The Good, the Bad, and the Civil Society

The Good, the Bad, and the Civil Society Jiahua Che Kim-Sau Chung Xue Qiao April 5, 2012 Department of Economics, Chinese University of Hong Kong, Hong Kong; jiahua@cuhk.edu.hk Department of Economics, University of Minnesota, U.S.A.; sau@umn.edu Department of Economics, Tsinghua University, China; qiaoxue@sem.tsinghua.edu.cn

The Good, the Bad, and the Civil Society Abstract There are three signature features of autocracies. First, there is wide variety across autocracies in terms of economic performance: some do much better and some much worse than democracies. Second, economic performance within a given autocracy is more sensitive to leader quality, resulting in higher volatility. Third, all autocracies, good or bad, tend to have weaker civil societies than democracies do. We attribute these features to the core of the autocratic political institution: incumbent leader selects future leader as opposed to citizens at large selecting future leader under democracy. We deliver our analysis in an overlapping-generation model where two kinds of dynamic free-riding problem arise. The first arises among different generations of citizens in implementing far-sighted policies. Political leaders come in two types, good ones aim to correct this first kind of dynamic free-riding problem, while bad ones do not care and aim to steal public assets. Both types need a weak civil society to achieve their goals, but a second dynamic free-riding problem arise among different leaders when it comes to suppressing the civil society. The autocratic leaderselection mechanism helps resolve this second dynamic free-riding problem, results in a continuously weakened civil society, which in turn generates more variation both among autocracies and within a given autocracy. A rich set of comparative statics is offered. Keywords: autocracy, democracy, political selection, civil society JEL code: D72, H11, P48

1 Introduction Almost half of the world is still governed under autocratic regimes, among which some have begun to embrace democracy, while some have been attaining remarkable economic achievement. Unfortunately, the literature studying autocracies remains disproportionally small compared with that studying democracies. Out of this emerging literature comes a number of salient empirical patterns. First, there is no systematic and robust evidence, either in cross-country comparisons or in within-country studies, that democracy brings about better economic performance than autocracy does (Przeworski and Limongi (1993), Barro (1996), Mulligan, Gil, and Sala-i-Martin (2004), Glaeser, La Porta, Lopez-d-Salines, and Shleifer (2004), Giavazzi and Tabellini (2005), Aghion, Alesina, and Trebbi (2007), Besley and Kudamatsu (2008)). Some autocracies do much better and some much worse than democracies. Second, however, economic performances of autocracies tend to be more volatile than those of democracies (Rodrik (2000), Quinn and Wolley (2001), Mobarak (2005)). Part of this extra volatility comes from the fact that political leaders exert a more powerful impact on economic performance in autocratic regimes than in democracies (Glaeser, La Porta, Lopez-d-Salines, and Shleifer (2004), Jones and Olken (2005)). Third, what could serve as restraints over political leaders, such as freedom of speech and civic activism, are more likely to be suppressed in autocracies than in democracies (Mulligan, Gil, and Sala-i-Martin (2004)). While some previous models take these features as fundamentals, and derive implications of them, 1 we seek to explain them as endogenous phenomena, and hopefully with a unified theory. We attribute these empirical patterns to the core of the autocratic political institution: incumbent leader selects future leader as opposed to citizens at large selecting future leader under democracy. We deliver our analysis in an overlapping-generation model where two kinds of dynamic free-riding problem arise. The first arises among different generations 1 For example, McGuire and Olson (1996) assume implicitly that democratic leaders are completely restrained (and hence incapable of wrong-doing in high office), and derive implications of that assumption. See Olson (1993) for a similar analysis. Similarly, Larsson and Parente (2008) assume exogenously that autocracies economic performances are more uncertain, and derives their analysis based on this assumption. 1

of citizens in implementing far-sighted policies. Political leaders come in two types, good ones aim to correct this first kind of dynamic free-riding problem, while bad ones do not care and only aim to steal public assets. Both types need a weak civil society to achieve their goals, but a second dynamic free-riding problem arise among different leaders when it comes to suppressing the civil society. The autocratic leader-selection mechanism helps resolve this second dynamic free-riding problem, results in a continuously weakened civil society. A continuously weakened civil society then leads to two kinds of variability in economic performance. First, within a given autocracy, a weakened civil society allows the leader s quality to have a larger impact. To the extent that leader s quality fluctuates, the country s economic performance fluctuates with it. Second, a weak civil society makes the choice of future leader more relevant, and hence this choice becomes more sensitive to primitive parameters of a given country. To the extent that these parameters vary across countries, the (endogenous) leader-selection outcome varies across countries as well. In our model, the cross-country variability is captured by the fact that, while there is only one kind of equilibrium in democracy, there are several possible kinds in autocracy, depending on the primitive parameters of a given country. The within-country variability is captured by the fact that most of the possible equilibria in autocracy are accompanied by a stationary distribution (of economic performance) that exhibits more volatility. Either of these two kinds of variability could have explained Besley and Kudamatsu s (2008) empirical finding that the distribution of economic performances among autocracies tends to be more diffused than that among democracies and has a longer tail on both sides. However, it is still important to conceptually distinguish the two, because the existence of within-country variability suggests that Besley and Kudamatsu s (2008) result will not completely go away even if we control for all country-specific characteristics. Although we build our theory on autocracy s leader-selection mechanism, and argue that this mechanism alone can already account for many important differences between autocracies and democracies, this does not imply that we dismiss the importance of other 2

