Feature: Colocation - A Game Worth the Candle?
Exchanges, networks and service providers are turning to the laws of physics in the race to reduce latency and win customers. But do colocation and proximity trading really offer substantial benefits worthy of the investment, or is it mainly a triumph of marketing over common sense?
The market's need for speed is genuine. In today's market milliseconds really do matter, although they matter more for some than others. What is more debatable is the best way to achieve a latency rate of milliseconds.
Colocation and proximity trading have become the latest area of interest for those firms seeking ultra-low latency. By way of definition it is commonly accepted that proximity trading refers to sites located as close to the venue as possible while colocation refers to the more recent practice of actually being located within the firewalls of the exchange itself.
For colocation services of this kind to flourish, the exchanges have had to get involved and look for partners to develop these services. Fortunately, the increasing pressure that they face from regulations such as MiFID and RegNMS in the US, plus a demand from the market for lower transaction fees has forced them to compete on price and product. Many have targeted the low latency market as a possible source of new revenue streams.
The London Stock Exchange is currently in talks with a number of brokers to have their algorithmic trading engines inside the exchange's firewalls. The New York Stock Exchange has colocation services offered within its data centres operated by SIAC Sector. Most recently, in the commodities and futures market the Chicago Mercantile Exchange has opened up its doors and through a venture with colocation provider Equinix is set to launch the Lnet Service in the first quarter of 2007 for users of the Equinix Chicago Internet Business Exchange centre.
"...it is not just the speed that is
important with colocation, it is also
about the reliability..."
"It is the first offering by a major commodities and futures exchange allowing a direct connection to its matching engine," claims Dan Walker, vice president, sales for the Western and Central regions of the US at Equinix. "The idea is to eliminate the middle man and remove any unnecessary latencies and network hops. The goal is to get the network latency as close to zero milliseconds as possible without residing in the engine itself."
Walker expects a great take up of these services, most likely from the sellside prop shops and high frequency hedge funds. Dmitry Bourtov, manager of the Solaris Market Neutral hedge fund uses colocation selectively for proprietary trading purposes. "The advantage is that it significantly cuts our messaging time so we are able to make our decisions on whether to buy or sell a lot quicker."
Bourtov sees developments such as the CME and Equinix Lnet service, where he can plug directly into the exchange's router, as a "big step forward" on the basis that the closer you are to the exchange's network then obviously the lower the latency. However, he also warns that there is more to consider and there is room for improvement.
"Sometimes it is not just the speed that is important with colocation, it is also about the reliability and how often the system goes down," he says. There is also the issue that latency can be lost elsewhere in the process chain. Because of the inefficiencies of the service providers involved, it reduces all the customers to the same latency so that even when you have a two millisecond connection to the execution venue you are not getting any gain. "This issue needs to be addressed on a vendor-by-vendor basis because it is extremely important," he says.
Elsewhere on the sell-side there are also concerns about whether there are significant gains to be made from investment in colocation services. Richard Balarkas, head of Advanced Execution Services at Credit Suisse accepts that speed is very important to the statistical arbitrage traders and ultra-high frequency hedge funds, but he also believes that the money being made in these markets from speed has been reducing.
"...it is getting harder to measure whether
changes in latency are really producing
markedly better performance.."
"Five years ago a hedge fund may have been able to say that each 50 milliseconds shaved off latency equated to $X,000 of revenue," he says. "But the process has been finessed so much that it is getting harder to measure whether those changes in latency are really producing markedly better performance."
For Frédéric Ponzo, managing director of financial technology consultant NET2S, the differences in distance and speed for a West End based hedge fund trading on the LSE being offered the chance to place its servers underneath the chair of the exchange's chief executive are so slight they are bordering on lunacy. "Light travels at 200km/sec in glass or optic fiber, therefore you are gaining 1 millisecond every 200km, so locating 10km nearer will save you 50 microseconds."
Milliseconds do matter, he says. For fixed income they matter in the order of 150-200 milliseconds. In equities it is a little lower and for arbitrage activity or small-cap funds it is still a case of 50-100 milliseconds and shaving 50 microseconds, or 0.1%, is not going to make any difference.
There is also a logical finale to the development of services for clients who have velocity as their modus operandi and where the way to beat your competitor is to get as close to the execution venue as possible, says Kevin Bourne, global head of execution at HSBC.
"Everyone is focusing on how quickly
they can get from their trading desk
to the exchange."
