QGI tracks geopolitical risk in every country
QGI builds its own geopolitical-risk reading for 215 countries from the global news record, and checks it against the published academic benchmark wherever one exists. Across the 44 countries that benchmark covers, our reading tracks the same rises and falls in risk.
What it is
Geopolitical risk is the drumbeat of war, unrest, and threats in the news. An influential academic index, the Caldara–Iacoviello Geopolitical Risk index, measures it by counting risk language across a set of major newspapers — but only for 44 countries. QGI rebuilds that idea on the open global news record (the GDELT Global Knowledge Graph), reading hundreds of sources in every language, so it can produce a reading for 215 countries rather than a few dozen.
The question we put to it was simple: when the published benchmark says risk is rising or falling in a country, does QGI’s independent reading move the same way? For most of the benchmark’s countries, it does. That makes QGI-GPR a confirmation layer — a second, wider-angle reading you can set beside any structural or news signal to ask whether the broader risk picture agrees.
What it does not do
QGI-GPR is not a copy of the academic index. That index is built from ten hand-curated newspapers; ours reads machine-coded news from hundreds of global sources. That difference in source material sets a ceiling on how closely the two can match in the absolute levelof risk — a ceiling we tested and confirmed three independent ways. What carries across that boundary is movement: when risk rises, QGI-GPR rises with it. Think of it as a compass, not a ruler — it points the right way and catches the sharp turns, rather than reproducing one institution’s exact number.
How we tested it
Every country with a published benchmark was scored against it, on the part that matters for a confirmation layer: direction and timing.
Movement, not level
We judged QGI-GPR on whether its month-to-month changes rise and fall in step with the benchmark’s, after stripping out the slow shared trend that flatters any comparison. Across the 44 benchmark countries, 84% cleared the bar.
Catching the spikes
We checked how many of each country’s sharpest risk jumps QGI-GPR also flagged as elevated. It caught 61% of them — about 2.4 times what chance alone would produce, the result that matters most for situational awareness.
Tested against itself, three ways
The headline finding held when we widened the risk vocabulary, and a purpose-built attempt to close the remaining gap did no better. The level ceiling is real and external — a property of the source material, not a bug we can tune away.
Honest by country
Not every country reads equally well. Each one carries a signal-quality tier, below, so a weaker reading is always labelled as weaker rather than quietly mixed in with the strong ones.
Signal quality, country by country
For the 44countries with a published benchmark, here is how well QGI-GPR tracks it, graded into three tiers. Each line traces the shape of that country’s risk movement across the period, on its own scale.
Each line is scaled to that country’s own range to show the shape of movement, not the level of risk; heights cannot be compared across countries. It ends at the latest reading.
Strong signal
16 countriesOur movements closely track the published benchmark.
Confirmed signal
21 countriesClear, reliable agreement on direction and spikes.
Weak signal
7 countriesOur reading and the benchmark move in the same direction here, but the agreement is weak. We label these indicative rather than confirmed.
For the remaining 171 countries the same method applies and a reading is produced; external validation is pending wherever no published benchmark exists to check against.
How we measure it
The two checks, in plain English
Movement match. We rank each month by how much risk changed from the month before, for both QGI-GPR and the benchmark, then measure how well the two rankings agree. Comparing changes rather than levels removes the slow shared trend that would otherwise make any two risk series look similar.
Spike catching.We take each country’s biggest risk jumps on the benchmark and check what fraction QGI-GPR also marked as elevated that month. A score of one in four would be chance; we catch far more than that.
Validated against the published benchmark · history through 2026 · generated 2026-06-14.