Waters Technology Roundtable Discussion on LatencyStats.com
Waters Technology Roundtable Discussion on LatencyStats.com
In the latest Waters Technology roundtable chaired by Jean-Paul Carbonnier, Mark Schaedel (NYSE Technologies) and Fergal Toomey (Corvil) discuss LatencyStats.com with Mark Reece (HSBC) who offers an end-user perspective.
WatersTech Panel: Raising the Standard for Latency Transparency
Author: JP Carbonnier
Inside Market Data 06 Aug 2010
With the importance of low-latency data to algorithmic and high-frequency trading, exchanges, alternative trading venues, and market data and technology vendors have all been keen to demonstrate their performance by quoting latency statistics. But without a clear and consistent methodology governing how latency is measured, these claims can be opaque and of little value to end-user firms.
This problem has prompted a number of industry initiatives aimed at establishing standards and promoting greater transparency around latency measurements, which are gathering pace-including FIX Protocol Limited's Inter-Party Latency Working Group and the Securities Technology Analysis Center (STAC).
However, challenges still remain to providing useful and usable data to end-users. "There are a lot of missing pieces to the puzzle as customers try to identify how they can improve their own latency and how they compare to others," said Mark Schaedel, senior vice president of global data products at NYSE Technologies, the trading technology and market data arm of NYSE Euronext, in a video briefing hosted by Inside Market Data's sibling news portal Waters Technology.
NYSE and Dublin-based latency management systems vendor Corvil recently launched Latencystats.com, a portal that provides latency statistics and full latency profiles for the exchange's US ArcaBook Equities and OpenBook Ultra feeds, updated every minute, along with a detailed methodology outlining how the statistics are calculated (IMD, June 21). While initially focusing on US feeds, Schaedel sees almost "infinite" potential to expand the breadth and depth of the portal's statistics, based on client demand.
In addition, NYSE and Corvil aim to provide more ways to access data from the portal. "We are considering making the data [from Latencystats.com] downloadable in a machine-readable format - for example, in a spreadsheet - which will allow people to integrate it more directly with their own performance management systems," said Fergal Toomey, chief scientist and co-founder of Corvil.
This ability to integrate latency data "in real-time, even away from the portal, so it can be fully integrated," could be valuable to end-user firms as an input for systems management or even order-routing decisions, said Mark Reece, e-trading solutions architect at HSBC. "I see that as the next big step for the banks in terms of understanding the trading process and the exact timings that matter to them," he said.
Making similar statistics for more market data feeds and trading venues available via Latencystats.com would also be useful, Reece added. In fact, NYSE and Corvil are already trying to enlist other service providers to publish their latency statistics via the portal, even if they don't use Corvil's latency measurement software.
"There are certain characteristics or capabilities in measurement technology that are desirable for latency measurement," Toomey said. "For example, to produce accurate measurements, you need a way to generate accurate timestamps, deal with the issue of clock synchronization and correlate a large number of market data events that are happening in different places. Those are quite technical issues... and to produce results on an apples-to-apples basis with those that are already there on Latencystats.com, we would like to ensure that those issues are being handled appropriately - but we wouldn't have any objections if a different vendor's technologies are used."
Indeed, if other exchanges and service providers publish data via the site using consistent, comparable methodologies, it could go some way to creating "a de facto standard of how [latency] data is published and for transparency around how those statistics are created," Schaedel said.
Other initiatives - such as FPL's Inter-Party Latency Working Group, which aims to create a common language for defining the technical components needed to monitor latency and delimit the various links in the trading chain to ensure common interpretations of measurement points - also need defined standards. "There is the technical side of measurement-accurate timestamps, how you measure, whether it is coming off the wire or from a log file or from a process - but it is also about where you measure... [and] defining the measurement points so that you can get that like-for-like comparison," Reece said.
In fact, these initiatives may yield mutually beneficial standards that encourage participation from other service providers. "Our hope is that Latencystats.com will increase interest in latency transparency and also add momentum to the desire for a more consistent approach to measuring latency," Toomey said. "The standards work is being done in the same general area, so hopefully each of those efforts will contribute to the other," he added.
Even if the industry succeeds in defining standards for measuring latency, comparative latencies may still remain veiled in secrecy. Technology already exists to generate anonymous latency league tables that rank each firm's end-to-end latency anonymously against its peers, though such solutions would require secretive trading firms to volunteer their own latency statistics.
Should privacy hurdles be overcome, it would be "useful to have a way to anonymously compare your latency to the rest of the market," Reece said, though he emphasized that such a service would need to be "a reasonably sophisticated measure" of latency and provide rankings based on latency profiles rather than just averages.
For now, sell-side firms are more likely to be guided by the demands of their buy-side clients, who will readily inform them if they begin to lag behind their competitors. "Typically, we have a set of customers telling us how well we are doing on latency where it matters to them, because they measure us in the same way that we measure the exchanges," Reece said.