Methodology
How the index is built
MTI converts multilingual regional reporting into a single, repeatable threat signal for the Middle East and North Africa. The pipeline is deterministic where it matters and auditable end to end.
{{ p.num }}
{{ p.title }}
{{ p.body }}
Category weights
Each headline is assigned one canonical category; each category maps to a fixed severity weight, ordered from open conflict down to de-escalation. The ordering follows event-data severity scales used in conflict research (Goldstein 1992; CAMEO). Weights never change between runs.
{{ w.label }}
{{ w.weight }}
Scoring formula
// per-country score from mean event weight
PerCountry = 1 + 9 · ( 1 − e−avg(w)/5 · 1.2 )
// composite is a weighted average across tracked countries
Composite = weighted_avg( PerCountry across 24 tracked countries )
avg(w) is the recency-weighted, credibility-weighted mean event weight, stabilised toward the country’s recent baseline. The score reflects severity, not how many times a story is filed.
Operational thresholds
STABLE
1.0 – 4.0
Routine reporting; no region-wide escalation.
ELEVATED
4.0 – 7.0
Sustained tension or instability warranting watch.
CRITICAL
7.0 – 10.0
Active conflict or security rupture with regional consequences.
Confidence
Each country and the regional composite carry a confidence between 0 and 1. It rises with the effective number of events, the diversity of sources reporting them, and corroboration across independent outlets; it falls when coverage is thin, stale, or single-sourced. Confidence is shown alongside every reading so a well-evidenced score is not read the same as a tentative one.
Sources
Reporting is drawn from regional outlets in their original language, plus aggregated news indices. Coverage is reviewed as feeds change.
Methodology basis
MTI follows established practice for news-based risk indices (Caldara & Iacoviello, Measuring Geopolitical Risk, AER 2022; Baker, Bloom & Davis, Measuring Economic Policy Uncertainty, QJE 2016) and for composite indicators (OECD/EC-JRC Handbook on Constructing Composite Indicators, 2008). Event-severity weighting draws on conflict event-data scales (Goldstein 1992; CAMEO); the forecast uses exponential-smoothing methods (Gardner & McKenzie 1985); market reactions use the event-study market model (MacKinlay 1997) with false-discovery control (Benjamini & Hochberg 1995). Full references are in the project’s METHODOLOGY_REVIEW.md.
Limitations
The index reflects the volume and severity of published reporting, not ground truth. Source coverage is uneven across languages, machine categorisation introduces noise, and runs that fail validation are withheld rather than published. Categorisation of non-English headlines depends on the optional language model; without it, local-language coverage contributes little to scores. Market correlations are short-sample, unstable, and indicate association, not causation. MTI is a monitoring aid, not an intelligence assessment.