Coup-Friendly Institutions and Apolitical Militaries: A Theory of Optimal Military in Uence
Coup-Friendly Institutions and Apolitical Militaries: A Theory of Optimal Military in Uence
Coup-Friendly Institutions and Apolitical Militaries: A Theory of Optimal Military in Uence
Abstract
Military coups often attract popular support. Certain constitutions even contain clauses that
implicitly allow the military to intervene in domestic politics. Also, the punishments for leaders of
failed coups tend to be mild. However, even in countries with coup-friendly institutions, military
o¢ cers are typically not allowed to participate in political debates. Hence, the institutions that
govern civil-military relations seem to simultaneously encourage and discourage military interven-
tions in domestic politics. This paper o¤ers a simple theoretical model that rationalises these
game with two-sided asymmetric information between two groups within the military. Both of
these groups care about the quality of government in a general interest dimension, but they may
di¤er on a second, factional dimension (e.g. ethnic or ideological). Each group receives an imper-
fect signal of the other group’s factional identity. In this set-up, optimal civil-military institutions
should allow for coups that are aligned with the general interest, while discouraging those coups
that are driven by factional interests only. The model shows that it is never optimal to allow open
political communication in the military as this will trigger coups that are mainly driven by fac-
tional interests. At the same time, the model shows that limited punishments for failed coups can
be optimal, but only if factionalisation is su¢ ciently strong. An extension of the baseline model
allows for "extremist" ideologies. The analysis highlights how a combination of factionalisation
and uncertainty about political views within the military can be an important driver of the gradual
demilitarisation of politics.
London School of Economics (STICERD and EOPP). I would like to thank Gerard Padro i Miquel for his support
and advice throughout this project. I also bene…ted from comments by seminar participants at LSE. All remaining errors
are mine.
1
1 Introduction
Surprisingly, the civil-military institutions that historically governed civil military relations simultane-
ously encouraged and discouraged coups. In 1970, fourteen Latin American countries had constitutional
clauses which allowed the military to intervene in politics (Stepan, 1971). Also, the punishments for
the leaders of failed coups were (and are) often surprisingly low: the leader of the failed 2000 coup
in Ecuador, Lucio Gutiérrez, was allowed to run in the presidential elections only three years later.1
Low punishments and institutionalised legitimacy of coups seem to encourage political interventions
of the military. These coup-friendly institutions could re‡ect the view that the armed forces have a
role to play in controlling the executive power. At the same time, even in contexts in which political
institutions are "coup friendly", o¢ cers are typically not allowed to openly engage in political debates.
The latter institution clearly constrains the political role of the military. This paper will explore the
optimal design of civil-military institutions, under the assumption that political elites or social planners
deem some control of the military over the executive power desirable.
This paper presents a game-theoretical model that rationalises observed civil-military institutions
such as the neutrality of the army and limited expected punishments for failed coup leaders. The
process of staging a coup will be modelled as a leader-follower game between two di¤erent groups within
the military. Both of these groups care about the quality of government in a general interest dimension,
but they may di¤er on a second, factional dimension (e.g. ethnic or ideological). Each group receives
an imperfect signal of the other group’s factional identity. The leader decides to stage a coup attempt,
but will only be successful if he is supported by the group of followers. Hence, the gains from staging a
coup will include the possibility to oust a poorly performing government and the chance to promote a
factional agenda. These gains will be weighed against the expected costs of staging a failed coup. By
manipulating the expected costs associated with failed coup attempts, civil-military institutions will be
able to shape the military’s coup behaviour. The welfare measure employed in this paper will assume
that coups against poorly performing governments should be encouraged, whereas coups targeting a
well performing government should be discouraged. According to this welfare criterion, my model
suggests that is never optimal to allow open political a¢ liations in the military. Open a¢ liations will
tend to promote coups driven by factional interests. As a corollary, the military is expected to perform
poorly if o¢ cers cannot hide their political a¢ liations (e.g. in case of ethnic cleavages). However, the
optimal punishments for failed coup leaders vary. If the military is severely factionalised in di¤erent
1 Shifter (2004).
2
political groups, then it may optimal to limit punishments for failed coups to encourage the military
to play its role as arbitrator in domestic politics. However, if political or ideological cleavages in the
military are limited, then high punishments for failed coup plotters are optimal. Further extensions of
the baseline model will account for endogenous communication, and the possibility of military take-
overs by an "extremist faction" (which is not preferred to a poorly performing incumbent). This last
extension highlights how a combination of factionalisation and uncertainty about political views within
This paper contributes to a growing literature that models the in‡uence of the military over policy
making. In a closely related contribution, Besley and Kudamatsu (2007) focus on institutional design
aimed at holding a government accountable for its general interest policy in a divided "society".
Whereas these authors compare stylised autocracies with democracies, I focus speci…cally on the role
focus on military interventions, Besley and Robinson (2011) model the strategic interaction between a
civilian government and the military. In their set-up, the size of the army and the decision to stage a
coup are the key decision variables. A central idea that my paper shares with these authors is the fact
that the military is politicised, i.e. that the military does not just promote its own corporate interests
but is a mapping from society at large. A di¤erent model of military interventions is developed by
Acemoglu, Ticchi and Vindigni (2009). These authors explore the moral hazard problem that arises
when the elite empower the military to repress democratisation attempts: a stronger military may
want to overthrow the elite. These authors assume that the military is always united and driven by
corporate interests (e.g. military spending), which is very di¤erent from my assumption of a politicised
and factionalised military. Leon (2009) focuses on career options for a representative o¢ cer (in the
army or in a military government) and links coup behaviour to the frequency of wars.2 These recent
contributions point at the rich interactions between coups and speci…c policy decisions, but they mostly
abstract from the mechanics of the coup process. This paper will analyse the internal coup process
within the military in more detail, by introducing factionalisation and civil-military relations that
shape the internal coup process. A contribution that shares my focus on the internal coup process is
Sutter (2003). This author develops a model of coups focusing on the collective action problem within
the military, which is di¤erent from the asymmetric information approach adopted in this paper. The
2 Severalauthors have conducted empirical research on coups, but focusing on di¤erent questions than mine. Such
econometric analyses of coups can be found in Belkin and Schofer (2003), Collier and Hoe- er (2007), and Londregan
and Poole (1990).
3
internal coup process is also the main focus of Geddes (2003), who models coups as a simultaneous
coordination problem between di¤erent factions in the military. This author shares my focus on a
factionalised military, but she does not explore the role of asymmetric information and civil-military
political transitions, although this author does not focus on military interventions in particular.3
While the existing literature has o¤ered a rich insights on the factors that could drive coup behaviour,
including both internal characteristics of the army and structural economic factors, it is striking that
few models incorporate the idea that the military is factionalised over political views. Nevertheless,
a substantial body of literature criticises the idea that the military can be considered as a unitary
actor.4 This paper will place the politicised and factionalised nature of the armed forces at the heart
of its analysis and explore the implications of this approach for the role and design of civil-military
institutions.
The paper is organised as follows. First, I will discuss certain stylized facts about coup behaviour
which the formal framework will incorporate. Second, I will informally describe the key assumptions,
the mechanism and the main results of the theoretical model. In a third section, the formal model
will be introduced and its results will be linked to relevant cases. The fourth section will present an
extension of the baseline model that o¤ers a microfoundation of earlier assumptions. The …fth section
discusses two important extensions of the theoretical framework. The sixth section o¤ers further
2 Background
Most militaries do not allow o¢ cers to discuss politics and require o¢ cers to remain "neutral" (Finer,
1962, p34).5 In several countries, this neutrality is underlined by constitutional clauses. For instance,
the 1982 Turkish constitution does not only prevent military personnel from becoming elected deputies
and being members of political parties, it also prohibits military personnel to vote at all.6 This institu-
tionalised neutrality makes it di¢ cult for coup plotters to recognise supporters. Hence, it discourages
3 This author models the decision to revolt by a domestic opposition who faces authoritarian leaders of unknown types
(hard-liner or soft-liner). Feaver (2003) models the agency problem that arises between a government and the military,
but he does not focus on military interventions.
4 See for example Stepan (1971, p7), Finer (1962, p34-42), Fitch (1998, p109-110), Geddes (2003), Loveman (1999,
army.
