Warconomy publishes forecasts as dated, immutable records with an explicit horizon, stated assumptions, named invalidation conditions, and a resolution source that decides the outcome. Probabilities are published only as coarse bands (very unlikely through very likely), each with a stated numeric meaning, because the reference class of comparable past events is usually too small to justify decimal precision. Every forecast names the model that produced it and links to that model's public card. Once issued, a forecast is never edited: a changed view becomes a new record and the original stays published so it can be scored honestly.
- Simplest defensible method wins — Warconomy does not deploy a complex model to look sophisticated.
- Every model must beat a simple naive baseline (no-change, no-revision, or persistence) to be worth running.
- Uncertainty is labeled by kind: a plausible range is not a statistical confidence interval, and is never presented as one.
- Forecasts are decision-support, not investment, legal, or policy advice.
What the probability bands mean
Warconomy publishes probability only in five coarse bands. A figure like “63.27%” would imply a calibration sample that does not exist for most conflict-economic events, so it is not published.
| Band | Numeric meaning |
|---|---|
| very unlikely | roughly 5-20% |
| unlikely | roughly 20-40% |
| roughly even | roughly 40-60% |
| likely | roughly 60-80% |
| very likely | roughly 80-95% |
Models in production
Each model has a public card documenting its inputs, assumptions, uncertainty construction, validation approach, known weaknesses, and the forecasts it produced.
Trend continuation (rule-based) v1.0
Takes the recent pace of change in an official series and keeps it going, assuming nothing dramatic happens. Simple, transparent, and easy to check — but it will miss sudden shocks.
Uses analyst judgment. Updated monthly.
Technical detail
Assumptions
- The most recently observed rate of change persists over the forecast horizon.
- No discrete shock (a major offensive, a ceasefire, a policy reversal) materially breaks the trend within the horizon.
- The publishing institution keeps its measurement definition unchanged over the horizon.
Known weaknesses
- Cannot anticipate discrete shocks — precisely the events that matter most in conflict economics.
- Uncertainty widens rapidly with horizon length; beyond roughly two publication cycles the interval becomes too wide to be decision-useful.
- No seasonality adjustment; series with strong seasonal structure need a different method.
- Interval width is set by analyst judgment anchored to historical revision size, not derived from a fitted error distribution, because most of these series have too few comparable published points to fit one.
Uncertainty: A plausible-range bracket, widened for longer horizons, reflecting the compounding risk that the constant-rate assumption breaks down. Not a statistical confidence interval — labeled as plausible-range throughout.
Validation: Compared against a no-change naive baseline on historical published points where two or more comparable rounds exist. See /forecast-track-record for the hindcast demonstrations.
Policy sequencing (institutional-process) v1.0
Looks at how an institution like the EU Council actually passes measures — who has to agree, how long similar steps took before — to judge whether a proposed measure lands by a given date.
Uses analyst judgment. Updated per event.
Technical detail
Assumptions
- The institution follows its stated procedural sequence (proposal, negotiation, adoption) rather than acting outside it.
- Historical timing between an equivalent proposal and adoption is informative about the current one.
- Publicly-stated intent by decision-makers is meaningful evidence, though not binding.
Known weaknesses
- Unanimity- or consensus-based bodies can be blocked indefinitely by a single holdout, which precedent timing systematically under-weights.
- Announced target dates routinely slip without any public correction.
- Relies on reported negotiating positions that are often second-hand and contested.
Uncertainty: Expressed as a coarse probability band only (very-unlikely through very-likely), never a decimal percentage, because the reference class of comparable past decisions is small.
Validation: Scored on resolution against the institution's own published legal act (its official journal or press release), and compared against a base-rate baseline derived from how often comparable proposals were adopted within a similar window.
External institutional forecast tracking v1.0
Watches whether a body like the EIA will move its own published forecast up or down at its next release, based on what has happened since it last locked its data.
Uses analyst judgment. Updated monthly.
Technical detail
Assumptions
- The institution publishes on its announced schedule.
- Its published forecast reflects information available up to its own stated cutoff — so evidence emerging after that cutoff is genuine new information about the likely revision direction.
Known weaknesses
- Predicting a forecaster's revision is not the same as predicting the underlying variable, and must never be presented as such.
- Institutions occasionally change methodology between releases, which can move a published number for reasons unrelated to the world changing.
- Highly sensitive to events landing in the gap between the institution's cutoff and its publication date.
Uncertainty: Direction-of-revision expressed as a coarse probability band; any accompanying level range is a plausible-range bracket around the institution's own published central value, never Warconomy's independent price model.
Validation: Resolved directly against the institution's next published release. Baseline: 'no revision' (assume the institution repeats its prior figure).
Threshold persistence v1.0
Asks whether something that is currently disrupted will still be disrupted by a certain date, on the reasoning that shipping and insurance decisions unwind slowly.
Uses analyst judgment. Updated monthly.
Technical detail
Assumptions
- The structural drivers of the current disruption (security risk, insurer pricing, contracted routing decisions) unwind more slowly than the headline news cycle suggests.
- Commercial operators change routing on multi-month planning cycles, not in immediate response to single events.
Known weaknesses
- Systematically biased toward persistence — it will be late to call a genuine normalization.
- Threshold choice is a judgment call; a differently-drawn threshold could flip the answer.
- Depends on the resolution source continuing to publish a comparable measure.
Uncertainty: Coarse probability band reflecting how far the current value sits from the threshold relative to its recent volatility.
Validation: Resolved against the named publishing authority's own figure at the resolution date. Baseline: 'persistence' (assume today's state simply continues), which this model must beat to be worth running.
How forecasts are scored
Every forecast names, at issue time, the exact criteria and the published source that will decide it. When that source publishes, the outcome is attached to the original record — the prediction fields themselves are never edited. Where a forecast cannot be scored (the source stops publishing, or changes definition so no comparable figure exists), it is marked unresolvable with a stated reason and excluded from scoring rather than quietly dropped.
Each model is measured against a simple naive baseline — no-change, no-revision, or persistence. A model that cannot beat its baseline is not worth running, and Warconomy will say so rather than present it as skillful. See the track record, which is honest from launch: it reports openly that no live forecast has resolved yet.
Related
Responsible forecasting policy · Observed vs. modeled values · Active forecasts · Track record