mechanisms. One particular mechanism that has been emphasized by previous studies is the reelection mechanism: autocratic and democratic leaders face different incentives to implement good policies because they face different reelection mechanisms. Implications of this difference have been studied by, for example, Robinson (2001), Bueno de Mesquita et. al. (2003), Maskin and Tirole (2004), Shen (2007), Miquel (2007), and Besley and Kudamatsu (2008). 2 According to Besely (2005), this mechanism has been studied at length, while the leaderselection mechanism has received far less attention despite the fact that the past 200 years of political history justify an emphasis on selection. Indeed, we made a deliberate effort to make sure that our analysis is driven purely by the leader-selection mechanism. In particular, we assume that each leader can serve at most one term (in both democracy and autocracy), and hence makes policy choices without the reelection consideration. There are also studies that, like ours, emphasize the importance of the leader-selection mechanism. Notable examples include Acemoglu, Egorov, and Sonin (2010), who study political selection of a broad range of political institutions that include democracy and autocracy as two extremes. Their model, however, does not generate enough variation either among or within autocracies to match the data. In particular, autocracies universally sink to the bottom, and are dominated by democracies. This is due to two assumptions made in their model: (1) there is no conflict of interests among voters (who in turn always elect the best possible leader under democracy), and (2) there is no term limit in political office (so that, under autocracy, eventually a bad incumbent will remain in power forever after eliminating other, better candidates). Our model eschew these two assumptions, and manage to generate more variation both across autocracies and within a given autocracy. A less related study is Caselli and Morelli (2004), who examine the determination of leadership quality. Their focus is on the decision of individuals with different qualities in joining the government, rather than on how different leader-selection mechanisms affect economic performances. 2 While these studies do not always explicitly refer to autocracies, their analyses are sufficiently general for readers to make such connection. 3

Our study also extends a number of existing studies in the political economy literature. Our first dynamic free-riding problem (the one arising among different generations of citizens) echoes the inefficiency in the representative democracy analyzed by Besley and Coate (1998). Our overlapping-generation model of leadership succession bears resemblance to Rauch (2001), although Rauch s (2001) interest is not on autocracies, but rather on the question of whether a bureaucratic structure can help the incumbent leader select a good successor. 3 The rest of this paper is organized as follows. The next section introduces the model. Section 3 analyzes first the equilibrium outcome under democracy, followed by a characterization of various equilibrium outcomes brought by autocracy. Section 4 discusses the intuition of, and draws implications from, the analytical results obtained in the previous section. Section 5 concludes. 2 The Model Consider an overlapping-generation economy. At the beginning of each period t, a unit mass of young citizens are born. Each citizen lives for at most two periods. With probability ɛ > 0, a typical citizen dies prematurely after one period. Therefore, in each period t, the economy is populated by a unit mass of young and a 1 ɛ mass of old citizens. The role of ɛ is to break the symmetry between the sizes of the young and old populations, making the young citizens the majority. Throughout this paper, we shall hence simplify algebra by treating ɛ as arbitrarily small while remaining non-zero. All payoffs will be measured in terms of a (perishable) numeraire good. At the beginning of each period, each old citizen is endowed with e units of this numeraire good, which can be thought of as income from his inelastic labor supplied when he was young. There is a government, run by a politician. is selected into this office under different political regimes. We shall explain later how a politician The government does two things, which we shall refer to as its welfare and investment policies. First, it implements 3 Rauch (2001) assumes exogenously that an incumbent wants to select a good successor. In our model, an incumbent s preference over successors is endogenous. 4

transfer payments among citizens. We assume that different young citizens have to be treated equally, and similarly for different old citizens. Therefore transfer payments can take only the form of taxing old citizens (uniformly) and transferring to young citizens (also uniformly). Transfer payments involve deadweight loss. The exact form of such loss matters little. Here, we assume that if τ units of numeraire good are taxed from the old citizens, a fraction δ (0, 1) of τ will dissipate in the process of transfer. Let λ = δe denote the total deadweight loss if each old citizen s full endowment e is taxed away. The government is also endowed with a certain amount of perishable goods at the beginning of each period, which can be thought of as income from natural resources. The second thing the government does is to invest this endowment. There are three mutually exclusive ways to invest it. First, the government can invest it in some long-term project, which is so long-term that none of the citizens, young or old, will live long enough to enjoy the benefit. More concretely, we assume that a long-term investment made in period t yields a public good in period t + 2, which in turn generates for each period-(t + 2) citizen a benefit equivalent to R units of the numeraire good. Alternatively, the government can invest this endowment in some short-term project, which yield benefits immediately. Specifically, a short-term investment made in period t yields a public good in the same period, which in turn generates for each period-t citizen a benefit equivalent to G units of the numeraire good. We assume G < β 2 R, where β is the discount factor. The third way to spend this endowment is on perks given to the politician in office. The perks generate for that politician a private benefit equivalent to Z units of the numeraire good, with Z > G. We refer to the politician in office in period t as the period-t leader. However he was selected, the period-t leader s choice of the government s period-t welfare and investment policies is subject to the constraint of public opinion, which we refer to as civil society. We use ω t to denote the strength of the civil society in period t, with ω t = 1 meaning strong and ω t = 0 weak. The period-t leader gets to choose the welfare and investment policies at his will only if ω t = 0; if ω t = 1, the government must give in to popular demand and as 5

