KASPR Datahaus research
Factor internet latency into news-driven trading strategies
Six years of nationwide round-trip-time measurements, matched to every news story and every price move for 1,166 US-listed companies. On days when connection quality is unusually uneven across the country, the same news moves prices measurably less.
Scroll to read ↓
The signal
National internet quality is not constant
KASPR measures round-trip time from many vantage points to millions of routable addresses across US networks. What matters here is not the average speed but the spread: on some days connection quality is uniform across the country, on others it is markedly uneven — some users see normal speeds while others wait.
We call that spread latency dispersion. It moves a lot from one day to the next, and the top quarter of days sits well above the rest.
The top quartile of days is our definition of “high latency”. It covers 25.35% of firm-days in the sample.
Daily value with a 7-day mean. The marker is the sample-wide 75th percentile; days above it are shaded as high-latency.
A daily line of latency dispersion from 2022 with a seven-day average; days in the top quartile, above the 75th-percentile marker, are shaded.The finding
The same news, a smaller price move
Take every firm-day where the news was clearly positive, and compare the same-day return on ordinary days against high-latency days. Then do the same for negative news.
Good news moves a stock 1.81% on a normal day. On a high-latency day it moves it 0.30 percentage points less — about a sixth of the response, gone. Bad news shows a larger proportional gap: a 1.32% drop on a normal day, dampened by 0.63 percentage points when the network is congested.
These are within-firm comparisons. Each estimate absorbs the individual stock, the year, the month and the day of the week, so it is not driven by which companies get news, or by good and bad periods for the market as a whole.
Event-day cumulative return by news polarity and network regime. Bars show the estimate; whiskers are one standard error.
Four bars: for both positive and negative news the high-latency return is closer to zero than the normal-day return.Who it lands on
Small caps and retail-heavy stocks absorb almost all of it
The effect is not spread evenly. Sort firms into five groups by market capitalisation and the attenuation is concentrated almost entirely in the smallest: for the bottom fifth, good news moves the price 1.19 percentage points less on congested days. For the largest fifth the estimate is 0.09 percentage points and indistinguishable from zero.
The same pattern appears when stocks are split by how much of their trading comes from retail investors. Stocks with above-median retail participation show a clear effect; those below the median show nothing at all.
This is the part of the picture that fits the intuition most directly. Institutional desks run co-located servers and dedicated feeds; they are insulated from consumer-grade network conditions. Retail investors are not, and the stocks they concentrate in are exactly where the effect shows up.
Bottom quintile: −1.19pp. Top quintile: −0.09pp, not distinguishable from zero.
By market-cap quintile
By retail participation
Positive-news estimate with its 95% interval. Hollow marks are not statistically distinguishable from zero.
Two coefficient plots for positive news: the effect is large and clear for the smallest firms and high-retail stocks, and near zero elsewhere.Does the price catch up?
Partly — and only after bad news
If congestion simply delays information rather than destroying it, the gap should close once the network clears. It does — but only after bad news.
Following negative news, the gap between congested and normal days narrows from 0.32 percentage points on the day itself to 0.24 percentage points by the third trading day. Following positive news it does the opposite: the high-latency group starts 0.24 percentage points behind and is 0.36 percentage points behind three days later.
Anchored at zero the day before the event. Solid lines are normal days, dashed are high-latency; bands are one standard error. The marker is the event day.
Cumulative-return paths from two days before to three days after the event, for positive and negative news on normal and high-latency days.From the paper to the dashboard
The data behind it, and how to build on it
What the dashboard provides
The internet-quality side of this study is KASPR Datahaus measurement. The Regional Intelligence dashboard publishes the same underlying series for the United States — national latency and jitter, daily and hour by hour — which this analysis reduces to a single daily dispersion measure. Those series are open to read and download there.
Open Regional Intelligence →What you bring to it
Network quality on its own says nothing about markets. To reach the result above, the KASPR series is matched to three external datasets: daily stock returns from CRSP, firm-level news sentiment from RavenPack, and intraday market microstructure from NYSE Daily TAQ. The join is by firm and trading day, across 1,166 US-listed companies from 2020 to 2025.
Sources
KASPR Datahaus · CRSP · RavenPack · NYSE Daily TAQ
Ackermann, K., Angus, S., Cui, B., & Raschky, P. A. (2026). News, Latency, and Stock Prices. Working paper, Monash University.
Estimates are within-firm interactions on a balanced panel of 1,166 firms, 2020–2025, absorbing stock, year, month and day-of-week, with standard errors clustered two-way on ticker-year and date. News coverage thins sharply after 2023, so news-based estimates lean on 2020–2023. Figures describe association, not a proven causal channel.