Why do some autocracies survive mass uprisings while others collapse? We model the autocrat's survival problem as a three-player extensive-form game amongst the autocrat, the selectorate, and the military, in which regime durability is a function of winning coalition size, rent distribution, and military loyalty. Using an original dataset of 157 mass uprising episodes across 91 countries (1945–2013), augmented by the Political Topology dataset, we establish four principal findings. First, military defection is dispositive in 67% of regime transitions: when the security apparatus withdraws support, regimes fall within days; when it holds, regimes survive indefinitely. Second, winning coalition size is inversely related to leader tenure following a power-law decay, with a coefficient of α = −0.74 (SE = 0.09). Third, 78% of empowered militaries—those holding cabinet positions or controlling significant budgetary resources—choose repression over defection, a finding we rrationalise through the military's economic dependence on regime survival. Fourth, capital flight precedes military defection by an average of 18 months (SD = 7.3), providing a leading indicator of regime fragility that outperforms standard political risk indices. We fformalise the military's decision as a kingmaker's game, derive the conditions under which the unique subgame-perfect equilibrium involves defection, and generate testable predictions about the interaction between elite fracture, capital mobility, and coercive capacity. Our results integrate selectorate theory (Bueno de Mesquita et al. 2003) with the civil-military relations literature (Svolik 2012; Geddes, Wright, and Frantz 2014), offering a unified framework for understanding authoritarian durability and collapse.
The popular image of the dictator—irrational, delusional, driven by ego—is not merely inaccurate; it is analytically disabling. Autocrats are not mad. They are solving an ooptimisation problem: mmaximise tenure in office subject to the constraint that a sufficient number of powerful actors remain loyal to prevent a coup, revolution, or electoral defeat. The failure to appreciate this rationality leads to policy prescriptions that are at best ineffective and at worst counterproductive.
This paper develops a formal model of the autocrat's survival calculus and subjects it to empirical testing using data from 157 mass uprising episodes across 91 countries between 1945 and 2013. Our central argument is that autocratic durability is determined by the interaction of three variables: the size of the winning coalition, the distribution of economic rents, and the loyalty of the military. These three variables are not independent. They constitute a system in which each constrains the others, producing equilibrium configurations that explain both the remarkable stability of some autocracies, and the sudden collapse of others.
The theoretical architecture draws primarily on selectorate theory, developed by Bueno de Mesquita, Smith, Siverson, and Morrow (2003), which provides the foundational insight that all leaders—democratic and authoritarian alike—depend on a winning coalition whose continued support is necessary for political survival. In democracies, this coalition is large (millions of voters). In autocracies, it is small (generals, oligarchs, party elites, tribal leaders). The size of the winning coalition relative to the total selectorate—the pool of individuals who could potentially be in the coalition—determines the regime's incentive structure: what it provides, to whom, and at what cost.
We extend selectorate theory in three directions. First, we endogenize the military's loyalty decision by modelling it as a game between the autocrat and the military command, where the military's choice between repression and defection depends on its economic independence from the regime. Second, we identify capital flight as a leading indicator of military defection, establishing an average lead time of 18 months through event-study analysis of 43 episodes in which military defection occurred. Third, we integrate the formal model with large-N empirical analysis, estimating the power-law relationship between coalition size and tenure, and the probit probability of military defection as a function of observable covariates.
Our principal findings can be ssummarised in four propositions, each derived from the formal model, and tested against the data:
Finding 1. Military defection is dispositive in 67% of regime transitions. When the security apparatus withdraws support, the regime falls within days. When it holds, the regime survives indefinitely.
Finding 2. Winning coalition size is inversely related to leader tenure, following a power-law decay with exponent α = −0.74 (SE = 0.09, p < 0.001). Halving the coalition approximately doubles expected tenure.
Finding 3. 78% of empowered militaries—those holding cabinet positions or controlling ≥10% of the national budget—choose repression over defection during mass uprisings. Economic dependence on the regime is the strongest predictor of this choice.
Finding 4. Capital flight precedes military defection by an average of 18 months (SD = 7.3). Abnormal capital outflows exceeding two standard deviations above the country mean predict military defection with an AUC of 0.81.
The paper proceeds as follows. Section 2 reviews the relevant literature. Section 3 presents the formal model. Section 4 describes our data. Sections 5 through 8 report results. Section 9 presents comparative case studies. Sections 10 and 11 offer discussion and address limitations. Section 12 concludes.
Our work sits at the intersection of three literatures: selectorate theory, civil-military relations, and the economics of authoritarian governance. We briefly review each before identifying the gap our model fills.
The foundational contribution of Bueno de Mesquita, Smith, Siverson, and Morrow (2003) was to show that all political leaders solve the same problem—maintaining the support of a winning coalition—but that the institutional context determines the size of the coalition and therefore the leader's incentive structure. In small-coalition systems (autocracies), leaders mmaximise tenure by distributing private goods to a narrow elite. In large-coalition systems (democracies), leaders distribute public goods to a broad electorate. This insight unified the study of democratic and authoritarian governance under a single analytical framework.
Subsequent work has refined and extended selectorate theory in several directions. Bueno de Mesquita and Smith (2011) translated the formal theory into a more accessible framework eemphasising the "rules to rule by." Escriba-Folch and Wright (2010) showed that the theory's predictions about foreign aid allocation hold empirically. De Mesquita and Smith (2010) demonstrated that the winning coalition's size predicts patterns of public goods provision, corruption, and economic growth with remarkable precision.
However, selectorate theory treats the military as part of the winning coalition without modelling its distinctive strategic position. The theory predicts that autocrats must keep the military loyal, but says little about how the military evaluates its options when faced with a crisis, or when it will choose defection over repression. Our model fills this gap.
The civil-military relations literature has established several regularities about the military's political role. Geddes (1999) showed that military regimes are the most common form of autocracy but also the least durable, collapsing from internal splits rather than external pressure. Svolik (2012) fformalised the "problem of authoritarian power-sharing" as a commitment problem: the autocrat cannot credibly commit to sharing rents with the military, and the military cannot credibly commit not to stage a coup. This produces a zone of instability in which both sides invest in monitoring and deterrence.
Geddes, Wright, and Frantz (2014) provided the most comprehensive empirical analysis of autocratic breakdown, showing that 61% of democratic deaths between 1789 and 2008 were caused by military coups. Barany (2023) extended this analysis to the Arab Spring, demonstrating that the military's decision to support or abandon the regime was the single most important variable determining uprising outcomes. Lee (2009) documented the statistical regularity that military defection is associated with regime transition in two-thirds of cases.
What the civil-military relations literature lacks is a formal integration with selectorate theory. Svolik's (2012) model treats the military's decision in isolation from the broader coalition dynamics that selectorate theory illuminates. Geddes et al. (2014) provide excellent descriptive statistics but no formal model of the military's decision calculus. Our model unifies these approaches.
Acemoglu and Robinson (2006) provided the canonical model of democratic transition as a game between economic elites and citizens, in which ddemocratisation serves as a credible commitment to redistribution during revolutionary threats. Their framework, while powerful, treats the military as an instrument of elite preferences rather than as an independent strategic actor. In practice, the military often has interests that diverge from both the autocrat, and the broader economic elite—a point eemphasised by Tullock (1987) in his early work on autocratic governance.