6 Articles 67, 68 and 76 of the 1982 Turkish constitution.
4
coups. While neutrality is the norm, there are exceptions. Sometimes, political a¢ liation is di¢ cult
to hide, for instance if the main political tensions are along ethnic lines (Luttwak, 1968, p59). There
is even one case in which political a¢ liations were deliberately open. In Brazil during the 1950ies
and 1960ies, military club elections were fought over political issues and vote results were published
Given that the neutrality of the military is common, it is surprising that a norm for the military to
abstain from intervening in politics was not wide-spread until recently. In 1970, fourteen Latin Amer-
ican countries had constitutions that implicitly allowed the military to intervene in politics (Stepan,
1971, p79). For instance, the Brazilian constitution of 1946 stipulated the military to be "obedient" to
the executive, but only "within the limits of the law". This clause was interpreted to give the military
the de jure power to intervene in politics as soon as the government was considered to transgress "the
limits of the law". In this sense, the armed forces took on the role of a "Supreme Court". Surprisingly,
the clause was deliberately included in the Brazilian constitution by democratically elected politicians
because they considered the military to be a good custodian of the state (Stepan, 1971, p79). These
ideas may also explain the surprisingly low punishments for failed coup leaders in several cases. The
leaders of the failed Brazilian coup of 1961 were the three generals who served as ministers of defence.
They were never prosecuted for their role in the coup and this mild punishment was institutionalised,
because their coup attempt could be considered a "ministerial decision".7 Even if political institutions
do not explicitly allow military control, military interventions often have a high degree of legitimacy.
The de facto role of the military as a "supreme court" could stem from a situation in which political
institutions are too weak to produce legitimate civilian arbitrators in political con‡icts. In line with
this idea, a number of very recent military interventions attracted strong popular support in Africa.8
However, the role of the military in Latin America has clearly declined over the past decades. All Latin
American countries have now adopted new constitutions that curtail the role of the military in politics
(Einaudi, 1996).9 Similarly, Turkey has recently adopted a package of constitutional amendments that
The main aim of this paper is to investigate how these institutional settings a¤ect coup behaviour
7 Another example: Ramon Barquin, the Cuban colonel who staged a failed coup attempt against president Battista
in 1956, was sentenced to a mere six years in jail. (Bonachea et al., 1974)
8 In Madagascar the military installed president Rajoelina in 2009 in response to growing public discontent (Ploch,
2009). In Egypt’s 2011 revolution, the army is thought to have been pivotal in ousting President Mubarak.
9 In Peru, the 1993 constitution states explicitly that coups can never be legitimate: "Power emanates from the
people. [...] No person, organization, Armed Force [...] may arrogate to themselves the exercise of such power. To do so
constitutes rebellion or sedition." (Art.45)
1 0 The Economist, "Erdogan wins again", 16 April 2010 (http://www.economist.com/node/17046653).
5
and which settings can be "optimal" for a social planner. This exercise will also point at the tensions
that are inherent to the design of civil-military institutions. In the …nal sections, I will argue that
these tensions can be understood within the formal framework of this paper. These tensions will be
able to explain shifts away from legitimising military control over politics.
3 Theoretical framework
In this section, I will introduce the key assumptions, mechanism and results of the theoretical model.
First, I propose a model in which coups can be "desirable". I assume that an incumbent government
has a default ideology (say, left wing). Moreover, government can be "low quality" or "high quality" in a
general interest dimension.11 The quality of government in the general interest dimension will be taken
as the welfare measure, which is not sensitive to ideology in the baseline model. If a military government
replaces the incumbent, I assume that it has a higher expected quality than a bad incumbent, but
a lower expected quality than a good incumbent. Under these assumptions, a coup against a bad
incumbent is "desirable" in the general interest dimension. It is not di¢ cult to …nd examples of
cases in which military coups attracted strong popular support because the military ousted unpopular
leaders. For instance, the 1964 coup in Brazil was enthusiastically welcomed in the editorials of all
major newspapers (Stepan, 1971, p110-12). In contrast, a coup against a well-performing incumbent
The explanation for why "undesirable" coups ever occur builds on the second assumption: the
military consists of o¢ cers who care unanimously about the government’s quality, but who are fac-
tionalised along ideological lines. Hence, o¢ cers either share the ideology of the incumbent (left-wing)
or hold rivalling views (say, right-wing). The idea that the military does not just narrowly promote its
corporate interests, but is factionalised along lines similar to those in society at large is supported by
a substantial body of literature.13 If coups are driven by factional interests, rather than the general
1 1 The idea underlying coup friendly constitutions is exactly that, sometimes, coups can be "desirable" in a gen-
eral interest dimension. Following Besley and Kadumatsu (2007), "good" governments might be thought of as being
characterised by low child mortality, high economic growth, high educational attainment and low corruption.
1 2 In Nigeria, gen. Murtala Mohammed, who was widely acclaimed for his e¤orts to cut corruption, was killed in a
coup in 1975. In this case, the interest of one particular faction within the military seemed to have driven the coup (Alli,
2000).
1 3 Stepan (1971, p7) writes for instance: "Descriptions of ideal military institutions which may emphasise such features
as military unity or national orientation often conceal more than they reveal about the interactions between the military
and the political system". Authors who emphasise the role of factionalisation within the military include Finer (1962,
p34-42), Fitch (1998, p109-110), Philip (1974,p75-99 and 1985, p190) and Loveman (1999, p207). Geddes (2003) argues
that the main reason why military regimes are relatively unstable is because the armed forces are typically factionalised
6
interest, they are undesirable.14 Once these rivalling political a¢ liations are introduced in the mili-
tary, staging a successful coup becomes a leader-follower problem, possibly with two-sided asymmetric
information. Successful coup leaders need the support of a critical mass within the military. However,
if political a¢ liations are kept secret, the leader cannot be certain that a coup will be supported and
the o¢ cers who are critical for a coup to succeed may be reluctant to lend their support to a coup
leader, if they do not know his true ideology. They fear that, once in o¢ ce, a coup leader is in an
excellent position to promote policies in line with his ideology. The importance of this coordination
problem may vary because of three main sources, which correspond to the key parameters of the model.
First, the strength of factionalisation will a¤ect the coordination problem. Second, the openness of
political a¢ liation can vary. Finally, the coordination problem becomes less severe if the punishment
Third, I assume it is impossible to condition these punishments for failed coups on the quality
of the incumbent government. There are three reasons for this assumption. First, if the incumbent
government can decide on the punishments it may not be optimal to make punishments conditional
on its quality.15 Second, the quality of the incumbent government may be incontractible, for instance
because it is observable to military o¢ cers, but not veri…able by a court.16 Finally, the organisation
that decides over punishments (e.g. the judiciary, a military tribunal or the executive power) may be
factionalised itself. More generally, low punishments for failed coup plotters could be considered as a
Based on the three key assumptions, I try to answer the question which institutional set-up will
su¢ ciently hold poorly performing governments accountable, while at the same time prevent coups
that are driven by factional interests only. This model yields three key results:
First, high punishments and low political openness achieve …rst best control of the military over
politics, which implies that a successful coup is staged if and only if the incumbent is poorly performing.
This requires that ideology does not dominate the utility function of o¢ cers. In particular, all factions
within the military must be willing to support a coup leader of unknown ideology against a bad
incumbent. The coup experiences of Peru (1948-1968) and Brazil (1948-1964) are consistent with
the positive results of this parameter range, although Brazil’s culture of open political debate in the
between opponents and proponents of military rule. The latter view corresponds to a speci…c interpretation of the
framework developed in this paper.
1 4 Finer (1962, p34-60) distinguishes between coups driven by "national interest" versus "sectional interests".
1 5 A government might implement high punishments to remain in power, regardless of its quality. If government quality
is not observable to the electorate, punishments can be a signal of government quality and a pooling equilibrium with
low punishments may arise.
1 6 This is a standard assumption in the incomplete contracts literature (Hart , 1993).
7
military is not consistent with the "institution design optimum". In contrast, if ideology becomes
more important, it may be optimal to implement low punishments for failed coups, while adhering to
the neutrality of the army. This is a second best solution, which will see coups by military leaders
who share the ideology of the incumbent, at the expense of allowing coups against well-performing
governments. This equilibrium requires that factionalisation within the military is su¢ ciently severe,
so that left-wing followers block a coup against a bad government out of fear that the initiators might
be right-wing. This parameter range is consistent with the recent coup history of Venezuela (1992).
