitted in the main text. 2 The Model This section presents the basic economic and political environment. The setting features two groups competing for power under different political regimes. The initial regime is a dictatorship controlled by one of these groups. This incumbent group enjoys a full monopoly of violence such that any organized opposition can be met with repression. To analyze how an electoral liberalization influences the path of political development, I consider two different electoral regimes, one in which the electoral institutions give a disproportionate weight to the votes of supporters of the regime, and another in which elections are free and fair. In this baseline specification the organizational capacity of the opposition is invariant to the regime. Also, parties are assumed to be purely office seeking in the sense that they do not have policy preferences and only care about holding power. 2.1 The Environment Consider an infinite-horizon model of political competition between two parties, A and B. These groups have identical preferences represented by W g = E t k=0 δk c g t+k for g {A, B}, where δ (0, 1) is the common discount rate, c g t is the consumption of g at time t, and E t is a standard expectation operator (conditional on the information available at 7

time t). The economy consists simply in the production of a natural resource good which generates some exogenous rents labeled R t. All rents are collected by the government; thus the incumbent enjoys rents in the amount of R t from holding power. For simplicity, suppose R t = R > 0 for all t 0 so the production of natural resources is constant across time. The political competition between A and B is Downsian in the sense that they do not have policy preferences and only care about holding office. Thus, the utility of group g at time t is given by where π g t W g = k=0 δ k π g t+kr, (1) is the probability that group g has of being the incumbent in period t. This probability will depend on the political regime and on the decisions taken by the parties within any given period. Voters are divided in two groups, the elite, superscripted E, and the citizens (or poor), superscripted P, with a continuum of individuals within each group. Let E and P denote the set of elite and poor voters respectively. The mass of voters is normalized to unity, and the share of total voters in each group is given by λ E and λ P, where λ E < λ P < 1. These voters are short-lived players, so each generation of voters is alive for only one period. As in the standard probabilistic voting model (Lindbeck and Weibull, 1987), voters preferences consist of two components: economic and ideological. Specifically, all voters within a group share the same economic preferences, but they may differ in their ideological support for a regime ruled by any given party. Formally, the instantaneous utility of voter i in j { E, P} when party A is in power at time t is U i (q, x A ) = u j(i) (q) + f( x A x i ) + ξ t, (2) where the indicator j(i) means that i is a member of group j. u j(i) is an indirect utility function, and q Q R K is a vector of economic policies where K 1. This function is defined over the set of feasible policies, strictly concave, and differentiable. The term x A represents the ideological bliss point of party A and x i the ideological bliss point of i. I assume that voters derive utility from any ideological proximity with the party in power (i.e., f is monotone decreasing). These ideological preferences capture the non-economic 8

benefits enjoyed by all voters from having group A in power. 5 Lastly, ξ t represents the average (relative) popularity of A in the population at time t. To analyze how different political institutions influence the path of political development, I consider two types of nondemocratic regimes. In the first one, the incumbent is not accountable to the voter population and the opposition party is banned from competition. This regime represents a closed dictatorship in which electoral competition is either not allowed or very low, and is modeled in a simplified way by assuming there are no elections. The second authoritarian regime considered represents a liberalized dictatorship. In this regime, elections take place and both parties can compete for the popular vote. The main difference between this system and a democracy is that in the former, the incumbent party can manipulate the electoral institutions in a way that creates an uneven competition field. The form of this manipulation can range from electoral rules creating malapportionment to ballot stuffing during the election day. The key element in this regime however is that the electoral system installed gives a disproportionate representation to the interest of the supporters of the regime. Moreover, this system is endogenous to the preferences of the incumbent group. From now on, this regime -based on lopsided elections- is refered to as competitive authoritarian (term coined by Levitsky and Way, 2002). The dynamics of the game are such that any type of nondemocratic regime, either closed or liberalized, is not stable since the opposition and the voters can organize and demand democracy. Following the framework of Acemoglu and Robinson (2006), I assume that the mobilization capacity of the opponents of the regime is stochastic and fluctuates over time. Explicitly, in every period of nondemocracy, opponents organize against the regime with probability p, and they fail to organize with probability 1 p. Periods in which the opposition sectors organize are labeled as high threat, since during these periods the opposition can disrupt the regular functioning of the economy and force a political reform. In this baseline specification, the mobilization capacity of the opposition is independent of the political regime. Periods of high threat may not be associated with democratizations since the group enjoying the monopoly of violence can employ repression and stop the liberalizing movement. If repression is used, it is always successful, and the group repressing pays a deterministic 5 This affinity between the citizens and the groups competing for power can be explained in terms of a partisan, ethnic or religious affiliation. 9

