Understanding Trading Latency Across Markets and Infrastructure
Summary
The document asks how order-processing latency varies across products and trading infrastructure. It defines realized latency as the interval from local order submission until processing, and points out that latency can matter in equities, options, futures, and foreign exchange. Co-location, direct market access, and intermediaries may produce different timings, while the question leaves fixed-income latency open for discussion.
The responses provide little current comparative evidence. One points readers toward specialist latency research and measurement resources. Another distinguishes matching-engine latency from broader order-path delays and gives a historical range spanning hundreds of milliseconds to hundreds of microseconds, with a 2010 example of a reported average round-trip time at the Singapore exchange. It suggests that instruments on the same venue may share a matching engine, though internal prioritization is possible. The figures are dated and do not establish typical latencies by product or infrastructure, so the document is better read as a framing of what must be measured than as a current benchmark.
Key ideas
- Realized latency is framed as the time from local order submission until processing.
- Latency may differ across products and infrastructure, including co-location, direct access, and intermediaries.
- Matching-engine latency is only one part of the order path.
- The cited latency figures are historical and do not provide current product-by-infrastructure benchmarks.
Tags
Full text
# What are the typical "realized latencies" across different products and infrastructures? # What are the typical "realized latencies" across different products and infrastructures? Latency has been a hot topic for a while, first in the industry and more recently also in academia. It is very common to hear about milliseconds, even microseconds. While most of the media attention is on equity markets, latency is an important variable in options markets, futures markets, and FX markets (arguably, it is not so important when dealing with fixed income instruments - comments on this are more than welcome). Latency is different across markets for different products. Also, within a product class, latency changes across different infrastructures (co-location, DMA, various types of mediation). Question: What are the typical latencies across different products and infrastructures? EDIT: To address one of wburzyns concerns, I changed the title to "realized latency". For lack of a better term, I use "realized latency" to mean "the time from when the order is submitted locally to when it is processed". As an answer prototype, below is a fairly outdated illustration dug out from a SUN white-paper, where different markets and market activities are binned according to the typical latency they face. I would like to know where do the markets stand now and see a similar classification for markets and infrastructures (not so much market activities). If you think this question is too broad, how would you break it up? ## Answer by Tim (score 5, accepted) https://quant.stackexchange.com/a/2566 You may want to look at STAC research, http://www.latencystats.com/ and other sites that you can find by searching for "latency" and "market data" I have no affiliation with those two links, however I am a principal in a company that measures and compares latency from exchanges. ## Answer by wburzyns (score 7) https://quant.stackexchange.com/a/2178 If this question is about matching engine latency then there are plenty of resources on the web. The short answer is that across various exchanges the latency is currently in range from hundreds of milliseconds to hundreds of microseconds. To name a few (as of 2010): AFAIK Singapore Stock Exchange is going to launch / already launched the world’s fastest matching engine with average round-trip time of 90 us. I've never heard about different values of latency for different classes of instruments on the same exchange. Although different classes, they are all serviced by the same matching engine. However I can imagine some internal prioritization/etc.
Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)
This summary was written by Stratmill's research agent from the original; it is not a copy of the source.