"The problem with this approach is that everyone can make the same engineering changes to their systems and the natural conclusion to this strategy is that everyone ends up at the same point. Then ultimately everyone will be able to process trades at the same speed. Once we reach that point then I think the question will be whether all that money that has been spent on colocation and performance improvement is in any way recoverable."
There are also additional or secondary costs that arise from low latency and the colocation model says Bourne. "If you are trading faster then you are likely to want to increase the number of transactions and that means you will need bigger servers. You will need the IT team to plan new systems and you will need to settle more trades in the back-office, which will have an impact - as will the increased fees from the exchanges. So economic advantage is only realised when all costs are taken into account."
Perhaps most importantly, though, there is the fact that proximity trading and colocation only address one part of the latency challenge despite what the providers may say and all that effort is wasted if the inefficiencies of the other parts of the process chain are neglected.
"There is a widespread misconception that latency is coming from the network," says Ponzo of NET2S. "You have traders saying that the network is slow but when you dig deeper, it is clear that it is the applications that cause most of the latency." Investing large sums in improving the network is akin to using a sledgehammer to crack a nut, he says. "If you have a slow truck and you take it on the motorway, it is still a slow truck," he says.
Firms should instead be addressing the way their applications are designed, the way they are coded and the way they are used within the organisation. But, says Ponzo, the people running the applications have been developing them for five years and are not about to take the blame for latency and blaming the lost milliseconds on the network is an easy concept for traders to grasp.There are also historical reasons for the latency in applications, namely that when they were developed five years ago, latency was not an issue except for a handful of users trading on a select group of asset classes. Instead the focus was on being able to handle multiple users and large volumes.
"It is only recently that there has been a
realisation that applications can cause a
few bottlenecks as well..."
"It is very rare that you will find a client or a broker where every single application in this chain has been designed with low latency as their primary objective," says Balarkas of Credit Suisse.
If firms really want to address latency in the applications for those asset classes and users for whom latency really matters, then they should be starting with a blank sheet of paper, says Ponzo. "Only then will they be able to squeeze every last bit of latency out of the process and out of the network. "Then colocation may shave off five or ten milliseconds but right now we are in the realms of microseconds.
" According to HSBC's Bourne, performance for performance's sake no longer gives an advantage. "Everyone is focusing on how quickly they can get from their trading desk to the exchange but not so many are focusing on how quickly they can decide whether they should go to there in the first instance. If I can get to a decision to trade one second ahead of you but you are 50 milliseconds ahead of me in your route to the exchange, I will still get to the exchange first, by a significant amount."
Kevin Covington, head of new product development at network provider BT Radianz believes it is unquestionable that firms have been looking at latency in isolation and looking only at the connectivity and route to execution. "It is only recently that there has been a realisation that applications can cause a few bottlenecks as well and over the last four to five months the more advanced organisations have been looking at the entire process."
It is no surprise then that Covington says that BT Radianz's proposition is all about addressing end to end latency rather than simply physical colocation. "We include the communications, the connectivity, the network infrastructure and the service levels. The service level is important because customers want it to be fixed and based on an agreed latency rate of specific milliseconds and for it to be managed over a long duration."
Covington also says that most firms are looking to deploy colocation in multiple venues rather than single exchanges. "For a firm based in London, it would be looking to deploy colocation in more remote locations rather than on their own doorstep so they are looking for a single service wrap from us to use in these places they cannot reach themselves."
The idea of firms using a single service for low latency trading in as many as 10 locations is a pertinent one with the implementation of MiFID just round the corner, even if the Fragmentation of liquidity and the creation of multi-lateral trading facilities would have happened anyway without MiFID - after all, an Order Book is not the answer to everything. What this does mean, though, is that it is likely there will no longer be single execution venues that hold a virtual monopoly on a particular asset class, which brings into question the wisdom of spending money on getting as close as possible to a single venue. Balarkas says that Credit Suisse's focus has been on ensuring they are ultra-fast in multiple venues and then finding the best venue to execute. "Doing that within 10 milliseconds is more important than blasting an order into one venue in six milliseconds and putting it on a marketing sheet," he says.
Aggressive marketing, it is felt, has been the dominant feature of the colocation industry. The speed of light cannot travel as fast as some colocation services are claiming. It has become the latest bells and whistles feature for a firm to have even if it brings no pecuniary advantage. But the expense is necessary because it is the cost of having people to do business with you, say those who have begrudgingly forked out. There is the perception that if you have the ability to get an order from point A to point B at the speed of light then you must be great at everything else too…