cost κ > 0. If repression is not used, the incumbent is forced to introduce democracy, and the society remains democratic forever. For simplicity, suppose that only A can repress and that the capacity to do so is not dependent upon A holding office. 6 The political regime in every period is treated as a state variable denoted s t. This variable can take three values: dictatorship (M), competitive authoritarian (C), or democracy (D). Similarly, the organizational capacity of the opposition in any period of nondemocracy (i.e., s t = M, C) is also a state variable labeled o t {l, h}, where o t = l for periods in which the opposition fails to organize, and o t = h during periods in which it organizes. To simplify the notation, let s h denote a political state during a period in which the opposition organizes, and s l a political state in which it does not. Thus, s h (s l ) is a political regime of type s facing a high (low) outside threat. Given that the opposition can only mobilize if the state is not a democracy, the set of relevant states is then S = {M l, M h, C l, C h, D}. The transitions between states are as follows. Starting from a threatened dictatorship (i.e., s t = M h ), if there is repression, the state transits to M h with probability p and to M l with probability 1 p. Similarly, starting from a competitive authoritarian regime under threat (i.e., s t = C h ), if there is repression, the state becomes C h with probability p and C l with probability 1 p. On the other hand, if the opposition organizes and there is no repression, in either s t = M h or s t = C h, the state automatically switches to democracy which is an absorbing state. To simplify the analysis, suppose that a closed dictatorship can only be reformed into a competitive authoritarian regime during periods of low threat. Thus, starting from a low threat dictatorship (s t = M l ), if a liberalization takes place, the state transits to C l with probability 1 p and to C h with probability p. Lastly, if a low-threat dictatorship is not liberalized, the state transits to M h or to M l with the same probabilities described above. The game form and the law of motion between high and low states is depicted in Figure 1 and Figure 2. 6 This aspect of the model can be interpreted as a case in which A is a military group or as a case in which the military is a perfect agent of group A. An alternative specification is to assume some form of institutionalized repression so only the group in power can repress (case in which the military acts like a perfect agent of the regime). This alternative would complicate the analysis without changing any of the substantive results. 10

2.2 Party Investments and Electoral Competition Starting from a closed authoritarian regime, if a liberalization or a democratization occurs, both A and B are able to participate and contest power in the new electoral regime. Naturally, in order to compete and take advantage of the electoral system, these groups will have to develop party organizations capable of wining elections. The establishment of these organizations is modeled as a process analogous to the adoption of a technology. As any standard production function, this electoral technology is associated with some startup costs, labeled θ, part of which are sunk investment costs. This means that once some of these costs have been incurred, they need not to be repeated, and neither can they be recovered. These costs, loosely called party investments, represent the resources necessary to develop organizations that are specific to the electoral competition. For example, θ can be interpreted as the full costs of setting up a party organization in the ground and a clientelistic network to mobilize new voters. 7 The amount of resources necessary to establish a party organization depends on how stiff the electoral environment is. In particular, in periods in which the opposition is organized, electoral competition is more intense relative to periods in which they are not. Following the same notation used for low and high-threat periods, let θ l represent the level of party investments associated with a low-threat period and θ h the investment for a high-threat period. These investments are such that θ l < θ h, which implies that it is more costly to develop a party organization in periods in which the opposition is organized (relative to periods in which is not). Without loss of generality θ l is normalized to zero. In any period in which elections take place, parties can credibly commit to any economic policy vector q Q, but they cannot make credible commitments regarding their ideological position (represented by x A and x B ). Therefore, voters base their decision both on the economic platforms announced by the parties and on the known ideological characteristics of these two groups. Specifically, voter i will vote for A when 7 These party structures are just some of the organizations inherent to electoral competition. For instance, to guarantee the fairness and transparency of the process, the introduction of electoral competition may require the creation of an independent electoral commission or an independent judiciary. The resources invested in this particular organization are not the ones highlighted, since any incumbent would not want to invest in such entities since this would lower her payoffs during a competitive authoritarian state. 11