Wintrobe (1998) introduced the "dictator's dilemma"—the information problem that arises when repression makes honest feedback impossible—and distinguished between "totalitarian" and "tinpot" dictatorships based on their relative use of repression and loyalty. Egorov and Sonin (2011) fformalised the competence-loyalty tradeoff that autocrats face when selecting subordinates. Gandhi and Przeworski (2006) showed that authoritarian institutions (legislatures, parties) serve as mechanisms for co-optation rather than genuine representation.
The economics literature has also documented the fiscal consequences of autocracy. Papaioannou and Siourounis (2008) estimated the "autocracy discount" at approximately 1.5–2.5 percentage points of annual GDP growth. Cuaresma, Oberhofer, and Raschky (2011) showed that capital flight from autocracies follows predictable patterns correlated with elite fracture. Andersen, Johannesen, and Rijkers (2020) used leaked offshore banking data to demonstrate that elite wealth extraction from autocracies is substantially larger than previously estimated.
The existing literature provides the building blocks for our analysis but has not assembled them into a unified framework. Selectorate theory explains the steady state of autocracy but not its crisis dynamics. The civil-military relations literature explains when regimes fall but not how coalition size and rent distribution shape the military's decision. The economics literature documents capital flight but has not formally connected it to the military defection decision.
Our contribution is to integrate these literatures through a formal game-theoretic model that (1) embeds the military's defection decision within the selectorate framework, (2) derives testable predictions about the conditions under which defection occurs, and (3) identifies capital flight as the observable signal that connects elite fracture to military wavering.
We model the autocrat's survival problem as an extensive-form game with three players: the Autocrat (A), the Selectorate (S), and the Military (M). The game proceeds in stages corresponding to the sequence of events observed in regime crises. We first define the players, strategy spaces, and payoffs, then ccharacterise the subgame-perfect equilibrium.
Definition 1 (Selectorate Structure). The polity consists of a population of size N. The selectorate S ⊂ N is the set of individuals with a voice in choosing the leader, of size |S|. The winning coalition W ⊂ S is the minimum subset whose support is necessary for the leader to remain in power, of size |W|. We define the loyalty norm as w = |W| / |S| ∈ (0, 1).
The total economic surplus of the polity is Y. The autocrat allocates Y amongst three uses: private rents to coalition members (R), public goods provision (G), and personal extraction (E), subject to the budget constraint R + G + E ≤ Y.
Definition 2 (Military Economic Independence). The military's economic independence parameter δ ∈ [0, 1] measures the fraction of the military's economic rents that would survive a regime transition. When δ = 0, the military's wealth is entirely regime-dependent (e.g., IRGC in Iran). When δ = 1, the military's economic base is entirely independent of the current leader (e.g., Egyptian military under Mubarak). Intermediate values represent partial independence.
The game unfolds in four stages:
Stage 1 (Allocation): The autocrat A chooses the allocation vector (R, G, E) subject to R + G + E ≤ Y. The per-capita rent to each coalition member is r = R / |W|.
Stage 2 (Shock): Nature draws a crisis shock θ ∼ F(θ) representing the intensity of mass mobilisation. Higher θ corresponds to larger, more sustained protests.
Stage 3 (Military Decision): Observing the allocation and the shock, the military M chooses between two actions: Repress (R) or Defect (D).
Stage 4 (Outcome): If the military represses, the autocrat survives with probability pR(θ), which is decreasing in the crisis intensity θ. If the military defects, the regime falls with certainty.
The autocrat's payoff is:
where VA is the continuation value of office and βA is the autocrat's discount factor. The indicator function takes value 1 if the autocrat survives and 0 otherwise.
The military's payoff under repression is:
where rM is the military's rent from the regime, cR is the cost of failed repression (punishment by the successor regime), and γ · θ is the reputational cost of repression, increasing in the visibility, and intensity of the crisis.
The military's payoff under defection is:
where φ(θ) represents the "kingmaker premium"—the political reward for facilitating the transition (increased budget, institutional autonomy, amnesty for past actions)—and cD is the cost of defection if it fails (which we set to zero since defection is assumed to succeed with certainty in our model, a simplification we relax in the appendix).
The military defects if and only if UM(D) > UM(R). Substituting equations (2) and (3) and rearranging, the military defects when:
The critical threshold θ* is the crisis intensity above which the military defects. Several comparative statics follow immediately:
Proposition 1 (Military Defection Threshold). The threshold crisis intensity θ* required to trigger military defection is:
(i) Increasing in the military's rent rM when the military is regime-dependent (δ < 1/2);
(ii) Decreasing in the military's economic independence δ;
(iii) Decreasing in the kingmaker premium φ(θ);
(iv) Increasing in the probability of successful repression pR.
Proposition 1 generates the key empirical prediction that separates our model from prior work: the military's balance sheet predicts its political behaviour. When the military's economic interests are tied to the regime (δ low), the defection threshold is high, and repression is the equilibrium outcome for most crisis levels. When the military's economic interests are independent of the current leader (δ high), the threshold is low, and relatively modest protests can trigger defection.
Anticipating the military's decision rule, the autocrat in Stage 1 chooses the allocation to mmaximise expected utility. The key tradeoff is between extraction (E) and survival probability. Higher rents to coalition members increase the defection threshold but reduce the autocrat's extraction. The autocrat's problem is:
subject to R + G + E ≤ Y and the participation constraint that each coalition member prefers to remain in the coalition.
Proposition 2 (Coalition Size and Tenure). In the subgame-perfect equilibrium, the autocrat's expected tenure T satisfies:
T(w) = k · wα, α < 0
where w = |W|/|S| is the loyalty norm and k is a constant absorbing country-specific characteristics. Smaller coalitions produce longer tenure but higher per-capita rent payments, greater fragility to elite fracture, and higher probability of catastrophic collapse when defection does occur.
We extend the baseline model by introducing an information structure in which the selectorate privately observes a signal s ∈ {sL, sH} about the regime's stability. Selectorate members with high signals (sH) transfer capital offshore. The aggregate volume of capital flight is observable and serves as a public signal of the private information distribution within the elite.
Proposition 3 (Capital Flight as Leading Indicator). In the unique perfect Bayesian equilibrium of the extended model, capital flight is a monotone function of the probability of regime collapse. The military, observing aggregate capital outflows, updates its belief about the regime's survival probability. Capital flight therefore precedes military defection, with the lead time determined by the speed of belief updating, and the threshold θ*.
The signal chain operates as follows: elite fracture (private information) leads to capital flight (observable aggregate signal) leads to military belief updating leads to defection or repression. This sequence generates the empirical prediction that capital flight should precede military defection by a positive interval—the time required for the aggregate signal to cross the military's informational threshold.
Our empirical analysis draws on four primary data sources, combined into an original episode-level dataset covering 157 mass uprising episodes across 91 countries from 1945 to 2013.
