Escalation Probability estimates the likelihood that a crisis will intensify to a more severe phase within the next 7–30 days. It synthesises military movement signals, diplomatic failure indicators, economic pressure points, and historical escalation patterns to produce a single probability score between 0 and 1.
The indicator draws on multiple event streams: troop mobilisation reports, airspace closures, naval repositioning, leadership rhetoric shifts, and the breakdown of backchannel communications. Each stream is weighted by its historical predictive power in analogous crises.
A logistic model combines six sub-signals: (1) event velocity — how quickly new military or diplomatic events are arriving; (2) actor concentration — whether a small number of actors are driving most activity; (3) rhetoric escalation — NLP sentiment shift in official statements; (4) economic stress — sanctions tightening, currency moves, trade disruption; (5) third-party involvement — new nations or organisations entering the crisis; (6) scenario trigger proximity — how close confirmed events are to known escalation thresholds.
Sub-signals are normalised 0–1 and combined with learned weights. The output is calibrated against a held-out dataset of 48 historical escalation events since 1990 to ensure the probability is meaningful rather than arbitrary.
High (above 0.7): Multiple convergent signals suggest escalation is probable. Historical analogue crises at this level escalated in roughly 65–70% of cases. Treat as a strong caution signal but not a certainty — diplomatic interventions can reverse course even at late stages.
Low (below 0.3): Signals are quiescent or actively de-escalatory. Ceasefire negotiations, troop withdrawals, or sustained diplomatic engagement typically push this indicator low. Sustained low readings over two or more weeks suggest a genuine cooling-off period.
Inputs come from AI-extracted events (RSS feeds, wire services), scenario confirmation data, and market stress proxies. The indicator is recomputed every six hours as new events arrive. Confidence in the output depends on data volume: low-signal days (fewer than 10 new events) widen the implicit uncertainty band significantly.
One email a week: the biggest scenario-probability moves, new cross-crisis connections, and what to watch next. Free, unsubscribe anytime.