u j(i) (q A ) + F (x i ) + ξ t > u j(i) (q B ), (3) where q A and q B are the policy platforms offered by A and B respectively, and F (x i ) f( x A x i ) f( x B x i ). This last term, which can take negative or positive values, is an individual-specific parameter and it measures i s ideological bias toward party A. To allow for within and between-group heterogeneity in the intensity of ideological leanings, let F (x i ) be an iid random variable uniformly distributed over the group-specific support [ µj(i) φ, 1 ] µj(i), φ with density φ > 0 and µ j(i) [0, 1]. The temporal shock ξ t is also a random variable, independent and identically distributed across time, characterized by a uniform distribution with support 1, 1. 8 These [ ] 2φ 2φ distributions, and the voter s policy preferences, are common knowledge to the parties. Without loss of generality, I limit the attention to the case in which the share of poor voters that has a negative bias towards A is greater than the share of elite voters having a negative bias towards the same party. This restriction is given by: Assumption 1. µ P > µ E. Conditional on the realization of ξ t, and given the distributional assumptions, the fraction of j { E, P} voters supporting party A is where u j u j (q A ) u j (q B ). (1 µ j )/φ u j ξ t φdi = 1 µ j + φ ( u j + ξ t ), (4) A key feature is that the incumbent in a competitive authoritarian regime can manipulate the electoral institutions in a way that creates an uneven competition field. 9 means that the actual share of supporters for A and B in a competitive authoritarian state 8 This shock captures aggregate or macro uncertainty during the election at any given point in time. 9 Naturally, any incumbent in this type of regime would like to give less importance to the votes of groups having a less favorable bias against the regime and to groups favoring the opposition. This 12

may not be equal to (4). I capture this process of electoral manipulation in a reducedform way assuming that the incumbent can influence directly the proportion of supporters within each group. Specifically, an incumbent in a competitive authoritarian regime can set a binary variable γ { γ, γ} such that the effective population weights during elections become λ E = λ E + γ and λ P = λ P γ, where γ λ E and γ λ P. 10 For cases in which γ (or γ) is high, an incumbent in a competitive authoritarian regime will have more power to distort the electoral process. For instance, given that A faces a greater bias among poor voters, γ is indicative of the capacity and power that A has vis-à-vis his opponent. Given that elections can only take place under a competitive authoritarian or under a democratic regime, the electoral probabilities are simply Π C and Π D, where Π s represents party A s probability of winning in state s. Assuming a majoritarian rule and conditioning on the value of γ, A s probability of winning in a competitive authoritarian regime is then given by 11 Π C (q A, q B γ) = Pr j {P,E} λ ( j 1 µ j + φ ( u j + ξ t ) ) > 1. 2 Using the distribution of ξ this probability simplifies to 1 if µ + φ λ j {P,E} j u j 0 Π C (q A, q B γ) = 1 µ + φ j {P,E} λ j u j 0 if µ + φ λ j {P,E} j u j 1 where µ = λ P µ P + λ E µ E., (5) As we see, the probability of victory for A in a competitive authoritarian regime depends not only on the exogenous characteristics of the electorate (e.g., the size and the responsiveness of each voter group), but on the endogenous choice of γ and q. The same is true for the competitiveness of elections in this regime. High levels of µ, which could be related to high levels of (endogenous) electoral distortions or to low levels of bias in the population, are associated with high levels of political competition. Similarly, cases in 10 A more flexible specification would be to allow for a continuum of γ and a cost for each level of manipulation. In this setting, given that there are only two groups of voters, the restriction on γ is without loss of generality. 11 The assumption of majoritarian rule is not crucial. If the objective of the parties is instead the maximization of the vote share (an objective that would correspond to a system of proportional representation) the results would be exactly the same. 13

