Given the growing focus on trading speeds at exchanges and other execution venues, a standard definition of latency would benefit the whole trading community, according to Peter Robin, senior director at technology firm Cisco Systems’ Internet Business Solutions Group.
“Lots of attention has been applied to latency but our first observation was that there is no clear definition of latency,” says Robin. “A standard-setting body or authority in the market could do a great deal of service to the trading community by producing a consistent set of definitions.”
During a recent study he produced for Cisco, Robin identified six different interactions between exchanges and their members where it is important to measure latency. These are: sending a price from the exchange to the member, which can account for between one and five milliseconds; taking the price from the members’ firewall to its price distribution mechanism; preparing an order for execution and sending it back to the exchange; placing the order on the exchange’s order book; generating the order acknowledgement; and finally, sending confirmation to the member that the order has been executed.
However, some platforms’ latency numbers only focus on some of these areas. “It would be beneficial for exchanges’ members to map out the whole chain of events,” says Robin. “These areas are well known by now because so much attention has been focused on improving latency, but some exchanges only talk about latency in terms of the time it takes to produce the order acknowledgment.”
This particular figure is important for algorithmic traders because as soon as the order acknowledgment is received, algorithms start to look for their next trade.
However, Robin believes that acknowledgement should not be the sole focus of platforms’ efforts to reduce latency. Another important area is order cancellation.
“One of the more surprising elements in the survey was the ratio between the number of orders that a member issues and the number of trades that actually go through the exchange,” he says. “For a typical equity order, a member generates five orders to one trade execution. For other types of asset classes such as commodities or derivatives, the ratio can be as high as 10,000 orders to one execution.”
Given this, Robin thinks platforms’ latency reduction efforts should also focus on cancellation. “If someone wants to cancel, this needs to get processed faster than the order confirmation itself,” he says.
Although Robin advocates a standard measurement for latency, he acknowledges that it will be difficult to decide who should be responsible for implementing it. “Exchanges themselves may need to take control of this and formulate a standard definition for trading times and latency,” he concludes.