If you want to beat the next outfit to the punch by a millisecond or a microsecond-as many high frequency traders aim to do-you're going to have to constantly wring out bottlenecks and delays in the way your instructions get sent to and executed by the various exchanges.
Enter the demanding exercise now known as latency budgeting.
High-frequency, low-latency tra-ding now accounts for 70 percent of all trading in equities, according to Aite Group. The high-frequency trading unit of the Chicago hedge fund Citadel Investment Group raked in almost $1 billion in trading profits in 2008 from the practice. And its 2009 numbers aren't in yet.
But the message is clear: Numbers like these mean that for those traders whose strategies require the highest speed of performance, every millisecond matters. And unless you manage by the numbers, lowering the latency in your trading systems every year, every month and every day, you will not be able to compete with better, more vigilant managers of latency "budgets."
It's not just time that must be budgeted. Time also is money. Buy side and sell side firms each must keep careful watch on IT costs. Because each microsecond saved has a calculable cost. And, in the words of the rock standard, it keeps getting higher and higher.
"The cost of lowering one's latency on any given component of the trading operation is rising. In particular, we see this trend in co-location costs and, for that reason, there is greater focus on the total latency management process," explained George Hessler, executive vice president at New York-based Lime Brokerage, an agency broker that works with many high-frequency trading firms.
Co-location has installation, maintenance and rental costs, as firms try to put their execution systems as near as possible to venues' matching engines. But every aspect of high-frequency trading matters.
"Latency budgets only truly came into vogue in the last four or five years. That's when we saw purely electronic trading or trading without a person in the mix," says Matt Meinel, global director of business development at 29 West, a Chicago-based provider of ultra low latency messaging platforms and solutions. Once automated trading systems got combined with electronic exchanges, the performance limits associated with human interaction were removed, he said. And the bar inevitably got raised on trading performance.
Time is Tight
Simply defined, a latency budget is kept in order to assess, monitor and then manage how much time a firm expends on each and every step of the trading process, in both seconds and dollars.
For a broker-dealer or hedge fund measuring equity trading, what gets watched is the round-trip time of an order-how long it takes for trading data to travel from your servers to an exchange's matching engine for execution and then back again.
This is sometimes referred to as end-to-end latency. The less latency, the faster trades are executed and the more trade fills you will see.
Alternatively, if you are an exchange, you are analyzing latency from the time you receive an order to the time you send a filled order back. This is door-to-door latency.
Managing a latency budget means determining precisely how much time in milliseconds or less is taken up by each component in the trade process and then determining if one wants to or can reduce latency at any step along the way.
For instance, says Meinel of 29West, a hedge fund may determine that it is not seeing as many trade fills as it would like on a particular high-speed trading strategy. The fund might want to lower latency in the hopes of filling more trades or completing more trades at better prices and thus, in either case, upping firm profits.
Technicians then get to set up the steps to track in the latency budget, arrive at averages for the current times each step takes to complete and help select stages where they judge if there is room for improvement. These can be, for instance, the time it takes data to travel from an exchange's matching engine to the servers on a fund's premises or the time it takes for a packet of market data to be received by a feed handler to the time it takes for that data to be acknowledged by the firm's messaging software.
According to John Panzica, vice president of facilities manager Switch and Data's financial services practice, this establishes a firm's "latency value chain."
Every step involves a combination of hardware and software, from time spent running through middleware, applications, servers, other processing equipment, network routers, and cables, right down to such details as the types: dark or lit fiber, Ethernet 10G or Infiniband. Then, there's the number of network points involved, the location of venues, the exchanges selected, etc.
There are customized approaches. Hessler of Lime Brokerage breaks down latency analysis into six groupings-communications-related latency, feed handler latency, latency related to complex event processing performance, risk management-related latency, market center processing and reporting latency.
And almost all budgets are custom-built and custom-managed. More than one source interviewed for this story asserted that given the broad range of trading strategies and technology configurations employed by various traders, it is nigh impossible to assign average latency ranges to each and every component of a trade execution. Hessler of Lime Brokerage, however, estimates that for those trading firms whose trading strategies require low latency, most of the steps have latencies ranging from microseconds to multiple milliseconds. Alternatively, U.S. market center latencies range from hundreds of microseconds to multiple milliseconds.
A microsecond is a millionth of a second. A millisecond is a thousandth of a second.
Be the first to comment on this post using the section below.