Securities Industry Article: Keys To Success for Remote Traders: Co-Location and More Co-Location
In this SIN article, the focus again is co-location. The entire article is below, but we want to use the article to again highlight the problem of co-location from a market structure perspective.
“Exchanges such as NYSE and the Boston Options Exchange have menus of different levels of service that they provide to traders who are encouraged to “co-locate” their servers on site.”
Some ramblings. If you opt for COLO Option 7, for reduced cost, is it even worth it, despite it being “cheaper”? Would you want to eat a $0.49 hamburger? The bigger prop shop you are, the bigger your budget, and the bigger your advantage in this proprietary front-running game. Some folks are inside the matching engines, and not just “close to it”.
And for those who persist in the argument that “anyone can co-locate if they choose”, it gets tiring. Can retail investors co-locate? At what service level? Does COLO OPTION 7 have a disadvantage versus COLO OPTION 2?
The Exchange Model: They need to maximize their revenues quarter to quarter, and this is done not by transaction fees, but by tape revenue and co-location revenue. And the largest revenue growth is hands down in co-location fees. Where they used to get paid in transaction commissions from brokers etc. The transaction fees are now nil. So check it out. Investors benefit because they pay $8.99 a trade. Cough. Except when you take into account the scalp-vig paid. Let’s see that talked about in the media. The Exchanges don’t mind; they still get paid more than ever. They just don’t get paid to execute a trade fairly anymore via a commission; they get paid to allow tiers of access for their preferred friends and partners, who ante up because they make money by co-locating their scalping prop algos, at the expense of those who do not co-locate (or who buy the $0.49 Hamburger).
They sell those that will pay co-location space. They sell those that will pay data feeds. They sell those that will pay data feeds with sensitive information. They do not care about investors; they care about who will pay the vig. Caring about investors now falls at the feet of the SEC and the SEC only. I sure hope they are up to the task.
By the way, in the below article, Richard Gorelick defends co-location as his firm pays for the service, and engages in “high frequency trading”. He actually rang the bell on the NYSE this morning. HFT firms are now DMM’s (Getco), and ringing the till, and now they also are ringing the bell.
OK. The SIN article :
May 3, 2010
With the need to act with speed so great in today’s securities markets, there’s only one option for traders whose offices sit hundreds of thousands of miles away from exchanges and alternative venues.
If they want to grab a piece of the profits in price movements that take place in thousandths and millionths of a second, they must put their trading engines under the same roof as the venues’ order-matching engines. Otherwise, they will suffer the indisputable laws of distance and physics. They won’t execute trades fast enough.
To make up for that distance, farflung traders must co-locate their server-resident trading algorithms near or in the exchanges.
Traders “want to be as close to the metal as they can. There’s no way around physical speed” and distance, says Karl Denninger, an independent trader and an Internet engineer who’s installed trading networks. “Every switch you are behind adds another level of latency. If I am two switches behind the matching engine, I am at a disadvantage. If I am two switches and 100 miles away, I am paying an even bigger penalty.”
So how does a high-frequency trader based in Austin, for instance, bridge the 1,750 miles between his or her office and the New York/New Jersey area where there’s the heaviest concentration of trading venues?
The answer is simple: They deploy their trading algorithms on servers “co-located” with the matching engines. That minimizes delays and puts them on equal footing with traders who work on the other side of the Hudson River.
Richard Gorelick, chief executive and co-founder of RGM Advisors LLC, is one such trader, who works from afar but competes with the most sophisticated outfits in New York.
Gorelick is a high-frequency trader based in Austin, the capital of Texas. He’s the G in RGM.
RGM doesn’t have individual traders per se. Like many high-frequency trading firms, RGM has researchers who write the trading algorithms and operations people who deploy them on the co-located servers.
“In some markets, we are [co-located] in the exchanges. In others, we feel like we are close enough if we are in the same region,” he says
“In the region” generally means within 25 miles of the trading venues’ matching engine. For example, one of Switch and Data’s 34 data centers is located in North Bergen, N.J., with Nasdaq 22 miles to the south in Carteret and the New York Stock Exchange’s new data center about the same distance to the north in Mahwah.