which voters are less sensitive to ideological concerns (i.e., high φ) are also associated with a stiffer electoral arena. In a democratic state, the incumbent cannot manipulate the electoral process, so we have the standard model of political competition. Given that the same two parties that participate in a competitive authoritarian regime compete also in a democracy, the fraction of elite and poor supporters for each party in a democratic state is also given by (4). Hence, the corresponding probability for A in a democracy is Π D (q A, q B ) = Pr λ j (1 µ j + φ ( u j + ξ t )) > 1. 2 j {P,E} By the same arguments, this probability is given by 1 if µ + φ Π D j {P,E} λj u j 0 (q A, q B ) = 1 µ + φ j {P,E} λj u j 1 0 if µ + φ j {P,E} λj u j 1, (6) where µ = j {P,E} λj µ j. This last term is the aggregate negative ideological bias towards A in the population. 2.3 Timing of Events and Equilibrium Concept The political competition in this economy is represented as a dynamic game between A, B and all voters in E P. In any period of non-democracy, A decides whether to repress or to democratize. In any period of low threat, the incumbent of an authoritarian regime can decide whether to liberalize or to maintain the status quo. In any period in which elections are introduced, A and B can decide whether to invest in party organizations and compete. Explicitly, the order of moves in any stage game is as follows: 1. Nature decides whether or not the opposition organizes against the regime. 2. If the opposition organizes and the regime is authoritarian (i.e., s t = M h ), A decides whether to repress or to democratize. If repression is used, the opposition cannot threaten the regime, and the stage game ends. 3. If the opposition organizes and the regime is competitive authoritarian (i.e., s t = C h ), A decides whether to repress or to democratize. If there is repression, the opposition cannot threaten the regime and competitive authoritarian elections take place. 14

4. In an authoritarian regime of low-threat (i.e., s t = M l ), the incumbent decides whether or not to liberalize. If a liberalization take place, the incumbent chooses γ { γ, γ}. 5. If a liberalization or a democratization takes place, A and B decide sequentially whether to invest in party organizations. 6. In any period of elections, A and B simultaneously and noncooperatively announce their electoral promises, ξ t is realized, and elections held. The winner implements his announced policy, collects all rents from office and becomes the incumbent in the next period. The standard solution concept in this type of game is subgame perfect equilibrium. In this concept, strategies played at any date may depend on the entire history of the game until that date. To simplify the analysis I restrict this history dependence and focus on pure strategy Markov-perfect equilibria (MPE). This implies that the strategies players play are restricted to those that depend only on the current state of the game and not on the entire history of play (except for relevant within-period histories). The focus on payoff-relevant strategies is attractive not only because these capture in a natural way the commitment problems stressed by North and Weingast (1989) and Acemoglu and Robinson (2000, 2006), but also, because past actions can only affect current payoff-relevant strategies if they have directly altered the physical environment of the game. In this particular setting, past liberalizations directly affect the current and future payoffs by determining the existing party organizations and the electoral regime. Therefore, MPE isolates the effect of past investments from other possible mechanisms by which a nondemocratic regime may retain power (e.g., promises of redistribution) in a simple way. A strategy for A is a function σ A (s). This strategy determines the action profile {ω, c, γ A, ϕ A, q A }. The variable ω denotes the decision to repress, which applies only in states M h and C h. ω = 1 indicates repression and ω = 0 indicates no repression. The variable c represents the decision to introduce competitive authoritarian institutions, a decision that applies only in state M l. Following the same convention, c = 1 implies the introduction of such institutions, and c = 0 implies no introduction. γ A { γ, γ} denotes the decision of how to manipulate the electoral contest in state C l. ϕ A {0, 1} represents the decision to 15