a result, period-t policies will in effect be dictated by the majority of the society (i.e., the period-t young citizens). The civil society is fragile: the strength of civil society is itself manipulatable by the government. However, to ensure that the constraint of public opinion does have a bite on the policy choice, we also treat civil society, like other institutions, as resilient. Thus, the period-t leader can determine only the strength of the civil society in the next period, ω t+1. For simplicity, we assume that the government s ability to manipulate civil society is not constrained by the strength of civil society; hence the period-t leader can set ω t+1 costlessly. In modeling this fragile and yet resilient civil society, we have in mind that, in an initially open and free society, tactics a government deploys to discourage civil activism will not be able to suppress populist demand immediately, but over time, the accumulated memory of government oppression will suffice to silence people, and that the frustrated civil society will rebound only long after the government has stopped applying these tactics. Note also that we speak of a strong or weak civil society only in relative terms. In particular, a strong civil society is not strong enough to change the political regime from autocracy to democracy, or vice versa; nor a weak civil society weak enough to allow the leader to make such changes. Similarly, a strong civil society is not strong enough to protect itself from being weakened by the leader, and a weak civil society is not too weak to rebound. We consider two political regimes: democracy and autocracy, which we regard as differing only in their constitutional rules of how a politician is selected into office every other difference comes from there. In autocracy, the leader is selected by the period t 1 leader; in democracy, the leader is elected by the citizens by a majority rule. We assume that, in democracy, the majority group (i.e., the period-t young citizens) act as a single selector. As such, we ignore any collective-decision problem among them. We also assume the constitutional rules to be exogenously imposed from the onset. As such, we ignore more general regimes such as those that allow switching back and forth between these two by way of, say, referendums. The selection of the period-t leader takes place at the beginning of period t. Regardless of the political regime, we assume that only young citizens are eligible to be selected into 6

office, perhaps because old citizens are not physically fit for the job. This assumption, together with the assumption of young citizens being the majority group, ensures that the selector in either political regime, i.e the young under democracy and the incumbent under autocracy, will calculate how their leadership choice will affect policies, and hence their very own payoffs, in the future. There are two types of citizens in each generation: the set of benevolent ones is nonempty but has measure 0, while the rest are selfish. A selfish young citizen s utility is U t = u y t + βu o t+1, with u y t = (1 δ)τ t + g t + r t and u o t+1 = e τ t+1 + g t+1 + r t+1, where τ t is the tax imposed on the period-t old citizens, g t = G if the government makes short-term investment in period t (g t = 0 otherwise), r t = R if the government makes long-term investment in period t 2 (r t = 0 otherwise), and β is the discount factor. Total welfare in period t is (recall that almost all citizens are selfish, and that the premature death rate ɛ is arbitrarily small) w t = u y t + u o t. A period-t benevolent citizen (young or old) maximizes W t = β s w t+s. s=0 Types are private information. At the beginning of each period t, before the selection process takes place, a heroic event may happen, in which a benevolent young citizen will rise to the occasion and reveal credibly his type to the society. Such a heroic event happens with probability q > 0. If and only if it happens, the selector has the option of selecting this identified benevolent candidate. Given the fact that the set of benevolent citizens has a measure of zero, a random candidate is almost surely selfish. It is not an invention of this model to recognize the possible presence of benevolent po- 7

litical leaders. When studying how re-election concern offers accountability for government behaviors, Tirole and Maskin (2004) as well as Besley and Kudamatsu (2007) for example, rely crucially on the possibility of leaders being benevolent. Different from these studies, however, we do not assume an exogenous probability of a leader being benevolent, rather in this model the probability is made endogenous as a result of political selection. Note that having benevolent leaders does not immediately imply supremacy of autocracy. It remains to be shown what types of leaders will be placed into power under either political regime. Nevertheless, to make our analysis more intriguing, we stack our cards against autocracy by assuming that, when the selector is a single person instead of the majority group, the selector can appoint a successor in exchange for a bribe. In other words, bribe-taking can take place under autocracy only. While this appears to be an additional difference between autocracy and democracy, such a difference can in fact be derived endogenously from the fact that the right to elect a leader rests with one single person in autocracy but is shared by all in democracy. We assume benevolent selectors/candidates do not take/pay bribe. Bribe-taking hence can take place only when the incumbent leader is selfish, and when he intends to select one out of infinitely many selfish young citizens as his successor. We do not try to explicitly model how any bribery contract may be enforced, or how a young citizen may borrow resources to pay bribe upfront. Instead, we collapse all these and other possible obstacles into a single parameter ˆb (0, 1), which we interpret as the maximum portion of perks a candidate can credibly pledge to share with the selector. Two implications follow immediately from this specification: the maximum bribe a period-t leader can collect from selling the office is b := ˆbZ if he has weakened the period-t + 1 civil society, which in turn allows his successor to collect perks in the following period, and is 0 if he has not. Finally, we assume that the strength of the period-t civil society by itself does not constrain the period-t leader s ability to sell office. We summarize the time line (within period t) as below: 1. a heroic event may or may not happens; 2. the selector (the majority group if we are in a democracy; or the period-(t 1) leader 8