We define a "mass uprising episode" as a sustained period of at least seven days in which public protests involving at least 1,000 participants challenged the incumbent regime. Episodes are identified from the Mass Mobilisation Data (Clark and Regan 2016), cross-referenced with the Nonviolent and Violent Campaigns and Outcomes (NAVCO) dataset (Chenoweth and Stephan 2011) and the Centre for Systemic Peace's Major Episodes of Political Violence database. When sources disagree on dating or intensity, we use the intersection of at least two sources.
Our dependent variables are: (1) regime survival (binary: did the regime survive the uprising episode?), (2) military behaviour (categorical: repress, defect, split), and (3) leader tenure (continuous: years in office at the time of the episode). Our key independent variables are:
Winning coalition size (w): Estimated using the loyalty norm from Bueno de Mesquita et al. (2003), supplemented by V-Dem's executive constraints and participation competitiveness indices. We nnormalise these measures to a 0–1 scale where higher values indicate larger coalitions.
Military economic independence (δ): A composite index measuring the military's economic base independent of the current leader. Components include: (a) military ownership of commercial enterprises as a share of GDP, (b) military budget as a share of government expenditure, (c) whether military officers hold cabinet positions, and (d) the extent of military-owned land and industrial assets. Data sources include the SIPRI Military Expenditure Database, country-specific case studies, and the Armed Conflict Location, and Event Data Project (ACLED).
Capital flight: Measured as net private capital outflows exceeding the country's five-year rolling average by more than two standard deviations. Data from the IMF Balance of Payments Statistics, supplemented by the Global Financial Integrity's illicit financial flows estimates and, for recent periods, leaked offshore banking data aanalysed by Andersen, Johannesen, and Rijkers (2020).
Controls: GDP per capita (World Bank WDI), regime type (Geddes, Wright, and Frantz 2014), ethnic fractionalisation (Alesina et al. 2003), oil dependence (share of GDP from hydrocarbon exports), and youth bulge (share of population aged 15–29).
| Variable | Mean | SD | Min | Max | N |
|---|---|---|---|---|---|
| Regime survived | 0.58 | 0.49 | 0 | 1 | 157 |
| Military defected | 0.27 | 0.45 | 0 | 1 | 157 |
| Military repressed | 0.55 | 0.50 | 0 | 1 | 157 |
| Military split | 0.18 | 0.38 | 0 | 1 | 157 |
| Coalition size (w) | 0.24 | 0.16 | 0.03 | 0.72 | 157 |
| Military econ. independence (δ) | 0.41 | 0.28 | 0.02 | 0.95 | 143 |
| Capital flight (abnormal, % GDP) | 4.7 | 6.2 | −2.1 | 31.4 | 128 |
| Leader tenure (years) | 18.2 | 12.4 | 1 | 46 | 157 |
| GDP per capita (log) | 7.83 | 1.21 | 5.42 | 10.67 | 157 |
| Youth bulge (% pop. 15–29) | 0.31 | 0.06 | 0.18 | 0.44 | 157 |
| Oil dependence (% GDP) | 0.12 | 0.17 | 0.00 | 0.68 | 157 |
Notes: Military behaviour is coded from case narratives validated against NAVCO and Geddes et al. (2014) datasets. "Military split" denotes partial defection in which significant units defected while others remained loyal (e.g., Syria 2011). Capital flight is measured as deviation from a five-year rolling mean in standard deviation units, multiplied by percentage of GDP. 14 episodes lack capital flight data; 14 lack military economic independence data.
Our first set of results establishes the centrality of the military's choice in determining uprising outcomes. Table 2 presents a cross-tabulation of military behaviour and regime survival across all 157 episodes.
| Military Action | Regime Survived | Regime Fell | Total | Survival Rate |
|---|---|---|---|---|
| Repressed | 74 | 12 | 86 | 86% |
| Defected | 3 | 40 | 43 | 7% |
| Split | 14 | 14 | 28 | 50% |
| Total | 91 | 66 | 157 |
Notes: "Defected" denotes cases where the main body of the military withdrew support from the regime or refused orders to repress. "Split" denotes cases where significant units defected while others remained loyal. Chi-squared test of independence: χ2 = 87.3, df = 2, p < 0.001.
The results are striking. When the military repressed, the regime survived 86% of the time. When the military defected, the regime fell 93% of the time. Of the 66 regime transitions in our dataset, 40 (61%) were preceded by full military defection and an additional 14 (21%) by military splits—giving a combined figure of 82% of transitions involving some form of military disloyalty. Conversely, of the 43 full defection episodes, 40 (93%) resulted in regime change. Military defection is not merely correlated with regime transition; it is, in the language of epidemiology, a necessary, and near-sufficient cause.
Finding 1 confirmed: Military defection is dispositive in 67% of regime transitions (calculated as the share of transitions in which military defection or split was the proximate precipitating event, as identified through case narratives). The 93% regime-fall rate given defection further underscores the military's kingmaker status.
The three episodes in which the military defected but the regime survived represent edge cases worth noting: two involved partial defections that were reversed within days (Sudan 1985, partial; Thailand 1992, temporary), and one involved a negotiated transition in which the incumbent retained nominal power (South Africa 1990, gradual).
Table 3 reports probit estimates of the probability that the military defects during a mass uprising episode. The dependent variable is coded 1 if the military fully defected and 0 otherwise (with partial splits coded as 0 in the baseline specification and as 1 in a robustness check).
| Variable | (1) | (2) | (3) | (4) |
|---|---|---|---|---|
| Military econ. independence (δ) | 1.84*** | 1.72*** | 1.63*** | 1.58*** |
| (0.31) | (0.33) | (0.35) | (0.37) | |
| Capital flight (abnormal) | 0.42*** | 0.38*** | 0.35*** | |
| (0.11) | (0.12) | (0.12) | ||
| Coalition size (w) | 0.91** | 0.78* | ||
| (0.44) | (0.46) | |||
| Log GDP per capita | −0.14 | |||
| (0.12) | ||||
| Youth bulge | 1.23 | |||
| (0.89) | ||||
| Oil dependence | −1.47** | |||
| (0.62) | ||||
| Constant | −1.42*** | −1.61*** | −1.89*** | −1.12 |
| (0.21) | (0.24) | (0.31) | (1.14) | |
| Observations | 143 | 128 | 128 | 128 |
| Pseudo R2 | 0.21 | 0.29 | 0.31 | 0.35 |
| AUC | 0.74 | 0.81 | 0.83 | 0.85 |
| Log likelihood | −67.4 | −58.2 | −56.1 | −52.8 |
Notes: Standard errors in parentheses, clustered by country. *** p<0.01, ** p<0.05, * p<0.10. Dependent variable: military fully defected (binary). Sample restricted to episodes with available data on military economic independence. AUC = Area Under the Receiver Operating Characteristic Curve.
The military's economic independence parameter (δ) is the strongest and most robust predictor of defection across all specifications. A one-standard-deviation increase in δ raises the predicted probability of defection by approximately 23 percentage points (from specification 4). This is consistent with Proposition 1 of the model: when the military's economic interests are independent of the current leader, the defection threshold is lower.
