LET'S BE REASONABLE
What would be a reasonable time for a round-trip order?
"If we assume that the client is in the same town as the broker, and the SwiftNet hub sends a small, single order on a quiet market day and time to the broker for instant execution on the bid or offer of the local exchange, I would expect the total round-trip time to be anything between half a second to three seconds, depending on the client's chosen connectivity solution," Wildenberg says.
The real issue is not whether one second is fast enough, but how fast, on a busy day, can SwiftNet process 1,000 messages, says Wildenberg. "Will 1,000 messages be processed in one second, or one second at a time?" he asks.
According to the investment-banking source, Swift's new Internet-Protocol (IP)-based network slows down the orders because Swift performs validation checks on the FIX messages.
Networks such as Bloomberg, Reuters, and Thomson's TradeRoute (which is being renamed Autex Order Routing), that are also hub-and-spoke networks, do not have this problem, says the source.
The advantage of a hub-and-spoke network is that brokers can maintain one connection to Swift's FIX server, which in turn manages all the FIX connections to buy-side counter-parties.
Swift also offers guaranteed message delivery, says the source, which he contends is unnecessary because the session layer within the FIX protocol does this already. "The session layer sends out a heartbeat or pulse to all your destinations, and that's how the message layer knows it has connectivity," says Ignatius John, senior systems and technology analyst at AXA Rosenberg Investment Management, based in Orinda, Calif. John, who is a member of the FIX Protocol Limited (FPL) equity committee, says no other validation is necessary.
"The protocol itself has so many checks within the message that if (the message) doesn't get there, we would know because a flag would be turned on," he says.
Several investment banks have asked Swift management to remove these extra components from the process which they believe are irrelevant to FIX messaging. But Swift won't budge, the source contends.
"We have slightly higher latencies than network providers do because we do more with the message. We add value," says Fox. Order-execution reports are considered high-value messages as opposed to IOIs, he explains. "If you look at an order to buy millions of shares of a security, we use non-repudiation," which means "that a message is subject to a lot more security going across our network," says Fox.
If there is a delay, AXA Rosenberg's John agrees,"Eight seconds is not acceptable. I would think that a complete round-trip would be two to three seconds. More than anything else, I would like my order to get to the floor instantaneously," he says.
When John refers to round-trip, he is speaking of sending an order to an electronic-communications network (ECN) or some automated electronic-execution system. One of the advantages of using SwiftNet is its plan to offer an automated FIX-certification process. Since AXA deals with 65 different brokers, instead of testing with each broker, it will certify once with Swift.
But John says the speed issue "is all theoretical now." That's because his firm plans to beta test FIX over SwiftNet to figure out the throughput speeds. "The proof is in the pudding," says John, who's expecting to begin beta tests in the fourth quarter and to do heavy testing early next year. "We'd like to send some large numbers of orders (to a broker) just to see how fast the network works. We will be able to determine that."
Right now, AXA uses Transaction Network Services (TNS) to communicate with brokers and ECNs and is very pleased with it, says John. TNS is a point-to-point network that requires brokers and buy-side firms to manage their own FIX sessions to each counterparty.
Some customers of Swift are not worried about the speed. "I don't think speed has been the most important factor in this," says Erik Einerth, head of trading at Enskilda Securities, a Swedish broker/dealer that agreed to be part of SwiftNet's early-adopter program.
The firm is waiting to go live in Europe and is interested in the efficiency of moving customer order flow into settlement instructions. "We use it for order routing and then we put (the orders) on an exchange," he says.
If an execution takes five to 10 seconds, it doesn't bother Einerth. "We could take up to 15 or 20 seconds to sort it out. It's not a huge market delay," says the head trader. If it takes a longer amount of time, he says, Enskilda "would more or less guarantee that a limit order would be filled at that price if it could not be executed instantly. Those orders will not be mishandled in any way."
Performance, however, may be a red herring, since it's not clear how much traffic is flowing over it and who is using it.
All of the firms WS&T contacted in late September, including Enskilda, AXA and Loomis Sayles, were not yet in production with the new IP-network for sending trades. "I actually don't believe that the SwiftNet network is getting any major level of traffic, and therefore its performance is unknown," says Wildenberg.
Fox insists, "There is a lot of traffic on the FIX service now," citing millions of messages comprised of orders, execution reports, allocations, etc., going across Swift's production network with a number of customers. Though he says he is not at liberty to share all their names, he points to Gartmore Investment Management and Dresdner Kleinwort Wasserstein, which began sending and receiving live allocation messages last December.
Chris Sims, Gartmore's London-based head of Information-Systems development, says the firm turned the allocations feature off because it was using a test infrastructure. It may turn allocations back on when it converts to the new Swift-IP infrastructure. "We're converting from X.25 to SwiftNet FIX IP network on Oct. 4 (for settlement instructions)," says Sims. "Once we've got to that stage, we will look at turning allocations over SwiftNet."
Fox expects the bulk of traffic will be orders and execution reports. "I think it's actually an exception that some of the early trading on the network is allocations," he says. "I think that what you will hear from customers is that they absolutely intend to use it for trading."
Fox admits a point-to-point session (like a TNS) would be better suited for "someone doing statistical arbitrage or very high levels of program trading where it's dependent on milliseconds for the quality of that execution." But, he says, certain types of trading are suited for a hub-based network such as SwiftNet. "I think your traditional buy-side to sell-side relationships are well suited to going over a hub."
Wildenberg says a client that wants high speed is typically not going to choose a hub network like Swift or Thomson. "I'm either going to suggest he moves his office next to mine, or I'm going to make sure he implements direct point-to-point connectivity." But the trading expert believes, "Swift will have all the credentials to be a player in that space."