In an extension of the baseline framework, it can be shown that assumptions on the information
structure of the game can be rationalised as outcomes of a communication game in which o¢ cers
can choose to reveal their types or not. In particular, leaders have an incentive to hide their political
a¢ liation, while followers would like to reveal their political a¢ liation. This result is important because
the incentive for followers to reveal their type implies that institutions need to be put in place in order
Third, there is a case in which institutions do not matter, because ideologies are perfectly observable
(for instance because they correspond to ethnicity). This yields a unique equilibrium, characterised
by high coup frequency and coup attempts against both good and bad governments. This result is
Finally, relaxing the assumption that any military government has a higher quality than a bad
incumbent government, the model can be altered to provide an account of "extremist" factions within
the military. This extension highlights that the moderator pattern of civil-military relations, which
relies on the willingness of military o¢ cers to support their leaders against poorly performing incum-
bents, necessarily puts countries at risk of take-overs by an extremist faction. In this set-up, it is
optimal to prevent any type of coups under very weak assumptions. This extension may explain the
gradual move away from the institutional and constitutional promotion of coups.
4.1 Set-up
Consider an incumbent government, indexed g, who is passive and has a left-wing ideology (Ig = L)
without loss of generality. The government can be high quality ("good") or low quality ("bad"), which
is indicated by a quality variable Qg 2 f 1; 1g. Qg is known to all players. To carry out welfare
8
analysis, a prior of the government being good is introduced: Pr(Qg = 1) = .
There are two players in the model: the leader and the follower, indexed i (for initiator) and f
(for follower) respectively.17 The leader moves …rst and decides whether or not to stage a coup, the
follower then decides whether or not to support the coup. A coup succeeds if an only if the leader
takes the initiative and the follower supports the coup. Both o¢ cers are characterised by an ideology
variable Ij 2 fL; Rg, j = i; f . Every o¢ cer knows his own ideology, but cannot observe the other
player’s ideology directly. The priors are common knowledge and given by:
Pr(Ii = L) =
Pr(If = L) =
The follower receives a private signal t 2 fL; Rg about the leader’s type, while the leader receives a
private signal s 2 fL; Rg about the follower’s type. The precision of the signal is common knowledge
1
Pr(t = xjIi = x) = = , x 2 fL; Rg
2
Pr(s = xjIf = x) = , x 2 fL; Rg
The signal about the leader’s type, t, is uninformative, i.e. = 21 , while the precision of the signal
O¢ cers derive utility from the government’s quality, Q, and from its ideology, through an indicator
function which takes on value one if an o¢ cer shares the ideology of the government. measures the
importance of ideology. The assumption that is the same for both left-wing and right-wing o¢ cers
1 7 An example of an initiator would be the general in charge of the army division stationed in the capital. We may
think of the follower as the o¢ cer in charge of the main army division outside the capital.
1 8 In section 2, an explicit microfoundation is provided for these parameter restrictions: in an extended model, leaders
want to hide their types, while followers would like reveal their types. Importantly, this implies that institutionally
restricting the openness of political a¢ liation matters, since it will counter the incentive of followers to reveal their
types.
1 9 This assumption can be relaxed without a¤ecting any of the (qualitative) results, as long as 1 < E(Qi ) < 1.
9
can be relaxed without changing the key results. A leader staging a failed coup faces a punishment
P > 0, which is not conditional on the quality of the incumbent. As argued above, P will depend on the
general legitimacy of coups: countries that allow for coups to occur under certain circumstances, will
…nd it harder to credibly punish the plotters of failed coups. It is assumed that there is no punishment
The expected utility of a military o¢ cer of type j if a government of type k is in o¢ ce is given by:
8
>
< P + Qk + (Ij = Ik ); if j = i and i stages a failed coup
Uj = ; j 2 fi; f g
>
: (Ij = Ik ) + Qk ; in all other cases
The action to stage a coup attempt is given by a 2 f0; 1g, where 0 stands for no coup attempt and
1 stands for a coup attempt. The decision to back up a coup attempt is given by b 2 f0; 1g, where 0
stands for the decision to resist a coup attempt and 1 for the decision to support a coup.
1. Nature draws the quality of the government, Qg , the type of the leader, Ii , and the type of the
follower, If .
In the model described here, "…rst best" control of the military over politics would imply that successful
coups are staged if and only if the incumbent is poorly performing. This outcome maximises the
expected government quality (E(Q)), which will be our welfare measure and is insensitive to ideology.21
It can be seen straightaway that ine¢ ciency could stem from two possible sources. First, by not being
able to condition punishments on government quality, there is a trade-o¤ between allowing the military
2 0 There are several cases of coup leaders being put on trial after they have left government. However, these trials
often relate to crimes (such as human rights abuses) committed while in o¢ ce, rather than to the act of staging a coup
in itself.
2 1 This assumption will be relaxed in section 4.
10
to intervene when it is desirable and allowing the military to intervene in case it is undesirable. For
punishments to have a bite, there needs to be uncertainty about the success of the coup, which is
introduced in this model by the uncertainty about the type of the followers. Second, even if punishments
could be made conditional on the quality of the incumbent, then the followers may not support a coup
against a poorly performing government (out of fear that the new military government will be of the
opposing ideology). The …rst ine¢ ciency may lead to either too many coups or too few coups, the last
ine¢ ciency can lead to too few coups. The next section contains a detailed description of the welfare
This section deals with the case in which ideology is unimportant. In this case, both left-wing and
right-wing o¢ cers may agree to oust a poorly-performing incumbent, even if there is uncertainty about
The combination of signal precision , the strength of punishments P and the importance of ideology
will determine the optimal strategies for di¤erent types of leaders and followers. A …rst result is
almost trivial: if ideology is less important than the quality of government ( < 1), then the …rst best
solution will always be reached. Leaders will stage coup attempts if and only if incumbents are bad,
followers will support coup attempts if and only if leaders are bad.
functions of military o¢ cers. As the followers do not have certainty about the type of the leader, they
do not necessarily want to support a coup by a leader of unknown ideology, even if the government
is bad. For the R followers this is not important, as they are always glad to see a poorly performing
government. In contrast, the L followers are worried that they may support a military leader of the R
faction. The following condition guarantees that the L followers prefer to support a random military
leader (who has a prior of being L), against a poorly performing government:
11
1
1< ) < (1)
1
This condition states that the pay-o¤ from the incumbent (on the left hand side) is smaller than
the expected pay-o¤ of the military leader (on the right hand side), which is the sum of the expected
quality (0) and the expected ideology pay-o¤. This latter component consists of the probability of
facing an L leader ( ) times the strength of ideology ( ). Hence, under condition (1), both types of
followers can be united against a poorly performing government. The leaders can now anticipate the
reaction of the followers, given that the latter believe that both leaders always stage a coup (which
made the relevant prior). Now, it can easily be seen that both the L leader and the R leader want to
topple the poorly performing incumbent if they would be supported in such attempts. They know that,
on average their rule will be better in the general interest dimension than the rule of the incumbent.
Also, they can anticipate the ideology of their own regime. Hence, there is a Perfect Bayesian pooling
equilibrium in which both types of leaders stage a coup attempt, which is always supported if followers
form beliefs consistent with these equilibrium strategies. This reasoning proves the …rst part of the
following result:
1
Proposition 2 If 1 < < 1 ("low factionalisation") and Qg = 1 ("bad incumbent"), there
exists a pooling equilibrium in which both L and R military leaders stage coup attempts against poorly
performing incumbents and both L and R followers support such attempts. This pooling equilibrium
However, this does not guarantee the …rst best solution, as R leaders may want to stage coups
against well-performing governments. This is because ideology is more important than government
quality. If the incumbent is good, L followers will never support a coup, whereas R followers will
always support a coup. L leaders will never want to stage a coup against a good government, but R
leaders may bet on the followers being R and attempt a coup against a good incumbent. Given these
incentives, any equilibrium with coup attempts against a good government will be separating in the
leader’s type.
1
Proposition 3 If 1 < < 1 ("low factionalisation") and Qg = 1 ("good incumbent"), the only pure
strategy equilibrium sees only R followers supporting coups and L followers never supporting coups.
12
The optimal strategy for the L leader is to never stage coups. The optimal strategy of the R leader is
determined by P and , for which there exist thresholds P; R (P ) and L (P ) such that:
¯
The positive results now easily allow us to identify the parameter region in which the …rst best solution
is achieved.
1
Corollary 4 If 1 < < 1 ("low factionalisation"), there exists threshold P; R (P ) and L (P )
¯
such that for P >P and < R (P ), the …rst best is achieved.
¯
This result can partly rationalise the institutions I presented under the stylized facts. While not
allowing information on political a¢ liation to be revealed is consistent with optimal control of the
military over politics in my model, it also prescribes punishments that are higher than those observed
in reality.