invest in party organizations, and lastly, q A is a policy announcement during elections which happens only in states C l and D. Similarly, a strategy for B is σ B (s) and it determines the profile {γ B, ϕ B, q B }, where γ B is the decision to manipulate the election in state C l when B is in power, ϕ B is the party investment decision, and q B is a policy announcement which applies only in states C l and D. 12 Then, a strategy combination (σa (s), σ B (s)) is a pure strategy MPE if σ A and σ B 3 Analysis are best responses to each other for all s in S. Pure strategy MPE profiles are characterized by backward induction within in any stage game beginning at some arbitrary date t and taking future plays as given. I start the analysis by solving for the equilibrium policy platforms in any period of elections taking the decision of γ as given. First, let ζ s = (τ A (s), τ B (s)) denote a MPE policy combination in state s where τ g : {C l, D} Q. If, starting in state M l, A does not undertake a liberalization, the Bellman equation describing the discounted payoff for such strategy is given by V A (M l ζ D ) = R + δ [ (1 p)v A (M l ζ D ) + pv A (M h ζ D ) ], (7) where δ is the discount rate and p is the probability of mobilization defined previously. This expression has a straightforward form. Since there is no electoral competition in a closed dictatorship, A enjoys all rents from office and holds power with probability 1 in the current period. 13 In the next period, the opposition fails to mobilize with probability 1 p, so the value V A (M l ζ D ) recurs. In the last term, the opposition mobilizes with probability p, so the continuation value is V A (M h ζ D ), the value of a threatened dictatorship. V A (M h ζ D ) is not necessarily equal to V A (M l ζ D ) since the value in this case is a function of the decision to repress or to democratize. 12 The action ω is not specified for B since A is the only group capable of employing repression. Yet, it is possible that B acting as the incumbent of an closed regime could decide whether or not to liberalize. This action is not specified for B since I assume that the society starts as a closed regime ruled by A. This implies that if B ever acts as incumbent, a liberalization has occurred in the past. Therefore, an action profile c for B is redundant. 13 Since parties have no policy preferences and there are no elections in a closed regime, the economic policy vector plays no role in this state (i.e., party A is indifferent between any q in Q). 16

Formally, V A (M h ζ D ) can be expressed as V A (M h ζ D ) = max ω {0,1} ω(v A (M l ζ D ) κ) +(1 ω) max ϕ A {0,1} ϕ A(V A (D ζ D ) θ h ), (8) where ω = 1 implies repression and ω = 0 implies democratization. For the case in which there is no repression, A decides whether or not to participate in the new democratic regime where ϕ A = 1 denotes participation. Is it clear to see how the decision to democratize depends crucially on the costs of repression κ, on the continuation value under democracy V A (D ζ D ), and on the level of party investments in a high-threat period. The value functions for A and B in any state s t {C l, C h ) are somewhat some complex but they follow the same logic as equation (7). In these states, we have to take into account that the incumbent may lose power if he is defeated in elections (even though the electoral institutions in a competitive authoritarian regime give a substantial incumbency advantage). With a slight abuse of notation, let τ g (C l ) = τ g (C) denote a MPE platform for party g in a competitive authoritarian election. Also, to keep the analysis tractable, suppose that once the competitive authoritarian institutions are installed for the first time they remained fixed for the rest of the game. Therefore, the relevant value functions in a low-threat competitive authoritarian state (ignoring party formation costs and taking the decision of γ as given) are and V A (C l ζ C, ζ D ) = max q Q {ΠC (q, τ B (C))R +δ[(1 p)v A (C l ζ C, ζ D ) + pv A (C h ζ C, ζ D )]}, (9) V B (C l ζ C, ζ D ) = max q Q {[1 ΠC (τ A (C), q)]r where the probability Π C (.) is given by (5). +δ[(1 p)v B (C l ζ C, ζ D ) + pv B (C h ζ C, ζ D )]}, (10) The main difference between V A (C l ) and V A (M l ) is that in a competitive authoritarian regime, A holds (or captures) power in the current period with probability Π C. The values V A (C l ζ C, ζ D ) and V B (C l ζ C, ζ D ) use the one-stage-deviation principle for infinite-horizon games (Fundenberg and Tirole 1991, Theorem 4.2), since it is assumed that both A and B will follow the same optimal strategies tomorrow if they face the same state as today. 17

