Latency (trading)
Latency is the time a message takes to travel through a trading system: from the exchange to your screen for market data, from your click to an acknowledgement for orders. It is made of distance, network hops and processing at each step, it varies from one message to the next, and it is only meaningful when you say which two points it is measured between.
Senzoukria · Glossary · Updated September 2026
Kinds of latency
- Market data latency: exchange event to your application, including the vendor's processing.
- Order round trip: your click to the gateway's acknowledgement.
- Tick-to-trade: a market event to your order reaching the exchange, the figure automated firms optimise.
- Jitter: the variation between messages, often more harmful than a constant delay.
Measuring it honestly
A single number hides the distribution. The median describes a typical order; a high percentile, such as the 95th, describes the slow tail that hurts in fast markets. A percentile needs enough samples: with very few, a 95th percentile is just the worst measurement.
Worked example: with 20 sorted measurements, the 95th percentile sits at position (20 − 1) × 0.95 = 18.05, counting from 0, that is between the 19th and 20th values. If those are 180 ms and 240 ms, linear interpolation gives 180 + 0.05 × 60 = 183 ms.
In Senzoukria
The trading panel measures order latency from the click to the gateway's acknowledgement, using a monotonic clock so that a system clock adjustment cannot produce negative or flattering durations. Placements and cancellations are kept apart because they do not travel the same path. It keeps the last 200 measurements and ignores those older than 30 minutes. The panel shows nothing until a first measurement exists, then the median of order placements with the number of samples, for example '38 ms median · 12 samples'; the 95th percentile, interpolated as above, is added only from 20 measurements. The hint is explicit: this is not the time to a fill, and the client path adds nothing measurable, so the figure is essentially the broker's round trip.
Common mistakes
- Quoting a best-case ping as the latency of order execution.
- Treating acknowledgement time as fill time.
- Optimising software paths that measurement shows add nothing, instead of the network or broker path.
Related
In the same section
This page in other languages
Frequently asked questions
- What latency matters for a discretionary order flow trader?
- Mostly consistency and the absence of long stalls. A stable round trip of tens of milliseconds is far below human reaction time; occasional multi-second delays during busy moments are what cause real slippage.
- Why does Senzoukria hide the 95th percentile at first?
- Because below 20 measurements it would simply be the worst one and would not change as more data arrives. It appears once there are enough samples to describe a tail.