if we are in an autocracy) selects the period-t leader; bribery may or may not take place; 3. if ω t = 1, the majority group choose the government s welfare and investment policies; 4. if ω t = 0, the period-t leader chooses the government s welfare and investment policies; 5. the period-t leader chooses ω t+1 ; 6. period-t payoffs realized. Before proceeding to the analysis, we observe for later reference that w t e λ =: w, and W t w/(1 β) =: W. The lower bound is achieved if the government always spends the public funds on the leader s perks and implements the maximum welfare transfer. Similarly, w t e + R =: w, and W t w/(1 β) =: W. The upper bound is achieved if the government always spends the public funds on long-term investment and implements no welfare transfer. 3 Equilibria Our solution concept is pure-strategy Markov-perfect equilibrium. The only payoffrelevant state variable is the strength of the civil society ω t. Before we study the equilibria in democracy and in autocracy, respectively, let s observe some common features across these equilibria. First, when ω t = 1, the majority group (i.e., the period-t young citizens) will dictate the government s welfare and investment policies. Regardless of the political system, in any Markov-perfect equilibrium, they must choose to tax the old citizens maximally (i.e., τ t = e, resulting in the maximum deadweight loss of λ), and to make short-term investment (generating payoff G immediately for every citizen). We shall refer to these as the populist policies. Second, when ω t = 0, and when the period-t leader is a benevolent one (hereafter a B-leader), he must choose not to tax the old citizens (avoiding deadweight loss), and to make long-term investment (generating payoff R for every period-(t + 2) citizens). 9

Third, when ω t = 0, and when the period-t leader is a selfish one (hereafter an S-leader), he must choose to tax the old citizens maximally (because he is himself a young citizen), and to invest in his own perks (generating immediate payoff of Z for himself). Therefore, equilibria in democracy and in autocracy differ in only two aspects: (i) how a period-t leader chooses ω t+1, and (ii) the kind of leader selected into office. We now turn to these questions. 3.1 Democracy Given the above observations, in democracy, an equilibrium boils down to a vector ( ΩB ( ), Ω S ( ), L Y ( ) ), where Ω B (ω t ) (resp. Ω S (ω t )) is a (period-t) B-leader s (resp. S- leader s) equilibrium choice of ω t+1 when the state is ω t, and L Y (ω t ) {S, B} is the (period-t) young citizens equilibrium choice of (period-t) leader when the state is ω t. Note that, if L Y (ω t ) = S, the young citizens intend to select a selfish candidate into office, and they will succeed with probability one (recall that they can always randomly pick one of the young citizens); however, if L Y (ω t ) = B, the young citizens intend to select a benevolent candidate into office, but they will succeed only with probability q. With probability 1 q, a heroic event does not happen, and they will have no means to identify a benevolent candidate. Markov-perfection requires that these choices depend only on the state but not on the history. Indeed, that the young citizens leader choice does not depend the history implies that Ω B and Ω S should be independent of the state; i.e., Ω B (0) = Ω B (1) = Ω B and Ω S (0) = Ω S (1) = Ω S. One immediately observes that, in any equilibrium, L Y (0) = S. Indeed, when it comes to choosing the next period s state ω t+1, an S-leader, being one of the young citizens, has the same preference as other young citizens have. And when it comes to choosing the current period s welfare and investment policies, an S-leader s choices (maximum welfare transfer, no short-term investment) are closer to the young citizens ideals than a B-leader s (no welfare transfer, no short-term investment) are. With L Y (0) = S, we must have Ω B = Ω S = 1, as shown by the following two lemmas. Lemma 1 L Y (0) = S implies Ω S = 1. 10

Proof. An S-leader s choice of ω t+1 affects only u o t+1. If he chooses ω t+1 = 0, then u o t+1 = 0 because the period-(t+1) leader will be an S-leader, who will implements maximum welfare transfer and spend public funds on perks. If he chooses ω t+1 = 1, then u o t+1 = G because populist welfare and investment policies will then prevail. Lemma 2 L Y (0) = S implies Ω B = 1. Proof. Suppose L Y (0) = S but Ω B = 0. Let W (ω t ) be the equilibrium discounted sum of future welfare starting from the initial state ω t. A B-leader s choice of ω t+1 affects only W t+1. If he chooses ω t+1 = 0, then W t+1 = W (0) = e λ + βw (Ω S ) = w + βw (Ω S ) because the period-(t+1) leader will be an S-leader, who will implements maximum welfare transfer and spend public funds on perks. By Lemma 1, W (Ω S ) = W (1). Therefore, W (0) = w + βw (1). If he chooses ω t+1 = 1, then W t+1 = W (1). Since Ω B = 0, we have W (0) = w + βw (1) W (1), (1) which implies W (1) W. But this is impossible, because W (1) = e λ+g+β ( qw (Ω LY (1))+ (1 q)w (Ω S ) ) > w + βw = W, a contradiction. But if both kinds of leaders have the same choice of next period s state ω t+1, then young citizens will be indifferent between them when ω t = 1, because populist policies will prevail anyway. This completes the characterization of all pure-strategy Markov-perfect equilibrium under democracy, and we have the following proposition. Proposition 1 In democracy, the unique 4 pure-strategy Markov-perfect equilibrium is that neither kind of leader weakens the civil society when it is strong, and both allow it to rebound when it is weak; young citizens are indifferent between the two kinds of leader when the civil society is strong, and will select an S-leader when it is weak. Regardless of 4 There are actually two equilibria, differing from each other in L Y (1), but they are observationally indistinguishable. 11