By the same argument as before, the values in a high-threat competitive authoritarian state are expressed as and V A (C h ζ C, ζ D ) = max ω(v A (C l ζ C, ζ D ) κ) ω {0,1} +(1 ω) max ϕ [ A V A (D ζ D ) (1 η)θ h], ϕ A {0,1} V B (C h ζ C, ζ D ) = ωv B (C l ζ C, ζ D ) + (1 ω) max ϕ [ B V B (D ζ D ) (1 η)θ h], ϕ B {0,1} where the parameter η (0, 1] measures the ratio of sunk investments to total party investments (in a high-threat period). A useful interpretation of this parameter is that it measures the effectiveness of party investments incurred in the past in a new democracy. For the extreme case in which η = 1, A and B compete for free in a new democracy since the party organizations in place, developed in a previous period, are sufficient to compete in a new democratic regime. It is clear to see how the value function for A for the case in which they face an organized opposition depends on their own choice about whether or not to repress. If repression takes place, A pay the cost κ and elections takes place. On the other hand, since A enjoys the monopoly of repression, the value for B in this state depends on A s decision to repress. If A represses, so ω = 1, they get the payoff V B (C l ζ C, ζ D ) and participate in competitive authoritarian elections. For the case in which there is no repression, both A and B decide whether or not to invest in a party organization to compete in fully democratic elections. I now characterize explicitly the equilibrium policies. Throughout this analysis, I assume that the ideological distribution and the electoral manipulation is such that no party will win with probability one. The following lemma characterizes the MPE policies in a competitive authoritarian state (all proofs are contained in the Appendix). Lemma 1 In any MPE, the unique electoral equilibrium involves τ A (C) = τ B (C) q C. For the case in which the equilibrium is interior, q C is characterized by (λ E + γ) u E (q C) + (λ P γ) u P (q C) = 0, (11) where u j (q) denotes the gradient vector ( u j q 1,.., u j q K ) T. 18

This characterization result highlights two points. The first one is that in a competitive authoritarian regime, the unique electoral equilibrium in any MPE involves both parties converging to the same policy platform. The intuition for this is that since parties do not have policy preferences, the objective function of B is symmetric to that of A. Thus, parties face exactly the same maximization problem. As it is well known, this result is not general and is a consequence of the fact that parties do not have policy preferences. 14 In the next section I relax this assumption and explore how the introduction of partisan politics influences the analysis. Second, and more important, the equilibrium policy caters to the preferences of the elite and the citizens. This is easy to see from (11), but it is also true for the case in which the solution is at the boundary of the feasible set. 15 Given that both λ P and λ E are strictly positive, both elite and poor voters have electoral power in a competitive authoritarian state. Yet, this power depends not only on the primitives of the population (i.e., their numbers and their economic preferences) but on the electoral design of the regime. In this state, the ruling group can decrease (or increase) the power of the citizens depending on how this affects their electoral prospects. For example, if A introduces such system we will have γ A = γ. This implies that a competitive authoritarian regime introduced by A will give disproportionate importance to elite voters. Lastly, the value functions for A and B in a democracy are given by { } Π V A (D ζ D D (q, τ B (D))R ) = max (12) q Q 1 δ {[ 1 Π D V B (D ζ D (τ A (D), q) ] } R ) = max. (13) q Q 1 δ These values take into account the fact that once democracy is created it becomes an absorbing state. An electoral equilibrium in this state is a policy combination {τ A (D), τ B (D)} such that τ A (D) solves (12) and τ B (D) solves (13). The following result is analogous to Lemma 1, and it characterizes these equilibrium policies. Lemma 2 In any MPE, the unique electoral equilibrium in a democracy involves τ A (D) = τ B (D) qd. For the case in which the solution is interior, this policy vector is characterized 14 Policy convergence between parties is not important for the main result, but it simplifies the analysis. 15 The interior solution characterization is useful however since it corresponds to the solution of a weighted utilitarian social welfare function with citizens receiving a weight λ E, and elite voters having a weight λ P. 19

