How we measure geopolitical risk

All four indices on the GPR Data Hub measure the same thing, the intensity of geopolitical risk in the news, with one design: screen news coverage for geopolitical-risk content, express it as a share of all coverage, and normalize it to a reference period. They differ in which part of the news record they read, because each region is best observed through different sources.

1. Definition

The indices adopt the definition of Caldara and Iacoviello (2022): geopolitical risk is the threat, realization and escalation of adverse events associated with wars, terrorism and tensions among states and political actors that affect the peaceful course of international relations. As in their framework, the African, European and MENA indices separate threats (anticipation and escalation) from acts (realization).

2. Data sources

Every index is built from the GDELT Project, an open database that monitors news worldwide. Two parts of GDELT are used:

3. Identifying geopolitical-risk content

Event-based screen (Africa, Europe)

GDELT's Event Database codes each reported event with the CAMEO taxonomy, which groups actions into four “quad classes”: verbal cooperation, material cooperation, verbal conflict and material conflict. Geopolitical-risk events are those in the two conflict classes:

• Verbal conflict (QuadClass 3) = threats: demands, disapprovals, rejections, threats, ultimatums, the reduction of relations.
• Material conflict (QuadClass 4) = acts: protests and blockades, coercion, assaults, fighting and mass violence.

Cooperative events (QuadClass 1 and 2) are excluded from the numerator but stay in the denominator, because they are part of a country's total news coverage.

Theme-based screen (MENA, South America)

In the Global Knowledge Graph, each article carries machine-assigned themes. An article counts as geopolitical risk if its themes match a fixed family: armed conflict and war, terrorism, military and security forces, rebellion, insurgency and coups, sieges, blockades and ceasefires, assassination, border tensions, weapons of mass destruction and nuclear matters, and sanctions. Because GDELT translates its theme vocabulary into English whatever the article's language, one screen applies identically across Arabic, Spanish, Portuguese, French and other languages.

Large-language-model check (MENA)

For the MENA index, a second track has a large language model read each Arabic article in its original language and score whether it is centrally about current geopolitical risk, adapting the approach of Iacoviello and Tong (2026). The transparent theme screen and the LLM reading correlate above 0.95 at daily frequency. Because they agree, the published headline uses the theme screen, which anyone can audit and reproduce at no cost.

4. From screened news to an index

1. Geographic assignment. In the Event Database, an event is attributed to the country where it took place (the action-location country). In the Knowledge Graph, an article is attributed to every country among the locations it mentions.
2. Article weighting. Events are weighted by the number of articles that report them, so episodes with wide coverage count more than those mentioned in passing.
3. Share, not count. The daily value is geopolitical-risk coverage divided by total coverage. This removes the long-run growth in the number of news sources and measures intensity rather than volume.
4. Normalization. Each series is scaled so its reference-period average equals 100. For country indices, each country is scaled to its own average, so 150 means risk coverage 50% above that country's normal. To compare levels across countries, use the coverage-share files.
5. Smoothing. Charts show 30-day moving averages, and monthly averages for long histories; the CSV files carry the raw daily values as well.

5. Decompositions

• By country: every country in each index (all countries).
• By region: UN subregions of Africa; Western, Nordic, Southern, Central-Eastern, Balkan and Eastern Europe; Southern Cone, Andean, Brazil and the Guianas.
• Threats vs acts: for Africa and Europe, verbal vs material conflict with a kinetic ratio; for MENA, threat vs act theme categories (military build-ups, nuclear, sanctions and ultimatums vs armed conflict, terror and assassinations).
• MENA only: country pairs, actors (armed groups and leaders), event types, city hotspots and oil-and-energy corridors (Persian Gulf and Hormuz, Red Sea and Bab-el-Mandeb, Suez, Eastern Mediterranean, Caspian).

6. Validation

• Known episodes. The indices rise at the right places and times: the Rwandan crisis, the 2011 Libyan war, the Sahel coups, the Tigray war and the Sudanese conflict in Africa; the 2022 invasion of Ukraine in Europe; the Gaza war and Israel–Iran confrontations in MENA.
• Benchmarks. The MENA index co-moves with the GPR and AI-GPR indices in monthly changes (correlation 0.61 with each) while differing in levels, and its oil component tracks the Middle East oil-risk index of Iacoviello at 0.82.
• Markets. In daily local projections, an Arabic oil-risk shock raises the realized volatility of Brent crude for about a month.

7. Limitations

• Early years are thinner. The Event Database before the 2000s rests mostly on English-language sources, so early daily data are noisy. Use monthly averages for historical work.
• Machine coding. GDELT's event codes, geocoding and translations are automated and contain errors; the share design limits, but does not remove, their effect.
• Small media markets. Countries with little local press (for example Guyana and Suriname) have noisier series.
• Coverage is not incidence. The indices measure how intensely the news reports conflict and threat, not casualties or events on the ground. For event counts, use conflict-event data such as ACLED alongside them.

8. Updates, versions and replication

The indices are updated monthly. Each refresh re-runs the full pipeline, so recent values can be revised slightly as late-arriving news is added. Keyword lists, CAMEO codes and country definitions are listed in the appendices of the working papers. Dataset DOI for the MENA index: 10.5281/zenodo.20813840.

How to cite

@techreport{iddouch2026mena,
  author      = {Iddouch, Karim},
  title       = {The Arabic AI-GPR Index: Measuring Geopolitical Risk from Arabic-Language News},
  institution = {International University of Casablanca},
  type        = {Working paper},
  year        = {2026},
  doi         = {10.5281/zenodo.20813840},
  url         = {https://karimiddouch.org/papers/mena-gpr-index.html}
}

@techreport{iddouch2026africa,
  author      = {Iddouch, Karim},
  title       = {The African Geopolitical Risk Index: A Daily, Country-Level Measure for Fifty-Four Countries, 1979--2026},
  institution = {International University of Casablanca},
  type        = {Working paper},
  year        = {2026},
  url         = {https://ssrn.com/abstract=7002698}
}

Please also cite the GDELT Project: Leetaru, K., & Schrodt, P. A. (2013). GDELT: Global data on events, location, and tone, 1979–2012. ISA Annual Convention.