the initial state ω 1, the country enters the absorbing state of ω t = 1 starting from period 2. And, subsequently, populist welfare and investment policies prevail in every period. 3.2 Autocracy Given the observations made at the beginning of this section, in autocracy, an equilibrium boils down to a vector ( Σ B ( ), Σ S ( ) ), where Σ B (ω t ) = ( Ω B (ω t ), L B (ω t ) ) {0, 1} {B, S} is the (period-t) B-leader s equilibrium choice of ω t+1 and (period-(t + 1)) successor when the state is ω t ; and similarly for Σ S (ω t ). As in the case of democracy, Markovperfection implies that Σ B and Σ S should be independent of the state; i.e., Σ B (0) = Σ B (1) = Σ B = (Ω B, L B ) and Σ S (0) = Σ S (1) = Σ S = (Ω S, L S ). Notice that, independent of B-leaders strategies, an S-leader is always indifferent between Σ S = (1, B) and Σ S = (1, S). His payoff when he becomes old will be G in both cases. How S-leaders break ties, however, will have implications on the long-run performance of autocracy. In this paper, we assume that they always break ties by favoring (1, B) over (1, S). This can be justified by a weak-dominance argument: if there is any risk that the period-(t + 1) civil society is weaker than expected, then a B-successor will deliver a higher u o t+1 than an S-successor will. We also observe for later reference that, in any equilibrium, L B = B. This is because the preferences of a B-leader and a B-successor are congruent. The dynamics in autocracy is much richer than that in democracy. Depending on parameters, different equilibria may arise. However, as we shall see, pure-strategy Markovperfect equilibrium always exists, and for generic parameter values it is also unique. We find it helpful to visualize the absorbing dynamics of any given equilibrium with a picture like Figure 1. In such a picture, the two columns represent the state of civil society, while the two rows the type of incumbent leadership, thus generating four cells, each corresponding to a generalized state that the economy potentially visit in a given period in an absorbing dynamics. For example, the cell in row B and column ω = 1 corresponds to the generalized state where, in a given period t, the civil society is strong, and a B-leader is in office. Entries in these cells, on the other hand, represent the incumbent leader s choice 12

of ω t+1 and (period-(t + 1)) successor, should the economy actually visit these generalized states in the absorbing dynamics of a particular equilibrium. Whether a particular generalized state will be reached in an absorbing dynamics, as we will show later, depends on what kind of an equilibrium prevails under certain parameter range. For example, in Figure 1, entries in row B indicate Σ B = (1, B) and entries in row S indicate Σ S = (0, B); these together describe an equilibrium, in which the economy reaches all the fours cells in its corresponding absorbing dynamics. In addition, we use arrows in a picture like Figure 1 to illustrate how the economy commutes from one cell to another. An arrow coming out of a cell tells us the generalized states the economy begins in period t and reaches in period t + 1. The dotted arrow corresponds to the evolution conditional on a heroic event, which happens with probability q; the solid arrow the evolution conditional on the complementary event. ω =1 ω = 0 B (1,B) (1,B) S (0,B) (0,B) Figure 1: the mostly-bad dynamics It is easy to show that there can only be five different absorbing dynamics (see the Appendix). Besides the one shown in Figure 1, the other four are shown in Figure 2. Of particular interest are Figures 2a, 2b, and 2c, where the economy reaches only some of the four generalized cells in the described absorbing dynamics. 3.2.1 The Democratic Dynamics We call the absorbing dynamics in Figure 2a the democratic dynamics. Its corresponding welfare performance is the same as that under democracy. In every period, regardless of the type of the leader in office, he refrains from weakening the civil society, 13

ω =1 ω = 0 ω =1 ω = 0 B (1,B) B (0,B) S (1,B) S (0,B) (a) the democratic dynamics (b) the good dynamics ω =1 ω = 0 ω =1 ω = 0 B B (0,B) (0,B) S (0,S) S (1,B) (1,B) (c) the bad dynamics (d) the mostly-democratic dynamics Figure 2: the other four possible absorbing dynamics and tries to select a B-successor whenever possible. Since civil society is always strong, populist welfare and investment policies always prevail. This economy, albeit being an autocratic one, is observationally indistinguishable from a democratic one. When will Σ B = Σ S = (1, B) arise as an equilibrium? For a B leader not to deviate to Σ B = (0, B), anticipating that any future leader will continue to play the equilibrium strategy, it must be that weakening the civil society for one period generates no greater expected welfare than the populist policies do: e λ + G q(e + β 2 R) + (1 q)(e λ) G q(λ + β 2 R) =: Q. For an S-leader not to deviate to Σ S = (0, S), it must be that the bribe he can collect from selling the office is no greater than the loss of short-term public good: G b. For an S-leader not to deviate to Σ S = (0, B), it must be that the benefit of having a B- 14