Both the cases of Chile (1971-1980) and Peru (1946-1970) …t into the "low factionalisation" parameter
range, of which the crucial characteristic is that di¤erent factions within the army can be united
against a poorly performing incumbent. In Peru, general Odria removed a mildly left-wing president
from power in 1948, on accusations of being too soft on violent militants. This latter position was
widely supported within the army. In 1962, the army brie‡y intervened to settle an election with
unclear results. It installed Belaunde as a president, a left-wing populist who later sought a coalition
with conservative politicians. In 1968, the army ousted Beluande, over a deal with the International
Copper Cooperation of which the unfavourable clauses were kept secret. This deal was unacceptable to
most Peruvians, and the leading general, Velasco, could easily …nd su¢ cient support for his coup. In
all cases, the interventions of the military were widely welcomed by the Peruvian public and Peru did
not see any coups against governments that were perceived to be well-performing. Also, Peru did not
13
1
see any failed coup attempts within this period. These two elements are consistent with the < 1
parameter range described above. In further support of our approach, Philip (1978) reports that the
Peruvian army was factionalised between left-wing radicals and right-wing conservatives. Nevertheless,
these political a¢ liations were e¤ectively kept secret. The key to the success of the coup attempts
against poorly performing governments was the fact that both factions were willing to take the risk of
supporting a coup leader of ex ante unknown ideology. This was a relevant trade-o¤, as is illustrated
by the fact that General Velasco turned out to promote radical policies which "[...] would have been
too extreme for most o¢ cers who supported the coup".22 Thus the relevant parameter range seems
1
to be 1 < < 1 , in which the faction of the army that was ideologically closest to the incumbent
Chile seems to have found itself in a similar equilibrium when Pinochet took power in 1973. His
move was widely supported by large segments of the population and a strong majority within the
military (Philip, 1985, p308). Historical evidence suggests that the main division within the military
was between o¢ cers with Christian democratic sympathies and right wing conservatives,23 but at the
time these a¢ liations were strictly secret (Philip, 1985, p307). The moderate Christian group seemed
to have had the strongest support, but once installed in power, Pinochet turned out to belong to the
second group. As Philip indicates, it is highly unlikely that Pinochet would have gained the support of
the military if his views were known before the coup. As in Peru, the relevant equilibrium range seems
1
to have been 1 < < 1 , where the faction ideologically closest to the incumbent government would
not support a coup from the opposing faction in a complete information world, but is willing to take
the risk of supporting a "random o¢ cer" if a coup is staged against a poorly performing government.
In contrast to the secret a¢ liation in the Peruvian and Chilean military, the Brazilian military had
"[...] in fact institutionalized military debate over public and political issues" (Stepan, 1971, p44).
The most important forum was the Military Club in Rio de Janeiro, of which the biennial elections
were traditionally fought over political issues. The three coups that succeeded in the 1950-1970 period
all targeted a government with low legitimacy (Stepan, 1971, p108). However, Brazil did also see
failed coups in this period. The two failed coups both tried to prevent a newly elected politician from
assuming power. As these governments had not revealed their quality yet, these coups are unlikely
to have been staged in the general interest. This suggests that Brazil was in an equilibrium with
2 2 Philip,
1978.
2 3 Assuming that the Christian Democrats shared Allende’s ideology, as we implicitly do here, is a simpli…cation.
However, Philip (1985) suggests they would have preferred Allende to Pinochet, if they had kown his agenda.
14
1
1 < < 1 , but its open political debate put in the range where > R. Therefore, factions
not sharing the government’s ideology (R) were willing to act upon a favourable signal (s = R) and
stage coups against governments that were not poorly performing and had a high degree of legitimacy.
This suggests that Brazil was not at its institutional optimum and could have prevented these coup
attempts (and the risk of them being successful) by having less open political a¢ liation.
In this section, ideology is so important that the military can no longer be united against a bad
government.
> max( 1 1 ; 2) implies that the L follower is no longer willing to back a coup attempt if both L and
R types always try to oust the bad government. As a consequence, there are no pure strategy Bayesian
equilibria in which the L follower supports a coup. The R follower, as before, will continue to support
any coup. Therefore, the parameters of the model will only a¤ect the optimal strategies of the leader.
Deriving these optimal strategies boils down to an exercise of comparing the expected costs and bene…ts
of staging a coup. The set of possible equilibria that can be attained by manipulating punishments
and signal precision can be represented in a diagram that delineates the optimal strategies for each
type of leader for each possible signal, the relevant quality of the incumbent and any combination of
P and .
The optimal strategy space of each type of leader in a particular situation is delineated by a
V-shaped curve in the ( ; P )-space. For example, the (^ L ; ^ R ) curves divide the strategy space in
di¤erent optimal strategies for an L leader who faces a bad government, upon receiving an L or R
signal respectively. In the region below the ^ L curve, signal precision and punishments are su¢ ciently
low, so that left-wing leaders are still willing to stage coups against bad incumbents regardless of the
signal (region A). As left-wing followers no longer support coups, the L leaders count on the support
of right-wing followers instead. Therefore, if signal precision rises above the ^ L and ^ R curves (regions
marked with a " "), the L leader is only willing to stage coups if he receives a favourable signal, i.e. that
the followers are right-wing . Finally, if we move underneath the ^ R curve (the regions marked with
" "), the L leader will never stage a coup as the signal has become unreliable and the punishments
for failure are too severe. Similarly the ( L; R) curves describe the optimal strategies of an R leader
15
σ
^R
σ
∼
D*
−R
σ
A*
B*
D**
∼
∼R
A
^ L ∼ B** ∼ E** σ
σ −L ∼L
σ C** σ F**
A**
1/2
0 P
Strategies of L leader:
** No attempt
Strategies of R leader:
F: No attempts
16
facing a good government. Finally, the (~ L ; ~ R ) curves describe the optimal strategies of an R leader
facing a bad government. Combined in the same diagram, the ordering of the (^ L ; ^ R ), ( L; R) and
(~ L ; ~ R ) curves is intuitive given their explanation: L o¢ cers (^ L ; ^ R ) are less coup prone than R
leaders facing a good government ( L; R ), who are in turn less likely to stage a coup than R leaders
Figure 1 clearly demonstrates the trade-o¤ between holding governments accountable and overshooting
by ousting good governments. The equilibria in which the L leader stages a coup if and only if the
incumbent is poorly performing (the …rst best L strategy), imply necessarily that the R leader will stage
coups against good governments. This follows from the fact that the utility functions are dominated by
the ideology component. Therefore, the …rst best is no longer achievable. Comparing the properties of
each possible equilibrium with the …rst best benchmark, it can easily be established that the parameter
ranges in …gure 1 can be ordered as follows in terms of the expected quality of government (E(Q)):
A > A >A
B > B
B > D
C > B >D
C > E >F
This leaves three candidates for a welfare optimum. First, there is an equilibrium with low punish-
ments and low signal precision (A), in which the L leader stages a coup if and only if the government
is bad and the R leader always stages a coup attempt. These attempts are successful whenever the
follower is R. Second, there is an equilibrium with intermediate punishments and high signal precision
(B ), in which both L and R leader condition their responses on the signal. Only if the signal is
favourable, the L leader attempts a coup against a bad government and the R leader attempts a coup
against the good government . Against a bad government, the R leader will still stage a coup attempt
regardless of the signal. Finally, there is an equilibrium with low signal precision and intermediate
punishments (C ), in which only R leaders stage coups and they do so if and only if the incumbent
is bad. The following proposition states that only (A) or (C ) can be optimal.