At the same time, Gorelick doesn’t underestimate the value of human intervention in high-frequency trading.
“There’s an important human role in automated trading. [Machines] follow precise instructions, quickly, repeatedly and without emotion. [Humans] have higher- level judgment and analysis they use to build and configure automated trading models, determine whether trading is going as planned and make decisions about what models to run under what conditions,” he says. “For the roles that humans perform well, it is fine to be located far from the exchanges and to be subject to the slight latencies caused by the speed of light and the laws of physics.”
Such judgment might be tweaking a trading algorithm with new parameters or pulling it entirely if it’s proving ineffective.
Exchanges such as NYSE and the Boston Options Exchange have menus of different levels of service that they provide to traders who are encouraged to “co-locate” their servers on site. John Goode, CIO at the Boston Options Exchange, likens a trader’s office to a cockpit in an airplane.
“[Offices] are more like a command and control cockpit and the only latency is between cockpit and the servers,” he says. “The algorithms allow them basically to put in the parameters and have the computer do the trading throughout the day.”
It’s largely the same drill for traders on other continents that want to be competitive in New York-area trading venues.
Take Kairos Asset Management, a hedge fund based in Sao Paulo, Brazil, where high-frequency trading has begun to emerge in early form since markets there became fully electronic in the past 18 months.
Kairos uses high-frequency algorithms and technology developer Lime Brokerage LLC based in New York for low-latency execution into New York-area venues. Kairos has been using co-located servers to trade derivatives in the BM&FBovespa exchange based in Sao Paulo. In capitalization, the B&MFBovespa claims to be the fourth-largest exchange in the Americas behind only the NYSE, Nasdaq and Toronto Stock Exchange.
“We are maybe one or two feet from [the B&MFBovespa’s] matching engine,” says Kairos chief operating officer and managing director Alberto Araújo. “High-frequency trading in equities has not taken off yet because unlike in the U.S., traders are charged for canceled orders,” he noted. In high-frequency trading in the U.S., for instance, as much as 90 percent of the thousands of orders that are placed every second actually get canceled, when the desired price or amount of stock can’t be found on a given stock at a given moment.
For the time being, Kairos uses its B&MFBovepsa co-location for trading futures, which demands speed. Brazil doesn’t have dark pools, either, and won’t unless the country’s disclosure regulations are relaxed, he says.
Several vendors and traders observed that co-location tends to get a bad rap because high-speed trading remains controversial. Indeed, one of its most vigorous critics is Denninger, who writes The Market Ticker, a popular financial blog. He believes traders constantly trying to jump in front of each other to eke out tiny spreads is a valueless proposition.
“I can code in assembly and could throw a couple of blades servers up there, [but] the only way to win that game is not to play,” he says.
No doubt, co-location is geared toward the speed game and gets criticism for giving certain traders an unfair advantage. But some vendors and traders say the complaining is unwarranted.
“It’s the great leveler and makes the markets more fair,” says RGM’s Gorelick, who adds that co-location is also relatively cheap. RGM, he says, generally pays a few thousand dollars a month for co-location space and power for a rack that can stack 42 blade servers together.
Co-location, according to vice president of NYSE Technologies Ken Barnes, doesn’t necessarily imply favoritism.
He argues that NYSE solicited feedback from all its customers for its new data center in Mahwah, N.J., which opens in August and into which customers begin installing co-location equipment this quarter. After the feedback period, NYSE Technologies made sure co-location space was not closed out to new customers.
“We came back made sure it was not oversold. Literally, not one person was denied access [to co-location],” he says. A NYSE spokesman explained that the data center will build out co-location “pods” of server racks as demand dictates over time.
The starting point for NYSE co-location services, which generally charges a 30 percent premium over third-party data center services given better proximity, is under $10,000 a month, according to Barnes.