News sensitivity by asset class.
Which markets react to news on which timescale. 33 instruments, 1,079 days, GDELT 2.0.
"News moves markets" is one of those statements that's true and useless. The useful version is: which news moves which markets, and on what horizon. We ran the numbers.
Universe: 33 instruments across seven asset classes — broad equities, sector ETFs, commodities, FX, rates, credit, crypto. News signal: GDELT 2.0 daily tone and volume across nine topics (general tone, war, inflation, central-bank action, stocks, oil, and so on). Window: 1,079 trading days, January 2023 through early 2026. For each instrument we regressed forward returns at horizons of 1, 1, 5, 21, and 63 days on the contemporaneous news features and pulled the joint R².
Equities are the fastest reactor. Broad ETFs peak same-day at 3.26% R². Sectors stay news-driven longer because oil, banking, and healthcare themes recur in headlines for days.
Credit lags equities by a day. HY and IG bond ETFs peak next-day, which matches the known slower price discovery in those markets.
Commodities and FX peak around a week. Physical markets move slower than financial ones. JPY had its highest R² at one week; energy ETFs almost the same.
Rates respond same-day and then forget. The reaction decays sharply from day one to day five.
Crypto is the outlier. R² stays under 0.5% across every horizon — by a wide margin the least news-driven asset class on the page. Crypto isn't responding to news in the broad sense; it responds to its own narrative cycle (ETF flows, regulation, BTC-specific events) that the general index doesn't capture.
3% R² means 97% of return variance is something else. The matrix doesn't tell you the trade. It tells you where to look for one.
The reason to publish the matrix is that the shape of it is the actual finding. News isn't one feature you bolt onto a model. It's a different feature at every horizon and for every asset class. A model that treats it as a single signal is using the wrong tool.