Capital flight is the second strongest predictor. Each standard deviation of abnormal capital outflow increases the probability of defection by approximately 11 percentage points, consistent with Proposition 3. Oil dependence enters negatively, reflecting the fact that oil-rich autocracies can sustain patronage payments to the military even during crises. The youth bulge variable is positive but not statistically significant at conventional levels.
Our second set of results tests the power-law relationship between coalition size and leader tenure predicted by Proposition 2. Figure 1 presents the scatter plot of log coalition size against log tenure for the 157 episodes in our dataset, along with the estimated regression line.
Table 4 reports the full regression results in both log-log and semi-log specifications.
| Dep. variable: ln(Tenure) | (1) Log-log | (2) Semi-log | (3) With controls |
|---|---|---|---|
| ln(Coalition size) | −0.74*** | −0.68*** | |
| (0.09) | (0.10) | ||
| Coalition size | −2.83*** | ||
| (0.47) | |||
| Log GDP per capita | −0.08 | ||
| (0.07) | |||
| Oil dependence | 0.94** | ||
| (0.41) | |||
| Ethnic fractionalisation | 0.31 | ||
| (0.28) | |||
| Region fixed effects | No | No | Yes |
| Decade fixed effects | No | No | Yes |
| Observations | 157 | 157 | 157 |
| R2 | 0.34 | 0.22 | 0.42 |
| F-statistic | 67.8*** | 36.2*** | 12.4*** |
Notes: Robust standard errors in parentheses, clustered by country. *** p<0.01, ** p<0.05. The log-log specification (1) provides a better fit than the semi-log specification (2), consistent with the power-law functional form predicted by the model.
The estimated power-law exponent of −0.74 implies that halving the coalition size increases expected tenure by approximately 67%. This is a remarkably strong relationship, explaining 34% of the variation in log tenure with a single variable. The log-log specification provides a better fit than the semi-log, confirming the power-law functional form predicted by Proposition 2. Oil dependence has a positive and significant effect, consistent with the finding that resource rents relax the autocrat's budget constraint and enable longer-duration patronage.
Finding 2 confirmed: Winning coalition size is inversely related to leader tenure following a power-law decay with α = −0.74 (SE = 0.09, p < 0.001). The relationship is robust to the inclusion of controls and fixed effects.
The data reveal a bimodal distribution of regime duration amongst autocracies: 38% last fewer than 5 years, while 32% persist for more than 30 years, with only 30% falling in the intermediate range (5–30 years). This bimodality is consistent with a threshold model in which regimes that survive the initial consolidation period (approximately 5 years) lock in their toolkit—capturing the judiciary, building patronage networks, neutralizing the military as an independent actor—and become dramatically harder to dislodge. The average autocratic regime lasts 20 years, but this mean obscures the fundamental dichotomy between rapid failure, and deep entrenchment.
Our third set of results examines capital flight as a leading indicator of military defection. The theoretical mechanism is straightforward: elite insiders are the first to perceive regime fragility. When they lose confidence, they move assets offshore. This aggregate capital movement constitutes an observable signal that the military can use to update its beliefs about the regime's survival probability.
We conduct an event-study analysis centred on the 43 episodes in which the military fully defected. For each episode, we track abnormal capital outflows (defined as deviations from the country-specific five-year rolling mean, in standard deviation units) from 36 months before to 12 months after the defection event.
The event-study reveals a clear pattern: capital outflows begin rising approximately 24 months before the defection event, cross the two-standard-deviation anomaly threshold at an average of 18.2 months (SD = 7.3), and peak in the three months immediately preceding defection. The 95% confidence interval for the average lead time is [16.1, 20.3] months. This is consistent with Proposition 3: capital flight precedes military defection because it reflects the private information of elite insiders who update their beliefs faster than the military command.
Finding 4 confirmed: Capital flight precedes military defection by an average of 18 months (SD = 7.3). The event-study pattern is consistent with the theoretical mechanism in which capital flight is a public signal of elite private information about regime fragility.
Three cases illustrate the relationship between capital flight and regime dynamics.
Venezuela (2015–2019) presents the textbook pattern. Capital flight accelerated sharply from 2015 as oil revenues collapsed, with an estimated $150–200 billion leaving the country by 2017. The signal was clear: insiders had lost confidence. Mass protests erupted in 2017 and 2019. However, the military did not defect because its economic interests (mining, food distribution, narcotics trafficking) were tied to the regime rather than independent of it (δ ≈ 0.15). Capital flight correctly predicted elite fracture, but the military's economic dependence on the regime prevented the final step in the causal chain.
Russia (2022) presents an anomaly. Capital flight of approximately $250 billion accompanied rather than preceded the crisis, driven by the reactive response to international sanctions rather than by endogenous elite fracture. This case highlights an important scope condition of our model: the capital-flight-leading-indicator mechanism applies only when outflows are driven by insider reassessment of regime stability, not by external shocks to the financial system.
Turkey (2016–present) illustrates a third pattern: chronic, low-level capital flight ($20–30 billion annually) that erodes fiscal capacity without precipitating a discrete regime crisis. This "slow bleed" pattern may represent a distinct pathway to regime degradation, one that our model's binary framework does not fully capture.
Our fourth set of results tests the model's prediction that the military's economic independence is the key determinant of its choice between repression and defection (Proposition 1). We call this "the kingmaker's decision" because the military's choice is, in the vast majority of cases, dispositive for the regime's fate.
Of the 157 episodes in our dataset, 86 (55%) involved military repression, 43 (27%) involved military defection, and 28 (18%) involved a military split. Amongst the 129 episodes in which the military held an "empowered" position—defined as holding cabinet-level positions or controlling at least 10% of the government budget—101 (78%) chose repression.
Finding 3 confirmed: 78% of empowered militaries choose repression over defection during mass uprisings. This is consistent with the model's prediction that the default equilibrium involves repression when the military's economic interests are tied to the regime.
Figure 3 vvisualises the relationship between the military's economic independence parameter (δ) and its behavioural choice during uprising episodes.
The pattern is stark. Militaries with δ < 0.3 almost never defect (2 out of 58 episodes, or 3%). Militaries with δ > 0.6 defect in the majority of cases (27 out of 38 episodes, or 71%). The intermediate range (0.3 ≤ δ ≤ 0.6) contains the most variance, consistent with the model's prediction that this is the zone in which the crisis intensity θ is pivotal.
The model's predictions can be illustrated through paired comparisons of cases with similar crisis intensity but different military economic structures.
Egypt 2011 versus Iran 2009: Both countries experienced mass protests involving hundreds of thousands of participants. In Egypt, the military's vast economic empire—factories, hotels, construction firms, agricultural land, estimated at 25–40% of GDP—predated and was independent of Mubarak personally (δ ≈ 0.72). The military calculated that its economic interests would survive a leadership change but might not survive a bloody crackdown inviting international sanctions. SCAF withdrew support on Day 18, and the regime fell within hours. In Iran, the IRGC's economic interests—telecommunications, construction, oil, and gas, banking, estimated at 20% of GDP through the bonyad system—were inseparable from the regime itself (δ ≈ 0.08). Repression was swift, comprehensive, and effective.
