Milliseconds Matter

Although every second counts, algorithmic trading is much like running a marathon - the runner who begins the race right before the finish line is likely to beat opponents traveling the full 26.2 miles. Likewise, the algorithmic trader whose server is located closest to the source of trading data is likely to achieve the fastest execution. But instead of a gold medal up for grabs, the prize for a quick algorithmic trader could be millions of dollars.

"What didn't make a difference a couple of years ago - because it was just a split second - has all changed," explains Robert Iati, a partner with Westborough, Mass.-based TABB Group. Advanced electronic execution platforms, such as algorithmic trading, have created a need for speed down to the millisecond. Firms that can capitalize on those milliseconds are likely to get the best execution in an ever-competitive algorithmic trading environment, he relates.

While a desire to enhance profits drives the need for speed, firms also are facing Reg NMS, which requires best execution at the best price. And data latency presents a hurdle to achieving immediate execution.

Many Wall Street firms have made great strides to reduce data latency internally by reorganizing data distribution within the enterprise to reduce bottlenecks or by investing in tick-data systems designed to process data quickly. But recently, an external option for reducing data latency has emerged. Firms are beginning to colocate servers armed with algorithmic strategies near or at market centers or electronic communications networks (ECNs).

How It Works

Colocation can occur in a variety of arrangements. In one scenario, some exchanges now offer firms the ability to rent space within the exchange's facilities to colocate servers.

Nasdaq has been offering this service for about a year and currently has 10 clients on board, says the exchange's chief information officer, Steve Randich. "By colocating, these firms are putting algorithms in the location that's going to get transaction turnaround in just a couple of milliseconds, as opposed to what could be 20 to 30 milliseconds for the data to travel across the country," he says. "Twenty milliseconds make a difference in terms of what price you get per order."

Trillium Trading, an affiliate of the Schonfeld Group, has offices in New York, Boston, Miami and Edison, N.J., and colocates servers at Nasdaq. Paul Famighetti, director of automated trading, rents a cabinet at the exchange for a monthly charge. The cabinet consists of a rack in the data center in which Trillium can fit about 43 servers as well as a self-cooling system to keep the servers functioning smoothly. Famighetti operates the site as a "lights out" data center - it is designed to be manipulated remotely with just basic maintenance provided by Nasdaq's employees.

In addition to Nasdaq, Trillium colocates at Archipelago and Instinet, and may soon do so at the Chicago Mercantile Exchange as well. Famighetti estimates his monthly costs for colocation at approximately $100,000, but he emphasizes that the profitability from faster execution outweighs the costs. And, if nothing else, he says, at least Trillium is on equal footing with other algorithmic traders that have servers at Nasdaq.

"You get proximity to [Nasdaq's] matching engine, which is important. The margins for profit are thin, so anything you can do to eliminate slippage due to geographical latency helps," Famighetti says. "When one of the automated strategies sees a change in the marketplace and wants to react, we get the first bite on the wire."

Famighetti concedes that Trillium is lucky to have a large enough technology budget to colocate at multiple marketplaces. "If you're committed to one place, you'll lose to people colocated at other exchanges," he admits.

Nasdaq's Randich notes that all of the exchange's colocation relationships are currently with "household names" on the sell side, but he doesn't rule out the potential for buy-side firms working their own quantitative strategies through algorithms to colocate at the exchange. Since firms participating in those strategies need to remain competitive, he says, "I do think we'll double our clients on this in the next year."

However, there will be an eventual cap on the market. "The universe of firms that is doing this is limited because speed is only necessary for firms with specialized strategies," Randich says.

The Options

For Wall Street firms with smaller budgets, opportunities for colocation do exist. Trillium's Famighetti notes that some of his firm's buy-side clients utilize space in Trillium's colocated servers for their own algorithmic trading. Another option for smaller sell-side and many buy-side firms is turning to agency brokers such as New York-based Lime Brokerage for colocation services.

Additionally, networking vendors also are dipping their toes into the colocating space. BT Radianz recently launched Radianz Proximity to host customers' algorithmic trading applications in a Northern New Jersey site located near Manhattan's execution venues. The service offers flexibility, says Radianz's chief technology officer, Brennan Carley. "Some companies want us to manage their computer processes entirely - you give us your algorithms and we'll run them on our hardware, back it up, restore it and monitor its applications," he describes. Other companies, he adds, feel their algorithms are "super secret and confidential, so its completely hands-off. They give us their server, but we don't look inside the box."

Either way, Brennan says, Radianz never examines a firm's algorithms for anything other than their operational status. "We don't trade, so we're neutral for them," he asserts.

Michel Debiche, president and CEO of Quantia Capital Management in Princeton, N.J., says neutrality would be a vital component in a colocation situation. "There has to be a level of trust that your provider is not going to try to figure out what your algorithms are doing, since many sell-side brokerages are also algorithmic providers," he explains.

Though Quantia, which runs an equity statistical arbitrage fund, is not currently colocating its servers, Debiche acknowledges that proximity is important to him under certain circumstances. As a result, there are times the buy-side firm chooses to use a broker partner's algorithms instead of Quantia's own models if that broker has the advantage of proximity to a marketplace.

As for colocating his own algorithms at an execution venue, Debiche says, "A platform with big pipes, fast servers and a facility for users to load their algorithms onto their platform - where the algorithms don't have to be visible to the platform providers - could add real value for a buy-side algorithmic trader."

The Drawbacks

After years of investing in disaster recovery capabilities, some Wall Street firms may be hesitant to establish servers close to, or even on top of, exchanges, according to TABB's Iati. He concedes that colocating servers presents "a conflict of interests," since firms have been encouraged to house a copy of their data a safe distance from the source. And while algorithmic trading may not be considered a core function like back-office record keeping, Iati says, "The more [firms] commit to program trading, the more those systems become fundamental to their profitability and the less they could survive if those systems disappear."

Though some sell-side firms and a few buy-side firms are beginning to find value in colocating servers, Iati projects the process will become more mainstream in just a year's time. Firms don't want to be left out - "The competition out there is only heightening," he says.

---

Colocating With ...

An execution venue:

Exchanges and ECNs provide the closest proximity - and thus the fastest speed - to the trade. However, for firms with arbitrage strategies crossing multiple exchanges, colocation and connectivity within their infrastructures could be costly.

A broker:

The sell side is realizing the value proposition of offering its clients closer proximity and faster speed. However, a buy-side firm risks the secrecy of its algorithm if revealed to its trading partner.

A third-party vendor:

Neutral parties can provide buy-side access to many exchanges in a closer location. However, no third party can offer as close proximity as an execution venue, which could be a deciding factor in an algorithm's success.