10
States Monitored
Sub-Saharan Africa
3
Collapsed States
↑ 1 since last quarter
3
Failing States
HMM State 2
82%
Mean Posterior Risk
↑ 6% from prior quarter
0
Stable (State 0)
No monitored states
4
Fragile (State 1)
Elevated risk threshold
3
Failing (State 2)
Active institutional breakdown
3
Collapsed (State 3)
No functioning government
State Distribution — 10 Monitored Sovereigns
State Definitions (Whitmore Ch.5 — HMM)
STATE 0 — STABLE Functioning institutions, rule of law, GDP growth positive, no active civil conflict
STATE 1 — FRAGILE Declining governance, economic stress, elevated localized violence, weak rule of law
STATE 2 — FAILING Active civil conflict, institutional breakdown, humanitarian crisis, mass displacement
STATE 3 — COLLAPSED No functioning central government, widespread atrocities, full territorial loss
STATE 1 — FRAGILE Declining governance, economic stress, elevated localized violence, weak rule of law
STATE 2 — FAILING Active civil conflict, institutional breakdown, humanitarian crisis, mass displacement
STATE 3 — COLLAPSED No functioning central government, widespread atrocities, full territorial loss
Bayesian Posterior State Probabilities
30-Day HMM State Transition Matrix — Dirichlet Posterior
Bayesian posterior transition probabilities: P(next state | current state). Prior = Dirichlet(α). High diagonal = regime persistence. Based on Whitmore (2026) Ch.5 — HMM inference with Gibbs sampling.
| FROM \ TO | STATE 0 STABLE | STATE 1 FRAGILE | STATE 2 FAILING | STATE 3 COLLAPSED | Persistence |
|---|---|---|---|---|---|
| STATE 0 — STABLE | 0.921 | 0.065 | 0.013 | 0.001 | 92.1% sticky |
| STATE 1 — FRAGILE | 0.038 | 0.831 | 0.118 | 0.013 | 83.1% sticky |
| STATE 2 — FAILING | 0.008 | 0.082 | 0.808 | 0.102 | 80.8% sticky |
| STATE 3 — COLLAPSED | 0.000 | 0.009 | 0.048 | 0.943 | 94.3% sticky |
Spectral gap = 0.074 (dominant eigenvalue = 1.0, second eigenvalue = 0.926). High persistence confirms regimes are slow-moving, justifying 30-day forecast windows. Once a state collapses (State 3), 94.3% probability of remaining collapsed — consistent with empirical fragile states literature.
98%
Highest 30d B-VaR
Sudan — Collapsed
±4%
Mean CI Width
95% credible intervals
82%
Portfolio Mean VaR
Across 10 states
Bayesian
VaR Methodology
Whitmore Ch.6
Bayesian VaR — P(Instability Escalation | Data) — 30/60/90 Day Windows
| Sovereign | HMM State | Prior | Posterior | 30d B-VaR | 95% Credible Interval | 60d | 90d |
|---|
Bayesian VaR computed as P(escalation | data) = ∫ P(escalation | θ) p(θ | data) dθ. Prior = Dirichlet(α) on HMM transition rows. Posterior sampled via MCMC (10,000 draws). Credible intervals show 95% posterior mass. Unlike classical VaR, this framework admits parameter uncertainty and quantifies confidence in the estimate — Whitmore (2026), Chapter 6.
20
Total Variables
Whitmore feature set
6
Political Variables
Governance collapse
5
Economic Variables
Fiscal fragility
9
Security + Social + Ext.
Conflict drivers
Posterior Feature Importance — Sovereign Instability Model — Bayesian RF Variable Analysis
+18%
Mean Belief Update
Prior → Posterior shift
3
Largest Upward Shifts
Sudan, Burkina, Niger
1
Downward Shift
Ethiopia — improving
Bayes
Update Method
Posterior ∝ L × Prior
Bayesian Belief Update — Prior → Posterior Instability Probabilities
Prior encodes historical base rate of instability (World Bank Fragile States Index). Posterior updates via Bayes' theorem: p(θ | D) ∝ L(D | θ) p(θ). Likelihood constructed from ACLED conflict events, IMF economic indicators, V-Dem governance data. Large upward shifts indicate new evidence strongly contradicts prior assumptions. — Whitmore (2026), Ch.1