successor implementing no welfare transfer does not compensate for the loss of short-term public good: G qe + (1 q)b. Proposition 2 The democratic dynamics is generated by the equilibrium Σ B = Σ S = (1, B), which arises when G Q and G max{e, b}. The economy is observationally indistinguishable from a democratic one, with welfare w dem = e λ + G in each period. 3.2.2 The Bad Dynamics We call the absorbing dynamics in Figure 2c the bad dynamics. In every period, the leader selected into office is an S-leader, who implements maximum welfare transfer, spends public funds on perks, weakens next-period s civil society, and sells the office to an S-successor. In every period, welfare is the lowest possible in this model: w bad = w = e λ. When will Σ S = (0, S) arise in equilibrium? For an S-Leader not to deviate to Σ S = (0, B), it must be that, by allowing the civil society to rebound and populist policies to prevail in the next period, his gain in short-term public good cannot compensate for his loss in bribery income: b G. Similarly, for him not to deviate to Σ S = (0, B), it must be that the benefit of having a B-successor implementing no welfare transfer cannot compensate for the loss in bribery income: b qe + (1 q)b b e. As long as b max{g, e}, and hence Σ S = (0, S), whether Σ B = (1, B) or Σ B = (0, B) does not affect the absorbing dynamics. In the appendix we show that either one will arise in equilibrium, depending on whether G Q or Q G, respectively. Proposition 3 When b max{g, e}, either ( Σ B = (1, B), Σ S = (0, S) ) or ( Σ B = (0, B), Σ S = (0, S) ) will arise as an equilibrium, depending on whether G Q or Q G, respectively. Both equilibria generate the bad dynamics, with welfare w bad < w dem generated in each period. 15

3.2.3 The Good Dynamics When most people think about autocracy, the bad dynamics is likely what appears in their mind. But it is not the only possible absorbing dynamics. Under certain parameter ranges, not only that autocracy can be as good as democracy (see the democratic dynamics above), it can even outperform democracy. The possibility of the latter case is demonstrated in Figure 2b, which we call the good dynamics. In every period, regardless of the type of the leader in office, he weakens the civil society and tries to select a B-successor whenever possible. Civil society is persistently weak, and the type of leader in office is an i.i.d. draw every period. With probability q he will be a B-leader, implementing no welfare transfer and making long-term investment. With probability 1 q he will be an S-leader, implementing maximum welfare transfer and spending public funds on perks. Single-period welfare, w t, swings wildly between the highest possible, w, and the lowest possible, w, levels. Occasionally, w t takes the intermediate levels of e λ + R (if the period-t leader is an S-type, but the economy enjoys the benefit of a long-term investment made by a B-leader two periods ago) and e (if the period-t leader is a B-type, but the period-t 2 leader is an S-type). In the limiting distribution, 5 the expected single-period welfare is w good = e + qr (1 q)λ. At first glance it is not obvious that w good > w dem. But we shall see that Σ B = Σ S = (0, B) arises as an equilibrium only if w good > w dem. For a B-leader not to deviate to Σ B = (1, B), anticipating that future leaders will follow the equilibrium strategies, it must be that allowing the civil society to rebound for one period generates no greater expected welfare: q(e + β 2 R) + (1 q)(e λ) e λ + G q(λ + β 2 R) = Q G. 5 In the expression of expected single-period welfare, where the expectation is taken with respect to the limiting distribution, the benefit of a long-term investment, R, is not discounted by β 2. The intuition is that a long-term public good appears q of the time on average. 16

The last inequality implies q(λ + R) > G, which in turn guarantees that w good = e + qr (1 q)λ = e λ + q(λ + R) > e λ + G = w dem. Similarly, for an S-leader not to deviate to Σ S = (1, B) or Σ S = (0, S), it must be that neither the benefit of short-term public good nor the bribery income from selling the office suffices to compensate for his loss from maximum welfare transfer: e max{g, b}. Proposition 4 The good dynamics is generated by the equilibrium Σ B = Σ S = (0, B), which arises when Q G and e max{g, b}. The performance of the economy fluctuates period by period, but on average is strictly better than that in democracy: w good > w dem 3.2.4 The Mostly-Democratic Dynamics We call the absorbing dynamics in Figure 2d the mostly-democratic dynamics because, under the reasonable assumption that q < 1/2, the economy spends most (more precisely, 1 q > 1/2) of the time in the state of strong civil society, where populist policies prevail and the economy resembles one in democracy. In every period, regardless of the type of the leader, he tries to select a B-successor whenever possible. The type of leader in office is hence an i.i.d. draw every period. A B-leader always weakens the civil society, while an S-leader allows it to rebound. In the limiting distribution, the expected single-period welfare is w md = (1 q)(e λ + G) + q 2 (e + R) + q(1 q)(e λ). Similar to the case of the good dynamics, although at first glance it is difficult to compare w md with w dem, one can verify that ( Σ B ) = (0, B), Σ S = (1, B) ) arises as an equilibrium only if w md > w dem (the proof is in the Appendix). Proposition 5 The mostly-democratic dynamics is generated by the equilibrium ( Σ B ) = 17

