Reaction Times to Economic News - US Edition
In a previous article , I had described how incredibly fast economic news releases trigger trades on Eurex - less than 100 ns after the official release time.
Eurex, however, is not the only exchange where this happens. It is just the one where the publicly available timestamps make such an analysis very easy. The earliest timestamp in the CME market data is the gateway-in timestamp. Its variance does not allow one to reverse-engineer the time when the request was sent by the participant - at least not to within a few nanoseconds.
Having said that, Fig. 1 below demonstrates how quickly the PMI releases are priced into the markets. The various release dates are on the x-axis. The y-axis is the order arrival time (gateway-in timestamp) past 10 am ET. Each marker represents a trade in one of the most liquid futures (list in the legend in the right-hand panel). The grey area represents an estimate for the minimum physically possible latency (length of cables, path through switches). This value had a few changes due to infrastructure changes and performance improvements .
We see a number of orders which must have been sent between 10:00:00.000000000 am ET and 10:00:00.000001000 am ET, i.e., within 1 μs of the release time. Later trades could be slower reaction to the PMI data or they could be reactions to the initial, fast reactions.
The right-hand panel shows the cumulative markout PnL (trade price vs. T+ 10 s mid) by product vs. the latency of the order. The chart only shows the first 10 μs, but the pnl curves are flat afterwards. This means that the news is fully priced in within just a few microseconds.
Truly amazing.
Quantitative Analyst at LaSalle Trading Group, LLC
2 个月This is very interesting. Do you have the same chart for CPI data?
Capital Markets Execution Consultant for Public Companies | Share Buy-backs | Group Exco | Global Head Execution Services | Algorithmic and Program Trading | Governance | Risk Managment | SM&CR Significant Person
3 个月And I would guess that something like 90% of the pnl accrues to the single fastest participant? There will always be a race to improve as the pnl prize is huge no?
Managing Director, LYO-X GmbH
3 个月Stefan Schlamp is this good or bad? Is there not a risk of creating closed loop decision systems that can amplify itself?
Quantitative Research | Ultra-Low Latency | ML
3 个月As always great analysis Stefan!