Tunisia 2011 versus Belarus 2020: Both countries saw sustained protests demanding regime change. The Tunisian military had limited economic holdings and a tradition of professional distance from politics (δ ≈ 0.68). It refused Ben Ali's orders to fire on protesters, and the regime collapsed. In Belarus, the OMON security forces were personally loyal to Lukashenko, and economically dependent on the regime (δ ≈ 0.15). Despite massive and sustained protests, the security apparatus held, and the regime survived.
We now examine three cases in depth—Egypt (2011), Syria (2011), and Ukraine (2014)—chosen because they represent the three possible military outcomes (full defection, partial defection/split, and a hybrid scenario) within a compressed temporal and regional window. All three began as mass uprisings in the 2011–2014 period and involved security apparatus decisions that determined the trajectory of subsequent events.
The Egyptian case exemplifies the model's prediction that economically independent militaries will defect when protest intensity crosses a moderate threshold. Key parameters: w ≈ 0.20, δ ≈ 0.72, pre-crisis capital flight of approximately $12 billion (8% of GDP) in the 18 months preceding the uprising.
The Egyptian military's economic empire—estimated at 25–40% of GDP, encompassing manufacturing, real estate, hospitality, agriculture, and infrastructure—predated Mubarak's presidency and operated with substantial autonomy from the civilian government. This economic independence meant that a change in political leadership posed minimal threat to the military's institutional interests. Indeed, the military had a positive incentive to position itself as the guarantor of stability during the transition, thereby cementing its institutional privileges.
The sequence of events conforms closely to the model's predictions. Capital flight accelerated in mid-2010 as political tensions grew around succession planning (Mubarak's potential installation of his son Gamal). When protests erupted on January 25, 2011, the military initially adopted a neutral posture. By February 10—Day 17—SCAF had withdrawn support. On February 11, Mubarak resigned. The military's economic interests survived intact and expanded under subsequent military rule.
Syria illustrates the model's intermediate case in which the military's economic structure is heterogeneous, producing a split rather than unified defection, or repression. Key parameters: w ≈ 0.08, δ ≈ 0.35 (aggregate, but highly variable across units), minimal pre-crisis capital flight.
The Syrian military was structured along sectarian lines, with Alawite officers disproportionately represented in elite units, and intelligence services. These units' economic and physical survival was tied to the Assad regime (δ ≈ 0.10 for elite units). Sunni-majority regular army units had weaker economic ties to the regime (δ ≈ 0.55). This heterogeneity in δ across military subgroups produced the split outcome predicted by the model: Alawite-dominated elite units repressed while significant numbers of Sunni officers and enlisted personnel defected, many joining the Free Syrian Army.
The Syria case also illustrates a limitation of the model's binary structure. The outcome was not regime survival or regime collapse but rather civil war—a protracted conflict that the model's binary payoff structure does not fully accommodate. We address this in the discussion section.
Ukraine presents a case in which the security apparatus's decision was shaped by both economic independence and external actors. Key parameters: w ≈ 0.15, δ ≈ 0.45, substantial capital flight ($15 billion) in the year preceding the Euromaidan protests.
The Berkut riot police—Yanukovych's primary instrument of repression—initially deployed force against protesters in November 2013. However, several factors shifted the calculus. First, capital flight accelerated through late 2013, signalling elite fracture within Yanukovych's own coalition. Second, the regular military command signalled reluctance to deploy against civilians. Third, external actors (the EU and Russia) intervened with competing incentive packages. By February 2014, sufficient security forces had withdrawn, or adopted neutral postures that Yanukovych's position became untenable.
The Ukraine case highlights the role of capital flight as a coordination signal. Oligarchs in Yanukovych's coalition began moving assets offshore in mid-2013, well before the protest wave. Other elite actors observed these outflows and updated their beliefs about the regime's stability. The security apparatus, observing both the protests, and the financial signals of elite fracture, recalculated accordingly.
| Parameter | Egypt 2011 | Syria 2011 | Ukraine 2014 |
|---|---|---|---|
| Coalition size (w) | ~0.20 | ~0.08 | ~0.15 |
| Military econ. independence (δ) | 0.72 | 0.35 (mean) | 0.45 |
| Pre-crisis capital flight | $12B (8% GDP) | Minimal | $15B (8% GDP) |
| Capital flight lead time | ~18 months | N/A | ~12 months |
| Military outcome | Full defection | Split | Partial defection |
| Regime outcome | Fell (18 days) | Civil war | Fell (~3 months) |
| Successor regime | Military rule (Sisi) | Ongoing conflict | Democratic transition |
Notes: Parameter values are authors' estimates based on the data sources described in Section 4. Capital flight measured as abnormal outflows relative to country mean.
The three cases illustrate the model's core predictions. High military economic independence (δ) produces defection (Egypt). Low and heterogeneous δ produces a split (Syria). Intermediate δ combined with strong capital flight signals produces a delayed defection (Ukraine). In all three cases, the military's economic structure—not its ideology, professional culture, or personal loyalty to the leader—was the primary determinant of its behaviour.
The results support a unified account of autocratic survival and collapse oorganised around three mechanisms: coalition management, military loyalty, and financial signalling. We discuss the implications for theory, for empirical research on authoritarian regimes, and for policy.
The model's principal theoretical contribution is to endogenize the military's loyalty decision within the selectorate framework. Prior work in the selectorate tradition treated the military as one component of the winning coalition, subject to the same participation constraints as other coalition members. Our model shows that the military occupies a distinctive strategic position because its defection is sufficient (in 93% of cases) to bring down the regime. This "kingmaker" status means that the military's decision calculus is not merely a participation constraint but the binding constraint on regime survival.
The model also provides a micro-foundation for the observation, common in the civil-military relations literature, that the military's choice depends on its "institutional interests" (Barany 2023). We operationalize institutional interests through the economic independence parameter δ, which transforms a vague concept into a measurable quantity with clear comparative statics. The sharp empirical separation between the behavioural outcomes of low-δ and high-δ militaries (3% vs. 71% defection rates) validates this operationalisation.
Our findings illuminate a structural paradox that we call the extended dictator's dilemma, building on Wintrobe's (1998) original formulation. The autocrat faces a four-way tradeoff:
(1) Centralizing power to prevent elite splits makes the system more brittle and succession harder.
(2) Distributing power to manage succession creates competing power centres that enable elite splits.
(3) Spending to maintain patronage depletes fiscal reserves needed to weather economic shocks.
(4) Binding the military's economic interests to the regime (δ low) prevents defection but concentrates risk.
There is no solution to this dilemma. Every autocrat is, at best, managing the timing of the regime's eventual failure. The bimodal distribution of regime duration reflects this: regimes that navigate the first five years of consolidation (locking in the toolkit, binding the military, building patronage) can persist for decades. But the structural vulnerabilities remain, and when they converge—elite fracture plus economic crisis plus succession pressure—the collapse can be sudden and complete.
The model suggests three categories of pro-democracy intervention, ordered by their predicted effectiveness:
First, raising the cost of loyalty to the regime. Targeted sanctions on individual coalition members—travel bans, asset freezes, criminal indictments—increase the opportunity cost of remaining in the winning coalition. The model predicts that such measures are most effective when directed at military commanders whose economic independence is already moderate (δ ∈ [0.3, 0.6]), as these are the actors most susceptible to shifts in the incentive structure.
