The AI-GPR Index

A daily measure of geopolitical risk constructed using large language models, covering 1960 to present.

Matteo Iacoviello  ·  Jonathan Tong

Last updated: March 31, 2026 with data through March 31, 2026.

About the Index

The AI-GPR Index measures geopolitical risk by reading newspaper articles with artificial intelligence. Building on the original GPR Index (Caldara and Iacoviello, 2022), it replaces keyword matching with semantic understanding: rather than searching for specific word combinations, the AI reads each article and assigns a risk score based on its geopolitical content.

The index is constructed at daily frequency using articles from the New York Times, Washington Post, and Chicago Tribune, spanning 1960 through 2026.

AI-GPR

The main index. Aggregates LLM-assigned geopolitical risk scores across all newspaper articles on each date, normalized by total newspaper article count. Captures the overall level of geopolitical risk as perceived in the news.

GPR Original (Keyword-Based)

A keyword-based GPR index constructed using the Caldara and Iacoviello (2022) proximity search methodology on the same three newspapers. Counts articles matching specific combinations of geopolitical and risk-related terms. Extended back to 1960.

GPR Oil Disruptions GPR

A sub-index identifying articles that discuss oil or energy supply disruptions driven by geopolitical events. Constructed using a second LLM classification layer applied to high-GPR articles containing oil-related keywords.

Methodology

Articles matching a broad keyword filter are scored by GPT-4o-mini on a 0–1 scale reflecting the intensity of geopolitical risk content. All indices are normalized to a mean of 100 over 1985–2019. See the paper for full details.

Interactive Chart

Use the range slider below the chart to zoom into specific periods. Hover over the chart for exact values. Double-click to reset zoom.

Country GPR Index (top 120 countries)
All
Initiator
Respondent
Spillover

Note: The AI-GPR Country Index uses machine learning to identify specific countries and roles (Initiator, Respondent, Spillover) within news text associated with geopolitical risk. Smoothing reflects a trailing moving average. Country-level data are available through March 2026. Scaling uses the same monthly factors as the overall AI-GPR index (mean=100, 1985-2019), so that values are comparable to the aggregate index.

GPR by Event Type

Note: Each series sums the AI-GPR sentiment score for articles classified under that Monthly GPR by event type.

Bilateral GPR Index (top 200 country-pairs)

Note: The bilateral index measures geopolitical risk in articles where both countries appear together as initiator and respondent (in either direction). Top pairs are ranked by total cumulative sentiment across the full sample. Scaled using the same monthly factors of the overall AI-GPR index (mean = 100, 1985–2019), so that values are comparable across pairs and to the aggregate index. Bilateral data are available through March 2026.

Downloads

Main Index

File Description Download
ai_gpr_data_daily.csv Daily AI-GPR index, 1960–present ↓ CSV
ai_gpr_data_monthly.csv Monthly AI-GPR index and components, 1960–present ↓ CSV
AI_GPR_PAPER.pdf Working paper describing the methodology ↓ PDF

Country Decompositions

File Description Download
ai_gpr_country_monthly.csv Monthly GPR by country (initiator / respondent / all roles) ↓ CSV
ai_gpr_eventtype_monthly.csv Monthly GPR by event type ↓ CSV
ai_gpr_bilateral_monthly.csv Monthly bilateral GPR for top country pairs (directed: initiator → respondent) ↓ CSV

Citation

If you use the AI-GPR Index, please let us know, and cite:

Iacoviello, Matteo and Jonathan Tong (2026). “The AI-GPR Index: Measuring Geopolitical Risk using Artificial Intelligence.” Working Paper, Federal Reserve Board of Governors.
Download Paper (PDF)

Contact: matteo.iacoviello@frb.gov  ·  jtong45@wisc.edu  ·  www.matteoiacoviello.com