17
Proposition 5 If > 1 , we can de…ne thresholds P^ ; P, P > 0 and ~ L > 12 , such that the
1
¯
combinations of P and yielding pure strategy Bayesian equilibria which (weakly) maximise E(Q) are
given by:
2. If > : P < P^ and < L < ~ L . In the corresponding equilibrium an R leader always stages
a coup. An L leader stages a coup if and only if the government is bad. An L follower never
This proposition implies that it is never optimal to have open political debate if both punishments
and signal precision can be manipulated. Note that the level of punishments does not a¤ect the
welfare function directly (it only a¤ects the equilibrium chosen), whereas the signal precision has a
direct impact by determining the likelihood of correct recognition of the followers (and, hence, the
probability of successful coups). Given that signal precision directly enters the probability of certain
equilibria, its e¤ect should be monotonic. This means that either the highest signal precision or the
lowest signal precision consistent with this equilibrium must be optimal. However, in terms of the
and strictly preferred, as soon as we rule out the cut-o¤ case of perfect precision ( = 1)). It is shown
in the appendix that limiting is only optimal within equilibrium B if it is better to discourage
Comparing the remaining candidates (A and C ), the advantage of the equilibrium C (interme-
diate punishments) is that a good government is never ousted. The precision of the signal is su¢ ciently
low and punishments are su¢ ciently high, so that the R leader only stages a coup attempt against
a bad incumbent, regardless of the signal. The disadvantage is that a leader of the same ideology of
the government will never challenge a bad government. He will …nd the punishments too high and the
information he has about the followers’political stance too scarce to ever attempt a coup. In equilib-
rium A (low punishments) the L leader is willing to act against a bad government. However, this also
implies that the R leader is willing to oust a good government under any signal. Which equilibrium
dominates depends on whether < . If < , the probability of the government being good is
18
smaller than the probability of the leaders being of type L. Hence, the optimal equilibrium has coups
against bad governments staged by the L leader at the cost of introducing R-led coups against good
governments. If > , the probability of the government being good ( ) is higher than the probability
of the leader being of the L type ( ). Hence, the optimal equilibrium will have no coups against the
good government at the cost of not having any coups by L leaders against the bad government.
These results are consistent with some of the stylized facts. Armies should be politically neutral,
and punishments for failed coups should be limited. The result on political neutrality is the most robust
…nding, as a best case for open political a¢ liation was made. Also, comparing the expected quality
of government under low and high factionalisation equilibria, the optimum under low factionalisation
clearly dominates the optimum under high factionalisation. Therefore, institution design could in theory
address the degree of factionalisation within the military, and there are institutions that seem to ful…l
exactly this role. For instance, military training places emphasis on the development of a "military
identity" that complements and possibly replaces any identities previously held by o¢ cers (Philip,
1985, p177-200). Also, recruitment policies may aim at attracting a homogenous set of candidates.
While I focus on the neutrality of the army at the level of individual o¢ cers, the neutrality of the
military as an institution may be seen an institutional arrangement that makes it easier for soldiers to
The case of Venezuela (1992) can illustrate these results. In the late eighties the Venezuelan economy
was severely hit by a decline in oil prices. Carlos Andres Perez’s government was alleged to be highly
corrupt and had reneged on campaign pledges to push through a series of unpopular reforms (Norden,
2001). Against this background, Hugo Chavez staged a coup attempt in 1992. Chavez’ coup failed
because he did not have the support of key players within the army. Importantly, political a¢ liations
other than links with the leading AD party (Acción Democrática) were not allowed to be shown within
the military (Norden, 2001, p121). Gott (2005, p64) indicates that Chavez only knew that he had
the support of about 10% of the army. He did not have any clues about the views of the remaining
90%. In terms of this model, it may have been that some conservative o¢ cers would actually like to
oust Andres Perez as well. The failure could then be explained by the fact that Venezuela was in the
high factionalisation range ( 1 1 < ). In this range, the majority of the army, which proved to be of
the same ideology as the government, chose not to back a coup leader of ex ante unknown political
19
a¢ liation: Hugo Chavez. In the aftermath of the failed coup, Hugo Chavez was put on trial and given
a long prison sentence, but the wide-ranging amnesty powers of the president allowed the successor of
President Perez to set him free after only two years (Gott, 2005, p119).
Evaluating the military institutions of Venezuela, the condition > may have been satis…ed in
Venezuela, implying that the average quality of incumbents is su¢ ciently high relative to the probability
of the coup leader sharing the ideology of the incumbent. As indicated in proposition 4, it is optimal to
have an equilibrium with political neutrality in this range. Also, the expected punishments for a failed
coup may have been su¢ ciently low to induce a coup attempt, assuming that Chavez anticipated
his relatively mild punishment. The failure of this coup, which was due to the strong position of
conservatives in the army, could not have been prevented by changing either P or .
This section develops a theory of the disclosure (or hiding) of political a¢ liation. This analysis endo-
genises "talking politics". It provides a microfoundation of earlier assumptions and suggests a speci…c
institutional framework which can implement the equilibria found before. The proposed game boils
5.1 Model
The model builds on the one introduced in section 1. The signals are now abandoned and replaced by
a message from the leader. To this aim, a stage is added to the game in which the leader can send a
message about his type: I^i 2 fR; L; N g . The option to announce N has to be interpreted as remaining
neutral. This option can be attractive for o¢ cers if there is a cost to lying about one’s type. This cost
8
>
< P + Qk ; if i stages a failed coup
Ui =
>
: (Ii = Ik ) + Qk (I^i 6= Ii ) ; in all other cases
The follower can also reveal his type I^f 2 fR; L; N g. For the follower, no cost of lying is assumed
(as this would only strengthen our results). The utility function of f is:
20
Uf = (If = Ik ) + Qk ; in all other cases
This game is now the same as before, augmented with a communication stage. The timing of the
game is now:
1. Nature draws the type of the leader, Ii , and the type of the follower, If .
2. The leader announces his type I^i and the follower announces I^f simultaneously.
Solving this problem by backward induction, o¢ cers in the communication stage anticipate the
positive results derived earlier. The following proposition states that followers prefer to reveal their
ideology.
Proposition 6 For > 1, the only pure strategy equilibrium communication is separating in the
follower’s type.
Proof. The R follower would like to reveal his type in order to attract coup attempts against the
incumbent. The L follower can always block a coup attempt so has no strategic incentives in the
communication stage. As soon as the L follower chooses a pure strategy message, there is an separating
equilibrium in which the R follower chooses a di¤erent pure strategy message. This result holds
Turning to the communication strategy of the leader, it is intuitive that all messages will be
uninformative in a pure strategy equilibrium. Suppose there is an equilibrium in which I^i (Ii ) = Ii .
L followers if the government is bad. Therefore, the R leader is strictly better o¤ by deviating and
announcing I^i (R) = L. Therefore I^i (Ii ) = Ii cannot be an equilibrium. This result holds regardless of
21
the communication strategy of the followers: even if the followers reveal their types, the leader cannot
rule out the followers being L when they decide to show their a¢ liation or not.24
1
Proposition 7 For 1 < < 1 , the only pure strategy perfect Bayesian equilibria that satisfy the
For > 1
1
, the only pure strategy perfect Bayesian equilibrium has I^i (R) = I^i (L) = N .
This proposition explicitly states an insight used in the previous section: the leaders have an
1
incentive to conceal their true types. For < 1 , coups against a bad incumbent are always supported
by any type of followers in this parameter range. This equilibrium relies on the secrecy of political
a¢ liation. The cost of lying (in combination with the intuitive criterion) makes sure that leaders pool
by both claiming to be L or to be neutral. From an equilibrium in which both types of leaders claim
1
to be R, only an L leader would strictly prefer to deviate under the belief that he is L. For > 1 ,
ideology is so important that L followers are not willing to support a coup if both L and R leaders
stage coups if Qg = 1. Again, the cost of lying allows us to rule out certain messages, in this case
only viable equilibrium is characterised by I^i (R) = I^i (L) = N . In this case, the equilibrium of both
o¢ cers claiming to be left-wing is not sustainable. As the R leader is not supported by L followers
anyway in equilibrium, he would deviate to claim to be neutral or R in order to prevent the cost of
The insight that coup plotters should refrain from having open political a¢ liations is emphasized in
Luttwak’s (1968) practical coup manual. An illustration of this theory can be found in the biographies
of generals Pinochet, Velasco and Chavez. None of these coup leaders revealed their a¢ liations before
they successfully assumed power. Velasco had clear links with the conservative elements in society
and concealed his radical left wing sympathies (Philip, 1978, p43). He seems to have been in the
pooling equilibrium where leaders claim to be of the incumbent government’s ideology. Chile’s general
Pinochet concealed his right wing ideology as well and kept a neutral pro…le (Philip, 1985). Hugo
Chavez did not have an openly left wing agenda either when he staged his 1992 coup (Norden, 1992).
2 4 The condition < + 1 guarantees that the cost of lying is not too large to prevent leaders of type R to lie.
2 5 All these results hold for an arbitrarily small > 0.
22
1
In line with the earlier hypothesis that Venezuela is best described by the > 1 region, Chavez was
neutral and had not clearly associated himself with conservative ideologies.