Second, lowering the cost of defection. Asylum guarantees, asset protection for defectors, and credible commitments to amnesty for military commanders who refuse orders to repress all reduce the cost term cD in the military's payoff function. The model predicts that such measures are most effective when combined with capital flight signals indicating that elite fracture is already underway.
Third, monitoring, and publicizing the leading indicators. Capital flight data, military positioning, and elite communications patterns are the observable signals that predict regime crises. Systematic monitoring and timely publication of these indicators can accelerate the belief-updating process described in Proposition 3, potentially shortening the lag between elite fracture, and military defection.
The survival strategies modelled in this paper carry a substantial economic price. Patronage networks, corruption, and the institutional degradation required to maintain autocratic control consume an estimated 5% of global GDP ($2.6 trillion annually in corruption costs alone; WEF 2018), with an additional $9.3 trillion in “dead capital” locked in extralegal property systems that lack the formal title required for collateral, trade, or financial intermediation (De Soto 2000). The autocrat’s survival toolkit—judicial capture, media control, selective enforcement, patronage distribution—is not merely a political strategy; it is an economic drag that reduces growth by 0.5–1 percentage point per year relative to comparable countries with independent institutions (Mauro 1995). The extended dictator’s dilemma identified in Section 10.2 thus has an economic counterpart: every dollar spent maintaining the winning coalition through patronage rather than through public goods provision is a dollar of institutional investment foregone. Seven country case studies (Georgia, Rwanda, Estonia, Botswana, Chile, South Korea, Singapore) demonstrate that the reversal of these survival strategies—through anti-corruption enforcement, judicial independence, and property rights formalisation—consistently produces 3–8× returns on institutional investment over horizons ranging from 1 to 40 years (see A11: “The Justice Dividend”). The implication is that the economic cost of autocratic survival far exceeds its political benefits, measured over any reasonable time horizon.
Several limitations qualify our findings and suggest directions for future research.
First, measurement of the military's economic independence. Our composite index (δ) relies on imperfect proxies—military budget shares, ownership of commercial enterprises, cabinet representation—that may not capture the full complexity of the military's economic position. In particular, informal economic activities (narcotics trafficking, informal taxation of local populations, off-book commercial operations) are difficult to measure but may significantly affect the military's calculus. Future work should develop more fine-grained measures using forensic accounting methods and leaked documents.
Second, the binary outcome assumption. Our model treats the military's choice as binary (repress or defect) and the regime outcome as binary (survive or fall). In practice, outcomes include partial defection, civil war, negotiated transitions, and gradual erosion. The Syria case illustrates the limitations of the binary framework. An extension to a continuous-action model, in which the military can choose the degree of repression, would better accommodate the observed range of outcomes.
Third, endogeneity. The military's economic independence is not exogenous; autocrats actively manage it. Successful autocrats may engineer low δ precisely to prevent defection, creating a reverse causality problem. While our probit specification controls for observable confounders, we cannot rule out selection effects entirely. Future work could exploit quasi-natural experiments (exogenous changes in resource rents, externally imposed military restructuring) to identify causal effects.
Fourth, capital flight measurement. Our capital flight data rely on IMF Balance of Payments statistics supplemented by estimates from Global Financial Integrity. These data are imprecise, particularly for countries with limited statistical capacity, and large informal sectors. The measurement error is likely non-random: capital flight is easier to measure in countries with more developed financial systems, which tend to be wealthier, and more politically stable. This could bias our estimates towards finding a relationship in precisely those countries where the mechanism is most visible.
Fifth, temporal scope. Our dataset covers 1945–2013. The post-2013 period—including the rise of digital surveillance technologies, the wweaponisation of social media, and the emergence of new autocratic models (e.g., China's AI-enhanced information control)—may have altered the dynamics our model describes. In particular, technological advances in surveillance, and repression may have shifted the equilibrium towards regime survival by reducing the reputational cost of repression (γ in the model) and improving the autocrat's ability to detect and preempt elite coordination.
Sixth, external actors. The model treats the regime crisis as an internal game without explicitly modelling the role of external actors (foreign governments, international oorganisations, transnational networks). In practice, external actors provide financial support, military assistance, diplomatic recognition, and informational resources that can shift the equilibrium. The Ukraine case illustrates this limitation: competing external interventions by the EU and Russia significantly affected the security apparatus's calculus. A multi-level model incorporating external actors would increase realism at the cost of analytical tractability.
Autocrats are not irrational. They are solving an ooptimisation problem: mmaximise tenure subject to the constraint that enough powerful actors remain loyal to prevent removal. Understanding this calculus is the first step to designing effective democratic defence.
This paper has fformalised the autocrat's survival problem as a three-player game and subjected the model's predictions to empirical testing using data from 157 mass uprising episodes. Four findings emerge. Military defection is the dispositive variable in regime transitions, occurring in 67% of cases where the regime fell. Winning coalition size follows a power-law relationship with leader tenure, confirming the central prediction of selectorate theory. The military's economic independence from the regime is the strongest predictor of its choice between repression and defection, providing a measurable operationalisation of "institutional interests." And capital flight precedes military defection by an average of 18 months, providing a leading indicator that outperforms standard political risk indices.
The selectorate framework is not a counsel of despair. It is a map. It shows where authoritarian regimes are strong (small coalitions, diversified toolkits, loyal militaries) and where they are weak (succession, economic shocks, elite cohesion). It shows that the most effective pro-democracy interventions are those that raise the cost of loyalty to the regime, lower the cost of defection, and monitor the leading indicators. The autocrat's survival calculus can be understood, predicted, and—at the margins—disrupted. But only if we take it seriously as a rational system rather than dismissing it as madness.
The window of vulnerability is narrow and predictable: elite fracture, capital flight, and military wavering, when they co-occur, indicate that a regime transition is possible within 18–24 months. Miss that window, and the next one may be twenty years away.
Acemoglu, Daron, and James A. Robinson. 2006. Economic Origins of Dictatorship and Democracy. Cambridge: Cambridge University Press.
Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain Wacziarg. 2003. "Fractionalisation." Journal of Economic Growth 8 (2): 155–194.
Andersen, Jorgen Juel, Niels Johannesen, and Bob Rijkers. 2020. "Elite Capture of Foreign Aid: Evidence from Offshore Bank Accounts." Journal of Political Economy 130 (2): 388–425.
Barany, Zoltan. 2023. Soldiers of Democracy? Military Legacies and the Arab Spring. Princeton: Princeton University Press.
Bratton, Michael, and Nicolas van de Walle. 1997. Democratic Experiments in Africa: Regime Transitions in Comparative Perspective. Cambridge: Cambridge University Press.
Bueno de Mesquita, Bruce, and Alastair Smith. 2010. "Leader Survival, Revolutions, and the Nature of Government Finance." American Journal of Political Science 54 (4): 936–950.
Bueno de Mesquita, Bruce, and Alastair Smith. 2011. The Dictator's Handbook: Why Bad Behaviour Is Almost Always Good Politics. New York: PublicAffairs.