(0, B), Σ S = (1, B) ), which arises when Q G and G max{e, b}. The performance of the economy fluctuates period by period, but on average is strictly better than that in democracy (i.e., w md > w dem ), although also strictly worse than that in the good dynamics (i.e., w md < w good ). 3.2.5 The Mostly-Bad Dynamics We call the absorbing dynamics in Figure 1 the mostly-bad dynamics because, for q small enough, the economy spends most (more precisely, (1 q) 2 ) of the time in the worst situation of having an S-leader who, being unconstrained by the civil society, implements maximum welfare transfer and spends public funds on perks. In every period, regardless of the type of the leader, he tries to select a B-successor whenever possible. The type of leader in office is hence an i.i.d. draw every period. An S-leader always weakens the civil society, while a B-leader allows it to rebound. In the limiting distribution, the expected single-period welfare is w mb = q(e λ + G) + q(1 q)(e + R) + (1 q) 2 (e λ). Unlike in the cases of the good and the mostly-democratic dynamics, the comparison between w mb with w dem is ambiguous (the proof is in the Appendix). Proposition 6 The mostly-bad dynamics is generated by the equilibrium ( Σ B ) = (1, B), Σ S = (0, B) ), which arises when G Q and b max{e, G}. The performance of the economy fluctuates period by period, and on average is strictly worse than that in democracy (i.e., w mb < w dem ) if G > q(λ + R), although also strictly better than that in the bad dynamics (i.e., w mb > w bad ). The parameter ranges covered by Propositions 2 6 form a partition of the whole parameter space under study. As a result, a pure-strategy Markov-perfect equilibrium always exists for any combination of parameters, and for generic combinations it is also unique. 18

4 Analysis We can summarize what we have learned from this simple model into seven lessons. 1. Democracies are alike, but autocracies come in different colors. Different economies have different local conditions. In terms of our model, they come with different primitive parameters: q, e, Z, R, G, ˆb, δ, etc. However, these differences notwithstanding, these economies behave similarly under democracy: populist policies prevail, which favor the majority population at the expenses of the minorities, and favor the current generation at the expense of future ones. Aside from that, the government is relatively clean, and the civil society relatively strong, and economic performance relatively stable. Different democracies with different parameters will have different economic performances, but small differences in those parameters lead to small differences in those performances. Small differences in primitive parameters will, however, drive big differences in economic performances among autocracies. This is because two autocracies with small differences in those parameters can find themselves in two very different kinds of equilibrium. This often amplifies the direct effect of a parameter change. For example, an increase in R increases the return of long-term investments, but also increases the likelihood of that an autocracy gets absorbed into the good or the mostly-democratic dynamics (Propositions 4 and 5), and hence indirectly increases the likelihood that these long-term investments will be made as well. Similarly, a decrease in e and G not only decreases the average pre-tax income and return of short-term investments, but also increases the likelihood that an autocracy gets absorbed into the bad or the mostly-bad dynamics (Propositions 3 and 6), and hence further decreases the autocracy s overall economic performance. Because of these amplification effects, the distribution of economic performances among autocracies tends to be more diffused than that among democracies and has a longer tail on both sides, as documented by Besley and Kudamatsu (2008). Autocracies diverge along other dimensions as well. Many autocracies have a weak civil society (Propositions 3 and 4), but some have a strong one (Proposition 2). Some autoc- 19

racies economic performances are relatively stable (Propositions 2 and 3), while others are more volatile (Propositions 4, 5, and 6). Some autocratic governments are persistently clean (Proposition 2), some are stubbornly corrupt (Proposition 3), yet others go back and forth between these two extremes (Propositions 4, 5, and 6). Some autocracies are characteristically short-sighted when it comes to public investment (Proposition 2), some occasionally manage to make longer-term investments (Propositions 4, 5, and 6), yet others are completely dysfunctional and little if any public investments are ever made (Proposition 3). A simple model like ours provides a unified framework to understand all these variations as different equilibrium outcomes in different parameter ranges. 2. There is no room for saints under democracy. In a democracy, majority voters have no taste of electing saints into office. Saints are saints because they protect the minority citizens from the predation of the majority ones, and care about the future generations at the expense of the current one. All these stand at odds with the majority s interests. In a democracy, where majority citizens decide who will be the next leader, saints stand no chance of being a leader. Even if they are elected, it is only because voters know that their hands will be so tied (by populist pressure) as to be ineffective in causing any harm. And, indeed, their hands will be as tied as those of any other leader, making different kinds of leaders virtually indistinguishable. In terms of our model, in the unique equilibrium under democracy, we have L Y (0) = S, and hence L Y (ω t ) = B only if ω t = 1. This means that when the civil society is weak (ω t = 0), and hence the leader s hands are not tied, only a selfish candidate will be elected into office, even if a benevolent one is also running. A benevolent candidate can possibly get elected in equilibrium only when ω t = 1, which is also a time when he cannot implement his favorite policies at will. In the bulk of the political economy literature, where the predominant focus is on democracies, the possibility of benevolent candidates is oftentimes assumed away. Our theory explains why this is indeed without loss of generality. However, assuming away the possibility of benevolent candidates is with loss of generality when we also study autocra- 20