by λ E u E (q D) + λ P u P (q D) = 0. (14) The characterization shows how in a democracy, both parties will converge to the same vector of policies. The intuition for this result is similar to the one explained previously. Also, the interior solution characterization shows clearly how the economic policy in a democracy is catered according to the size and responsiveness of each group, with bigger and more responsive groups having more importance in the policy making process. Moreover, the economic equilibrium in such state is optimal in the sense that it maximizes the average utility in the voter population. I now analyze the optimal choice of ω. This indicator variable will determine whether A decides to repress or democratize in a high-threat period. This decision depends crucially on the relationship between the repression cost κ and the party investment θ h. For cases in which κ and θ h are too high, A will never repress, nor will they want to invest in a party to compete in a democratic election. Certainly more interesting are cases in which, given that democratic elections are introduced, A and B will find it optimal to compete and the investments entailed in such competition are considerable high. A sufficient condition for this to be true is given by the following restrictions. Assumption 2: θ h < µr 1 δ and θh < (1 µ)r 1 δ. This assumption is simply a participation constraint for A and B in a democratic state. 16 Under these restrictions, starting from a closed authoritarian regime, both A and B will find optimal to participate in the new democratic system. Moreover, given that the costs of setting up a party organization are lower if a democratization occurs in a competitive authoritarian state (relative to a closed authoritarian state), these restrictions imply also a participation constraint for A and B in the case in which a democratization takes place in a competitive authoritarian regime. Thus, under Assumption 2, any MPE profile will involve ϕ A (D) = ϕ B (D) = 1. 16 An alternative formulation of Assumption 2 is { θ h < µr/(1 δ) if µ < 1/2 (1 µ)r/(1 δ) otherwise. 20

I now define some values necessary to describe the complete equilibrium profiles. In particular, I focus on some critical values for κ that will determine the decision to repress or democratize in different states of the game. In order to derive these repression thresholds, suppose that A, acting as the incumbent in a closed dictatorship, is playing a strategy of repressing every period in which they face an organized opposition. 17 Let V A (O, M l ) the payoff of such strategy, where O denotes oppression (the letter R is already used to denoted the rents from holding power). Using (1.8) we get that V A (O, M h ζ D ) = V A (O, M l ζ D ) κ. Substituting in (7) and solving for V A (O, M l ζ D ), we get that The interpretation of (15) is straightforward. V A (O, M l ζ D ) = R δpκ 1 δ. (15) This value is the per-period return from having an infinitely-lived dictatorship discounted to the present. It takes into account the fact that the regime will pay the cost of repression κ a proportion p of future periods. Let κ be a repression cost such that in any high-threat state, a closed dictatorship ruled by A is indifferent between repression and democratization. Under Assumption 2, this critical value is implicitly given by V A (M l κ ) κ = V A (D ζ D ) θ h. Using (12), (15), and the result of Lemma 2, this critical value of the repression cost is given by κ = 1 [ (1 Π D (ζ D ))R + (1 δ)θ h], (16) 1 δ(1 p) where ζ D = (q D, q D ) and Π D (ζ D ) = 1 µ. Therefore, starting in a closed regime, if κ κ we will have ω = 1, so repression will occur in any period in which the opposition mobilizes. Similarly for cases in which κ > κ democracy is preferred so A chooses ω = 0. Comparative statics on κ reveal that if p is high, which means that the opposition is well organized across time so they pose a constant threat, repression is less attractive and democracy more likely (i.e., κ / p < 0). Also, for cases in which the expected rents for office are high and the party formation costs in a high-threat period are low, an increase in 17 This derivation is without loss of generality since any conjectures we make about future actions are not important. If instead we use a one-shot deviation principle approach, assuming that repression would take place today and never in the future, we would get exactly the same repression threshold κ. This is a property of this type of games and of dynamic programing more generally (see Fudenberg and Tirole 1991). 21