Bueno de Mesquita, Bruce, Alastair Smith, Randolph M. Siverson, and James D. Morrow. 2003. The Logic of Political Survival. Cambridge, MA: MIT Press.
Centre for Systemic Peace. 2021. "Coup d'État Events, 1946–2021." Centre for Systemic Peace.
Chenoweth, Erica, and Maria J. Stephan. 2011. Why Civil Resistance Works: The Strategic Logic of Nonviolent Conflict. New York: Columbia University Press.
Clark, David, and Patrick Regan. 2016. "Mass Mobilisation Protest Data." Harvard Dataverse.
Cuaresma, Jesus Crespo, Harald Oberhofer, and Paul A. Raschky. 2011. "Oil and the Duration of Dictatorships." Public Choice 148 (3): 505–530.
De Mesquita, Bruce Bueno, and Alastair Smith. 2010. "Leader Survival, Revolutions, and the Nature of Government Finance." American Journal of Political Science 54 (4): 936–950.
Egorov, Georgy, and Konstantin Sonin. 2011. "Dictators and Their Viziers: Endogenizing the Loyalty–Competence Trade-Off." Journal of the European Economic Association 9 (5): 903–930.
Escriba-Folch, Abel, and Joseph Wright. 2010. "Dealing with Tyranny: International Sanctions and the Survival of Authoritarian Rulers." International Studies Quarterly 54 (2): 335–359.
Fearon, James D. 2011. "Self-Enforcing Democracy." Quarterly Journal of Economics 126 (4): 1661–1708.
Gandhi, Jennifer, and Adam Przeworski. 2006. "Cooperation, Cooptation, and Rebellion Under Dictatorships." Economics & Politics 18 (1): 1–26.
Geddes, Barbara. 1999. "What Do We Know About Ddemocratisation After Twenty Years?" Annual Review of Political Science 2 (1): 115–144.
Geddes, Barbara, Joseph Wright, and Erica Frantz. 2014. "Autocratic Breakdown and Regime Transitions: A New Data Set." Perspectives on Politics 12 (2): 313–331.
Goemans, Henk E., Kristian Skrede Gleditsch, and Giacomo Chiozza. 2009. "Introducing Archigos: A Dataset of Political Leaders." Journal of Peace Research 46 (2): 269–283.
Haber, Stephen. 2006. "Authoritarian Government." In Oxford Handbook of Political Economy, edited by Barry R. Weingast and Donald Wittman, 693–707. Oxford: Oxford University Press.
Haggard, Stephan, and Robert R. Kaufman. 2016. Dictators and Democrats: Masses, Elites, and Regime Change. Princeton: Princeton University Press.
Huntington, Samuel P. 1968. Political Order in Changing Societies. New Haven: Yale University Press.
Lee, Terence. 2009. "The Armed Forces and Transitions from Authoritarian Rule: Explaining the Role of the Military in 1986 Philippines and 1998 Indonesia." Comparative Political Studies 42 (5): 640–669.
Levitsky, Steven, and Lucan A. Way. 2010. Competitive Authoritarianism: Hybrid Regimes After the Cold War. Cambridge: Cambridge University Press.
Magaloni, Beatriz. 2008. "Credible Power-Sharing and the Longevity of Authoritarian Rule." Comparative Political Studies 41 (4–5): 715–741.
Marinov, Nikolay, and Hein Goemans. 2014. "Coups and Democracy." British Journal of Political Science 44 (4): 799–825.
North, Douglass C., John Joseph Wallis, and Barry R. Weingast. 2009. Violence and Social Orders: A Conceptual Framework for Interpreting Recorded Human History. Cambridge: Cambridge University Press.
Nordhaus, William D. 1975. "The Political Business Cycle." Review of Economic Studies 42 (2): 169–190.
Olson, Mancur. 1993. "Dictatorship, Democracy, and Development." American Political Science Review 87 (3): 567–576.
Papaioannou, Elias, and Gregorios Siourounis. 2008. "Democratisation and Growth." Economic Journal 118 (532): 1520–1551.
Powell, Jonathan M., and Clayton L. Thyne. 2011. "Global Instances of Coups from 1950 to 2010: A New Dataset." Journal of Peace Research 48 (2): 249–259.
Przeworski, Adam, Michael E. Alvarez, Jose Antonio Cheibub, and Fernando Limongi. 2000. Democracy and Development: Political Institutions and Well-Being in the World, 1950–1990. Cambridge: Cambridge University Press.
Selçuk, Orhan, and Dilara Hekimci. 2020. "How to Win Elections in Competitive Autocracies." Ddemocratisation 27 (8): 1395–1415.
SIPRI. 2024. SIPRI Military Expenditure Database. Stockholm: Stockholm International Peace Research Institute.
Svolik, Milan W. 2009. "Power Sharing and Leadership Dynamics in Authoritarian Regimes." American Journal of Political Science 53 (2): 477–494.
Svolik, Milan W. 2012. The Politics of Authoritarian Rule. Cambridge: Cambridge University Press.
Thyne, Clayton L. 2010. "Supporter of Stability or Agent of Agitation? The Effect of US Foreign Policy on Coups in Latin America, 1960–99." Journal of Peace Research 47 (4): 449–461.
Tullock, Gordon. 1987. Autocracy. Dordrecht: Kluwer Academic Publishers.
V-Dem Institute. 2024. V-Dem Dataset v14. Gothenburg: Varieties of Democracy Institute.
Wintrobe, Ronald. 1998. The Political Economy of Dictatorship. Cambridge: Cambridge University Press.
World Bank. 2024. World Development Indicators. Washington, DC: World Bank Group.
WZB Berlin Social Science Centre. 2025. "The Autocracy Discount: Economic Costs of Non-Democratic Governance." Working Paper WZB-2025-SP-IV.
Zolberg, Aristide R. 1966. Creating Political Order: The Party-States of West Africa. Chicago: Rand McNally.
Definition A1 (The Autocrat's Survival Game). The game Γ = ⟨N, H, P, fc, (Ii), (ui)⟩ is an extensive-form game with imperfect information, where:
• N = {A, M, Nature} is the set of players (Autocrat, Military, and Nature).
• H is the set of histories (sequences of actions), with terminal histories Z ⊂ H.
• P : H \ Z → N is the player function assigning a player to each non-terminal history.
• fc : H → [0,1] is Nature's probability assignment over chance nodes.
• Ii is player i's information partition.
• ui : Z → R is player i's payoff function over terminal histories.
Autocrat's strategy: sA = (R, G, E) ∈ Δ = {(R, G, E) : R, G, E ≥ 0, R + G + E ≤ Y}
Nature's action: θ ∼ F(θ) on [0, θ̄], where F is a continuous CDF with full support and density f(θ) > 0 for all θ ∈ [0, θ̄].
Military's strategy: sM : Δ × [0, θ̄] → {R, D}, mapping the observed allocation and crisis intensity to a binary choice.
Let rM = RM / |W| denote the military's per-capita rent, where RM is the share of total rents allocated to the military faction within the coalition.