An important conclusion is that this section justi…es my so far implicit assumption that institutions
can alter the actual openness of political a¢ liation (i.e. changing ). As followers have an incentive to
reveal their types, only appropriate institutions will prevent them from doing so. In principle, these
institutions could be developed entirely within the armed forces, in line with Huntington’s ideal of a
professional, autonomous military (Huntington, 1981). However, the analysis in this section points
at the tension between an independent army and the incentives that shape political communication
within the army. This may call for targeted civilian control over the army. For instance, purging powers
can be granted to a government that aims to prevent coups. As coups are more likely if the followers
show their a¢ liation, a government that wants to avoid coups will have an incentive to discourage
o¢ cers from revealing their (diverging) political views. As indicated earlier, the military in Brazil
(1950-1970) and Venezuela (1980-2002) di¤ered strongly with regard to their openness of political
a¢ liation. One can compare the promotion structures of both militaries to illustrate the hypothesis
that di¤erent degrees of open political a¢ liation can be implemented by varying the government’s
control over promotions. Throughout the period 1958-1998, Venezuelan o¢ cers were expected to be of
the leading political parties’ideology or not to hold any political views at all. The congress had strong
powers over military appointments, which was seen as an explicit strategy to reduce the armies’threat
to intervene (Norden, 2001, p126; Trinkunas, 2001, p171). In this context, Chavez’activities, which
are reported to have started in the late seventies, had to remain secret (Norden, 2001, p122). Chavez
had little information about his overall support within the army and Norden ascribes the failure of
the coup partly to the unwillingness of the coup plotters to approach extra o¢ cers (Norden, p120). In
contrast, at least until 1962, the Brazilian army had a largely independent promotion structure mainly
based on achievement in the military academy (Stepan, 1971, p51). The limited direct in‡uence of the
government on army promotions was necessary to allow o¢ cers to take open political stances.
6 Extensions
Finally, I will present the positive results of this model for the case in which institutions do not matter
because types are directly observable. This seems an appropriate assumption if ethnicity or regional
23
origins form the most important political dimension, as these types cannot be concealed. Going back
to the full model of section 1 and reintroducing the drawing of Ii , this implies: = = 1.
Proposition 8 If < 1, then there is a coup attempt if and only if Qg = 1 and followers always
support this coup. If > 1, then there is a coup attempt from an R leader whenever the follower is
This proposition is a direct corollary of earlier results. To summarise, in countries where the main
con‡ict is along ethnic lines, military coup attempts are expected: (1) to target both well-performing
and poorly performing governments; (2) to be frequent, possibly in spite of severe punishments; and
An illustration of this case is Nigeria. Nigerian politics have been dominated by ethnic con‡icts ever
since its independence. The Nigerian military recruited in all regions and, as is clear from the account
of Alli (2000), the regional a¢ liation of military o¢ cers was easily observable. Typically, Northern
o¢ cers dominated the army and used this position to block the South to gain more in‡uence in politics
(Luckham, 1974). First, as predicted by my model, Nigeria saw coups both against poorly performing
governments and well-performing governments.26 For instance, Nigeria’s …rst coup, in 1966, was
entirely driven by the sectional interests of Ibo o¢ cers who aimed to overthrow a government which
was dominated by Muslim leaders. However, Northern o¢ cers also removed the Northern General
Gowon from o¢ ce in 1975, because he did not deal e¤ectively with the economic crisis. A second
prediction of our model is a high frequency of coup attempts. Nigeria faced nine coup attempts in the
period 1966-2000, even though the leaders of the two (o¤ equilibrium) failed coups were all executed.
Also, of these nine coup attempts, only two failed. Therefore, the case of Nigeria closely …ts the
predictions of our model. Also, the analysis may explain the empirical result of Collier and Hoe- er
(2007) that ethnic dominance, in the sense that one ethnic group accounts for more than 45% of the
population, leads to signi…cantly higher coup risk. The authors do not provide an explanation for this
result, but their variable ethnic dominance can be expected to be correlated with ethnic polarisation,
as I discussed here.
In conclusion, the case of observable types yields several benchmark predictions that naturally lend
themselves to empirical testing. However, the scope for institution design is extremely limited in this
2 6 The coup attempts I consider are: January 1966, July 1966, 1975, 1976, 1983, 1990, 1993, 1995. Of these attempts
24
case, as political a¢ liations are open by default.
Relaxing the parametric assumptions in earlier sections, coups can be undesirable for two main reasons:
(1) coups could target a well-performing government, or (2) coups could replace any government with
a regime that is even worse.27 The assumption that a military government is always better than a
bad incumbent means that I discarded the latter possibility earlier. The welfare function that was
proposed was not sensitive to ideology. However, it is conceivable that a faction of the army does not
promote the general interest and less so than the incumbent government. Such factions can be thought
The model can now be re-interpreted by assuming that the military consists of left-wing moder-
ates and right-wing extremists, whereas the government is moderately left-wing. The positive results
derived earlier will carry through. However, a new welfare function will now assign a positive value
to the ideology of the incumbent, for instance by including ( > 1). Under this assumption, the
institutions that maximise expected government quality will promote coups by moderate o¢ cers if and
only if the incumbent is bad but discourage coups by right-wing o¢ cers. This means that the …rst
best is no longer achievable, even if factionalisation within the military is low, because the right-wing
extremists stage coups in the underlying equilibrium whenever the government is poorly performing.
This same solution, which was characterised by high punishments and low political openness, will now
be the second best optimum though, as no other institutional setting prevents coups more e¤ectively in
the case of low factionalisation. The concept of "extremism" may appear to be at odds with the notion
1
of "low factionalisation" ( < 1 ). Nevertheless, ideological views within the army could be close
to each other, in particular if the military’s corporate interests are taken into account. Furthermore,
the "low factionalisation" parameter range could be consistent with substantial ideological di¤erences
(high ) in combination with a su¢ ciently low ex ante probability of an extremist coup leader (low
1 ). In a richer game, followers could gradually learn about the distribution of in the universe
of coup plotters. If extremism enters the welfare function, the high factionalisation range will have
a di¤erent optimal solution in comparison to the earlier results. It can be seen straightaway that it
may no longer be optimal to limit punishments, as an equilibrium with high punishments and political
2 7 The strong economic growth of Chile under Pinochet contrasted sharply with the economic crisis during the Allende
regime. Nevertheless, it would be reasonable to adopt a social welfare function that gives more weight to the human
rights abuses of the Pinochet regime than to Chile’s economic growth experience.
25
neutrality has no coup attempts by right-wing o¢ cers and beats an optimum with limited punishments
in which a right wing leader stages a coup if and only if the incumbent is poorly performing.
As an illustration, Brazil was characterised by repeated take-overs (in 1945 and 1956) initiated by
moderate factions within the military who quickly organised a return to democracy until the 1964 coup
saw the military taking over government for the next 21 years. The military government of Brazil was
responsible for at least 339 disappearances and assassinations, on top of wide-spread torturing and
jailing of political opponents. While the initial coup leader, General Castello Branco was moderate
and committed to democratisation (Stepan, 1971, p234), his position in the army was to weak to
prevent hard-line o¢ cers to in‡uence his policies and eventually replace him in 1968. It can be argued
that Brazil’s policies did not consider a welfare improvement over the -arguably poorly performing-
Goulart government the military replaced in 1964. My analysis suggests that this outcome could
result from the same equilibrium that underlaid the relatively benign "moderator role" of the military
in earlier years. If the L faction is interpreted as prodemocratic elements within the military and the
R faction as o¢ cers with antidemocratic ideals and a repressive agenda, then my model suggests that
any equilibrium in which L followers are willing to oust bad governments, also allows R followers to
oust bad governments. Therefore, the break-down of the "moderator pattern", in which the military
intervenes for short periods without fundamentally changing democratic institutions can be understood
military. Once "moderate" o¢ cers learn about the distribution of "extremist ideologies", they may no
longer be willing to support any coup leader. This may shift the equilibrium to the high-factionalisation
range, in which coups are rarer and more e¤ectively prevented by institutionalising high punishments
7 Interpretation
An important limitation of the theoretical framework introduced in this paper is its seemingly para-
doxical assumptions on the strength of political institutions. It was argued that the military could be
a de facto moderator if political institutions are su¢ ciently weak. However, the subsequent analysis
argued that for the military to be an e¤ective arbitrator, political institutions must be su¢ ciently
strong to generate credible civil military institutions. Such institutions can both ban open political
26
a¢ liations and credibly implement punishments for failed coup plotters. Given that the incumbent
could have an obvious incentive to severely punish failed coup plotters, restraint in punishment could
be a sign of relatively strong political institutions. In the example of Nigeria discussed earlier, all failed
coup plotters were executed. Therefore, the poor coup performance of Nigeria could partially be due
to poor political institutions as well. As a corollary, the military will be a poor arbitrator of domestic
politics if political institutions are very weak. Hence, my theoretical framework suggests an ordering
of political institutions. The analysis in this paper is most relevant if political institutions are in an
intermediate range, su¢ ciently strong to constrain the behaviour of the military, but su¢ ciently weak
to legitimise military interventions in politics. These institutional capabilities may change over time
and the next subsection considers the recent break-down of the moderator pattern in more detail.