cies, as there is room for these candidates under autocracy. 3. Civil society is always strong in all democracies. In a democracy, regardless of the initial state ω 0, and regardless who initially occupy the office, civil society will stay at or rebound to a strong level in the next period, and then stays strong thereafter. This is true not only on the equilibrium path, but off the equilibrium path as well. Even when a democracy is being knocked off the equilibrium path by accident, civil society will quickly rebound to its originally strong level afterward. In this sense, a strong civil society is a very robust signature of democracies. The reason why civil society is always strong in democracy is rooted in something seemingly unrelated, namely that majority voters have no taste of electing saints into office (Lesson 2). Understanding that any future leaders must be normal selfish beings, neither a benevolent nor a selfish current leader dares to suppress the civil society. Both are afraid that a weakened civil society will make it easier for future leaders (who are definitely not saints) to engage in self-dealing (Lemmas 1 and 2). This highlights a feature of democracy that we think is under-appreciated, namely that the checks and balances afforded by a strong civil society do not come easily. They are definitely not guaranteed by fiat. They can be damaged and suppressed by those in power, just like how they are damaged and suppressed in autocracies. However, they are, instead, nurtured and protected in democracies, because everyone understands that democracies do not elect saints to office, and these checks and balances are these societies last line of defense against self-dealing leaders. 4. Civil society is often weak in autocracies, including those that out-perform democracies. If a strong civil society is a signature of democracies, the opposite is true for autocracies. This is true not only for the worst types of autocracies, but for the best types as well. Again, this is so not by assumption, but is a equilibrium phenomenon instead. Indeed, regardless of the political institution, leaders always prefer a weaker civil society. However, a dynamic free-riding problem arises among leaders when it comes to suppressing the civil society, and the autocratic leader-selection mechanism helps resolve this problem. 21

A weak civil society, in turn, explains the existence of both good and bad autocracies. It allows selfish leaders to engage in self-dealing, which squanders public funds at the expense of the society (Propositions 3 and 6), and enables benevolent leaders to defy populist demands and implement policies that are better for the society in the long run (Propositions 4 and 5). 5. Even when an autocracy outperforms a democracy, it s performance tends to be volatile. In the best type of autocracy, the economy fluctuates between good and bad times. This fluctuation is not due to business cycles, nor other aggregate economic shocks. It is due to the fluctuation in governance quality instead. Because civil society is persistently weak (Lesson 4), the governance quality is sensitive to what kind of leader is in office. As the availability of benevolent candidates fluctuate over time, the country s economic performance fluctuates with it. However, what makes a good autocracy good is that, whenever a benevolent candidate is available, he will be selected into office. Hence the economy spends a fair share of time living with a good government as well. It is worth emphasizing that, the best type of autocracy out-performs democracy not because it is lucky enough to stumble on a succession of benevolent leaders. Indeed, luck has nothing to do with that comparison. The best type of autocracy out-performs democracy even when we take the average between good and bad times. What makes a selfish incumbent leader willing to select a benevolent successor (when one is identified) instead of selling the office to a selfish one? Self protection is the main reason. Specifically, in our model, an incumbent leader would like to protect himself against populist welfare policies in the future, which a selfish successor will implement because he will be a member of the majority citizens. A more general message we bring about beyond our model is that incumbent leaders may find themselves vulnerable after their retirement as their interests then can be incongruent both with those of a selfish successor and with populist demands. They will then have the urge to seek self-protection from a benevolent successor who is ready to stand up against populist demands. If the self-protection motivation is strong enough (e big enough in our model), or the market of political offices 22

inefficient enough (ˆb and hence b small enough in our model), then an incumbent leader will willingly refrain from selling the office to a selfish successor. 6. The worst autocracies are worse than democracies. This is probably obvious by now, but it is still worth stating explicitly. Democratic institutions may not bring the best to an economy, but they protect it against the worst. The worst form of autocracy functions as follows: any incumbent leader is a normal selfish being who, thanks to a weak civil society, engages in self-dealing at the expense of the society. Moreover, he is not afraid of continuing the suppression of the civil society. Although doing so makes it easier for future selfish leaders to engage in self-dealing as well, which the current leader does not like, he is more than compensated for that by being able to sell the (now delicious) office at a better price. A reasonably efficient market of political offices definitely plays some role in facilitating this kind of autocracy, as captured by the condition b max{g, e} in Proposition 3. 7. The best and the worst autocracies may look like each other initially. This is probably the most troubling lesson we have learned. Suppose we are studying a particular autocracy and have the following findings: its civil society is deliberately suppressed, its top government officials are surprisingly uncorrupt given the lack of checks and balances, there are no widespread evidences that they are engaging in self-dealing, its public investments are farsighted, its environmental policies are the envy of many if not all, and its economic growth leaves the rest of the world in awe. What can we say about the long-term performance of this autocracy? Can we say for sure that this is an example of the best type of autocracy? All we can say is that this autocracy seems to be in the generalized state where ω t = 0 and a B-leader is in office. This is consistent with the good dynamics in Figure 2b, where the autocracy is currently in the generalized state at the top right corner. But it is also consistent with the top right corner in Figure 2c, where the autocracy is waiting for the inevitable fate of being absorbed into the self-perpetuating bad dynamics. In 23