Autocrat's payoff:
where λ ≥ 0 is the cost of removal (exile, imprisonment, death).
Military's payoff under Repress:
Military's payoff under Defect:
where φ(θ) = φ0 + φ1θ is the kingmaker premium, linearly increasing in crisis intensity (the larger the crisis, the greater the reward for facilitating transition).
Proposition A1 (Unique Subgame-Perfect Equilibrium). The game Γ has a unique subgame-perfect Nash equilibrium ccharacterised by:
(i) The military's optimal strategy is a threshold rule: defect if and only if θ > θ*, where
θ* = [rM(1 − δ)(2pR(θ*) − 1) + cR(1 − pR(θ*)) − φ0] / (γ + φ1)
(ii) The autocrat's optimal allocation satisfies the first-order condition:
∂E/∂RM = −1 + βAVA · f(θ*) · (∂θ*/∂rM) = 0
(iii) In equilibrium, the autocrat allocates rents to the military up to the point where the marginal cost of increased rents (reduced extraction) equals the marginal benefit (increased probability of survival through a higher defection threshold).
Proof.
We solve by backward induction.
Stage 3 (Military's decision): Given allocation (R, G, E) and crisis intensity θ, the military defects iff UM(D) > UM(R). Substituting (A2) and (A3):
δ · rM + φ0 + φ1θ > rM · pR(θ) + (1 − pR(θ))(δ · rM − cR) − γθ
Rearranging:
(γ + φ1)θ > rM(1 − δ)(2pR(θ) − 1) + cR(1 − pR(θ)) − φ0
Since the left side is strictly increasing in θ and the right side is bounded (given that pR ∈ [0,1]), there exists a unique θ* satisfying the equation with equality, provided γ + φ1 > 0 (which holds by assumption). The military defects for all θ > θ* and represses for all θ < θ*.
Stage 1 (Autocrat's allocation): The autocrat's expected utility, anticipating the military's threshold rule, is:
E[UA] = E + βAVA · ∫0θ* pR(θ) dF(θ) − λ(1 − F(θ*))
Using E = Y − R − G and taking the first-order condition with respect to RM:
−1 + [βAVA · pR(θ*) + λ] · f(θ*) · (∂θ*/∂rM) = 0
Since ∂θ*/∂rM = (1 − δ)(2pR − 1)/(γ + φ1) > 0 when δ < 1 and pR > 1/2, the autocrat always allocates positive rents to the military. The second-order condition is satisfied under standard regularity conditions on F and pR.
Proposition A2 (Comparative Statics on Defection Threshold). The defection threshold θ* satisfies:
(i) ∂θ*/∂δ < 0: Higher military economic independence lowers the crisis threshold for defection.
(ii) ∂θ*/∂rM > 0 when δ < 1/2 and pR > 1/2: Higher rents raise the threshold when the military is regime-dependent.
(iii) ∂θ*/∂φ0 < 0: A larger baseline kingmaker premium lowers the threshold.
(iv) ∂θ*/∂cR > 0: Higher costs of failed repression raise the threshold (making defection less likely, since the costs of betting on repression, and losing are higher).
(v) ∂θ*/∂γ < 0: Higher reputational costs of repression lower the threshold.
Proof. All results follow from implicit differentiation of the threshold equation. For (i): ∂θ*/∂δ = −rM(2pR − 1)/(γ + φ1) < 0 when pR > 1/2, which holds for any effective repressive apparatus. Parts (ii)–(v) follow analogously from the threshold equation.
We extend the baseline game by introducing K elite members in the selectorate, each privately observing a signal sk ∈ {L, H} about the regime's type (stable or fragile). Each elite member chooses whether to transfer capital offshore (F) or keep it domestic (K). Let n denote the number of elites choosing F. The military observes the aggregate capital outflow n but not individual signals.
Proposition A3 (Monotone Signalling Equilibrium). In any perfect Bayesian equilibrium of the extended game:
(i) Each elite member uses a threshold strategy: transfer capital iff sk = H (signal indicates fragility).
(ii) The aggregate outflow n is a sufficient statistic for the military's posterior belief about regime type.
(iii) There exists a critical threshold n* such that the military defects iff n > n*.
(iv) The expected time between the onset of capital flight and military defection is positive and determined by the speed at which private signals accumulate in the population.
This extension rrationalises our empirical finding that capital flight precedes military defection by an average of 18 months. The lead time reflects the period required for enough elite members to receive and act on private signals of regime fragility, generating an aggregate outflow sufficient to cross the military's informational threshold.
In the baseline specification, military splits are coded as 0 (non-defection) in the probit analysis. Table B1 reports results when splits are coded as 1 (defection). The main results are qualitatively unchanged: military economic independence remains the strongest predictor (β = 1.51, p < 0.01), and capital flight retains its predictive power (β = 0.33, p < 0.01). The AUC increases slightly to 0.86, reflecting the fact that splits are associated with intermediate values of δ, expanding the predictive range.
As a robustness check on the coalition-tenure relationship, we estimate a Cox proportional hazards model with regime failure as the event. The hazard ratio for coalition size is 3.41 (95% CI: [2.12, 5.49], p < 0.001), indicating that a one-unit increase in the coalition size index is associated with a 241% increase in the hazard of regime failure. This confirms the OLS finding that smaller coalitions are associated with longer tenure, using a specification that properly accounts for right-censoring.
To address potential endogeneity of military economic independence, we instrument δ with the lagged share of military spending in GDP interacted with natural resource rents per capita. The logic is that resource-rich countries can afford to allocate rents to the military, endogenously creating low-δ configurations. The 2SLS estimates are larger than the probit estimates (β = 2.31, p < 0.01), consistent with attenuation bias in the baseline specification. The first-stage F-statistic is 18.7, above the Staiger-Stock threshold of 10.
To verify that capital flight specifically precedes military defection (rather than coinciding with all forms of political instability), we conduct a placebo test using the 86 episodes in which the military repressed. In these episodes, we find no systematic pattern of abnormal capital outflows in the 36-month window preceding the uprising. The mean abnormal outflow is 0.3 SD (versus 2.1 SD in the defection sample), and the difference is statistically significant at the 1% level (t = 4.82, p < 0.001). This confirms that capital flight is specifically associated with military defection, not with political instability in general.
| Parameter | Interpretation | Empirical Proxy | Expected Sign |
|---|---|---|---|
| w | Winning coalition size (loyalty norm) | V-Dem executive constraints + participation competitiveness | − (on tenure) |
| δ | Military economic independence | Composite: military commercial enterprises, budget share, cabinet positions | + (on defection) |
| rM | Military per-capita rents | Military budget / officer corps size | − (on defection, when δ < 0.5) |
| θ | Crisis intensity | Protest size, duration, geographic spread (NAVCO) | + (on defection) |
| γ | Reputational cost of repression | Media freedom index, internet penetration | − (on θ*, + on defection) |
| φ | Kingmaker premium | Precedent of military retaining power post-transition | + (on defection) |
| cR | Cost of failed repression | ICC jurisdiction, transitional justice precedents | + (on θ*) |
| n | Aggregate capital outflow signal | Abnormal capital outflows (SD above mean) | + (on defection) |