It seems appropriate to revisit the central idea that military coups can e¤ectively hold governments
accountable. I argued that this idea was widely spread and de facto military in‡uence over domestic
politics remains strong. Therefore, I believe that this is an adequate assumption to make in mod-
elling coup behaviour and understanding the evolution of civil-military institutions. Public opinion
has recently become more sceptical about the bene…cial e¤ects of military intervention, as exempli…ed
by the increasing constraints to the legitimacy of military power over politics. My analysis provides
three explanations for such a shift in ideas. First, my model implicitly assumed a failure in holding
governments accountable by other means than coups. If democracies have become more e¤ective, then
military coups may have become super‡uous. Second, it may be that political elites or social planners
have overestimated the expected quality of military interventions in the past. The experience of the
brutal military regimes in Latin America or, more recently, in Burma, may have lowered the expec-
tations about military rule. Finally, my model also suggests that the so-called "moderator pattern
of civil military relations", in which the military intervenes for short periods without fundamentally
changing democratic institutions, is only one possible outcome of the underlying equilibrium. Once
o¢ cers are willing to support a leader of unknown quality against a bad government, they take the
risk of supporting a leader with a hidden extremist agenda. The coup experiences of Brazil and Chile
demonstrated how relevant this risk is. Ultimately, this risk is weighed against the cost of a poorly
performing government. For this reason, coups may become less attractive as societies become more
prosperous and moderate o¢ cers are no longer willing to bet on the ideology (or quality) of a military
27
coup leader.
8 Conclusion
This paper analyses the circumstances under which coups are aligned with the general interest. The
analysis is based on three crucial assumptions. First, a military leader is assumed to be of average
quality, so that a military intervention leads to higher government quality if the government is poorly
performing. Second, it is assumed that the military is factionalised and politicised. Third, punishments
The main results are derived under the assumption that political a¢ liation is not directly observ-
able. Focusing on the expected government quality as the objective function, the military is best
in performing its role as a check on the government if ideological cleavages within the military are
su¢ ciently limited and coup plotters are su¢ ciently likely to share the government’s ideology. Once
ideology becomes more important, the optimal institutional set-up has low or intermediate levels of
punishments and no open political a¢ liation. This setting prevents coups against good governments
that are driven by the factional interests of the opposition, while the su¢ ciently low punishments can
induce the army to intervene if the government is poorly performing. In the special case of perfectly
observable political a¢ liation, the model yields an equilibrium with high government turnover, in
which coups are staged both against well performing and poorly performing governments.
The most important limitation of the current paper is probably its strong assumption on the form
that punishments can take. It would be interesting to collect systematic data on how leaders of
failed coups are punished and to develop a model that endogenises punishment and explains empirical
patterns. The framework developed in this paper could provide a useful starting point for this analysis.
A second, very natural extension would be to endogenise the quality of government (e.g. as a moral
hazard problem). Nevertheless, the results of such an exercise can be expected to be qualitatively
The analysis in this paper can help to understand the factors that contribute to civilian control
over the army. It was argued that the risk that is inherent to the moderator pattern of civil-military
relations can explain the breakdown of this pattern. It also highlights how a combination of factionali-
sation and uncertainty about political views within the military can be an important driver of gradual
demilitarisation of politics.
28
References
[1] Acemoglu, Daron, and James A. Robinson (2006). Economic Origins of Dictatorship and Democ-
[2] Acemoglu, D., Tichi, D., Vindigni, A., (2009). A Theory of Military Dictatorships, American
[3] Alli, M.C. (2000). The Federal Republic of Nigerian Army. Malthouse Press Limited. Ikeya, Nige-
ria.
[4] Belkin, A., Schofer, B., (2003) Toward a Structural understanding of coup risk, Journal of Con‡ict
[5] Besley, Timothy and Masayuki Kudamatsu (2008), Making Autocracy Work, in Elhanan Helpman
[6] Besley, T., Robinson, J., (2011). Quis Custodiet Ipsos Custodes? Civilian Control over the Mili-
[7] Bonachea, R.L., San Martín, M. (1974). The Cuban insurrection: 1952-1959, Transaction Pub-
lishers.
[8] Cho, I.K., Kreps, D.M. (1987). Signalling Games and Stable equilibria. The Quarterly Journal of
[9] Collier, P. and Hoe- er, A. (2007). Military Spending and the Risk of Coup d’Etats, mimeo,
Oxford University.
[10] Crescenzi, M.J.C. (1999). Violence and Uncertainty under Transition. The Journal of Con‡ict
[11] Einaudi, L.R. (1998). The politics of security in the Western Hemisphere, Parameters - the US
[13] Finer, S.E. (1962). The Man on Horseback - The role of the Military in Politics. Transaction
Publishers, London.
29
[14] Fitch, J.S. (1998). The armed forces and democracy in latin america, John Hopkins University
Press, Baltimore.
[15] Geddes, B. (2003). Paradigms and Sandcastles: Theory Building and Research Design in Com-
[16] Gott, R. (2005). Hugo Chávez and the Bolivarian Revolution, Verso, London.
[17] Hart, O., (1993). Incomplete contracts and the theory of the …rm. In The Nature of the Firm, eds.
Oliver Williamson and Sidney Winter, pp. 138-158. New York: Oxford University Press.
[18] Huntington, Samuel P. (1981), Soldier and the State: the Theory and Politics of Civil-Military
[19] Janowitz, M. (1960). The Professional Soldier: A Social and Political Portrait. Glencoe: Free
Press.
[20] Kennedy, G. (1974). The Military in the Third World. Charles Scribner’s Sons.
[21] Leon, G.J. (2009). Soldiers or Bureaucrats? Con‡ict and the Military’s Role in Politics, mimeo,
LSE.
[22] Londregan, John and Keith Poole (1990). Poverty, The Coup Trap, and the Seizure of Executive
[23] Loveman, B. (1999). For la Patria - Politics and The Armed Forces in Latin America, Scholarly
[24] Luttwak, E. (1968). Coup d’Etat - A Practical Handbook. The Penguin Press, London.
[25] Luckham, Robin (1974). The Nigerian Military: A Sociological Analysis of Authority and Revolt
[26] Norden, D.L. (2001), Paths to Power in Venezuela, in (ed.) Pion-Berlin, D. (2001). Civil-Military
Relations in Latin-America, University of North Carolina Press, Chapel Hill and London
[27] Nordlinger, E.A. (1977), Soldiers in Politics - Military Coups and Governments. Prentice Hall,
New Jersey
30
[28] Phillip, G.D.E. (1978). The Rise and Fall of the Peruvian Military radicals: 1968-1976. Athlone
Press, London
[29] Phillip, G.D.E. (1985). The Military in South American Politics, Croom Helm, Beckenham (UK)
[31] Ploch, L. (2009) Madagascar’s 2009 Political Crisis, Congressional Research Service, Washington.
[32] Shifter, M. (2004). Breakdown in the Andes, Foreign A¤ airs, September/October 2004.
[33] Stepan, A. (1971). The military in politics: Changing Patterns in Brazil. Princeton University
Press, Princeton.
[34] Sutter, D. (2003) A Game-Theoretic Model Of The CouP D’État, Economics & Politics, Volume
[35] Trinkunas, H.A. (2001). Creating Civilian Control in Argentina and Venezuela, in (ed.) Pion-
31
9 Appendix
Proof of proposition 1
E [Uf (1; L; L; Qi )] = ,
Uf (0; L; L; 1) = 1, Uf (0; L; L; 1) = + 1.
E [Uf (1; R; L; Qi )] = 0,
Uf (0; R; L; 1) = 1, Uf (0; R; L; 1) = 1.
Therefore both types of followers support a coup attempt by L i¤ Qg = 1. Anticipating this, the L
leader has:
Ui (1; L; L; 1) = Ui (1; L; R; 1) = ,
Ui (0; L; R; 1) = Ui (0; L; L; 1) = 1,
Ui (0; L; R; 1) = Ui (0; L; L; 1) = + 1.
Ui (1; R; L; 1) = Ui (1; R; R; 1) = ,
Ui (0; R; R; 1) = Ui (0; R; L; 1) = 1,
Ui (0; R; R; 1) = Ui (0; R; L; 1) = 1.
Proof of proposition 2
See main text for a proof of the existence of the pooling equilibrium. Note that this equilibrium holds for any
out-of-equilibrium belief. There is also a pure strategy separating equilibrium in which only the R type stages a
coup attempt and only R type followers support coups, and there can also a mixed strategy equilibrium. These
equilibria are all pareto dominated by the pooling equilibrium from the perspective of the military leaders, as in
the pooling equilibrium bad governments are always ousted by both L and R leaders, whereas in the separating
equilibrium only R leaders will attempt a coup if Qg = 1, which only succeeds if they are supported by R
followers. It is clear that both L and R leaders are worse of in this equilibrium. The same reasoning can be
applied to prove that mixed strategy equilibria are dominated. Also, the mixed strategy equilibria break down
for lower values of than the pooling equilibrium, as they imply a probability of a coup leader being L which
is smaller than .
32
Proof of proposition 3
If Qg = 1, only R followers will support a coup. Ancipating these reactions, the optimal strategy of the leader
can be derived.
Proof of proposition 5
1
> 1 implies that the L follower never supports any coup attempt in equilibrium. For a pooling equilibrium,
1
this result follows from the proof of proposition 3. We can rule out separating equilibria as well for > 1 . If
the L follower only supports a coup with some probability as to make the L initiator indi¤erent for s = R(L),
then the R initiator will always stage a coup for s = R(L) (as the R follower has a stronger incentive to stage
a coup) and the expected utility of supporting a coup for an L follower will be even lower than in the case
where both types always stage a coup. If the L follower tries to make R indi¤erent at s = R; L, then L will
no longer stage coups at this signal as L’s outside option is better than R0 s. This implies that the expected
pay-o¤ for the L follower from supporting a coup is lower than in the case where both types always stage a
coup.
Knowing that L followers will never support a coup and R followers will support any coup, I derive the
optimal strategy for the leaders. I focus on the R leader …rst. Write Ui (a; Ii ; s; Qg ) and Uf (b; If ; Ii ; Qg ja =
33
(1 )( +1)
E [Ui (1; R; L; 1)] > U i (0; R; L; 1) , < P +(1 )( +1)
~ L (P )
P
E [Ui (1; R; R; 1)] > U i (0; R; R; 1) , > P +(1 )( +1)
~ R (P )
(1 )( 1)
E [Ui (1; R; L; 1)] > Ui (0; R; L; 1) , > P +(1 )( 1) L (P )
P
E [Ui (1; R; R; 1)] > U i;R (0; R; R; 1) , > P +(1 )( 1) R (P )
(1 ) (1 ) (1 ) +(1 )
p
1 2
De…ne P ,P and P (1 ) 1 if > 1, P 0 if < 1.
¯
It can be shown that P< P < P for > 1. The following results can be derived:
¯
1
P < P) ~ R < R < 2 < L < ~L
¯
1
P >P ) L < ~L < 2 < ~R < R
1
P< P < P ) L < ~R < 2 < ~L < R
¯
1
P < P < P > P 2 ) ~R < L < 2 < R < ~L.
Similar thresholds can be derived for the L leader, for the case in which the L follower does not support
(1 )
E [Ui (1; L; L; 1)] > U i (0; L; L; 1) , < P +(1 )
^ L (P )
P
E [Ui (1; L; R; 1)] > U i (0; L; R; 1) , > P +(1 )
^ R (P )
(1 )
De…ne P^ = , it can now be shown that:
A number of results restricting the set of possible parameter combinations can be derived by comparing
P > P^ ; 8
^R > R , > 2; 8P
34
p
^R > ~L , P > 1
(1 ) +1 P•
p ...
^R > L , P > real( 1 (1 ) 1; 0) P
...
P^ < P , >2
1
I will now derive the optimal strategies for each player in all parameter ranges for maxf2;g < .2 8
1
...
Using the parameter restrictions derived above, it can be shown that this parameter range has: P^ < P <
...
2. P < P and max(^ L ; ^ R ) < < L < ~L:
L always stages a coup i¤ incumbent is bad and s = R. R stages a coup if incumbent is bad or if s = R.
(B )
...
5. P < P < P , max( L; R) < < min(^ R ; ~ L ):
7. > max f~ L ; ^ R g ; 8P
E(Q) = + (1 )( (1 ))
28 I do not consider the parameter range 1 1 < < 2 in the main section. This range P< P^ < P < P .
¯
Perfect accountability is no longer achievable. 1 1 < implies that there can be no equilibrium in which the
L follower supports a coup attempt. The optimum will have both R and L leaders staging coups if and only
if the government is bad. This requires P< P < P^ and < minf^ L ; R ; ~ L g. The last result rules out the
following parameter range: ^ R < ~ L < R¯ .
35
8. P• < P , max(~ L ; R) < < ^R:
It can be seen easily that the following ranking must hold between these equilibria if we focus on expected
government quality E(Q): A > A > A , B > B ,B >D ,C >B >D ,C >E >F .
E(QjA) = (1 (1 )(1 )) + (1 )( )
E(QjB ) = (1 (1 )(1 ) ) + (1 )( (1 ) (1 ))
dE(QjB )
d = (1 )( )
E(QjD ) = + (1 )( (1 ))
Comparing (A) and (B ), we …nd that for = 1, both equilibria are equivalent. This is intuitive, as
under perfect information coups will only take place if the leader observes that the follower is of type R and
all coups that are attempted will succeed. Therefore, an equilibrium in which L always stages a coup if the
government is bad and R always stages a coup becomes equivalent in terms of the outcome to an equilibrium
in which L always stages a coup if the government is bad and the followers are R and the R leader always
stages a coup if the followers are R. As soon as the precision of the signal becomes weaker in equilibrium
(B ), there are two e¤ects. First, the weaker signal decreases the number of coups that the R leader attempts
against the good government. On the other hand, the weaker signal also decreases the number coups that the
L leader stages against the bad government. If it is optimal in equilibrium (B ) to have low signal precision
dE(Q)
(because d < 0 ), then we can look at the extreme case of = 21 , even though it is inconsistent with this
equilibrium. It can be veri…ed that equilibrium (C ) is preferred to the hypothetical equilibrium (B ) with
1
= 2 if and only if > . As a consequence, the same will hold for the best possible equilibrium (B ). It
(B )>(A)() <
(C )>(A)() <
36
This implies that either (C ) or (A) are optimal. (C ) is optimal for > , (A) is optimal for > .
Proof of proposition 7
1
First consider < 1 . It follows from the analysis in the main text that pure strategy equilibria can only
take the form of pooling equilibria, in which I^i (R) = I^i (L) = ! where ! 2 fL; R; N g and appropriate out-
of-equilibrium beliefs are speci…ed. Of these equilibria, ! = R fails to meet the intuitive criterion (Cho, Kreps,
1987). Suppose we have an equilibrium with I^i (R) = I^i (L) = R , supported by the out-of-equilibrium belief
of the L follower Pr(Ii = L j I^i = L) = 0. Then L receives a strictly higher utility from announcing L if the
belief would be be Pr(Ii = L j I^i = L) = 1, as L does no longer pay cost . R would not receive higher utility
from announcing L under the same belief Pr(Ii = L j I^i = L) = 1, as his coup attempt is supported anyway
1
in the original equilibrium and he has to pay a cost for lying. Now consider > 1 . In this parameter
range, it is never optimal for the L follower to support a coup attempt against a bad leader if both L and R
types stage such coups in equilibrium. Suppose we have an equilibrium with I^i (R) = I^i (L) = R. Then L
can save by deviating to announcing L. The out-of-equilibrium beliefs of the L follower cannot prevent L
from taking such an action, as the L follower does not support a coup against the bad leader in equilibrium.
The same holds for R in an equilibrium where I^i (R) = I^i (L) = L. Therefore, the only pure strategy pooling
equilibrium has I^i (R) = I^i (L) = N . Finally, the equilibria described in proposition 7 need to be supported
by appropriate out-of-equilibrium beliefs. The belief that the probability of the deviator being L is a supports
all equilibria.
37