It's Not Always About Pure Performance, Usability is Key for Trading Technology, Agrees BTLLT Panel

Building a low latency trading architecture is not just about high performance and speed, it is also about usability, agreed panellists debating ‘Deploy Fast, And Build to Last’ at this week’s Business & Technology of Low-Latency Trading (BTLLT) event in London. However, there was some debate between panellists from TraderServe and Sybase about the shelf life of an algorithm: figures ranged from four days to four years.

Sybase’s business development manager for EMEA Stuart Grant contended that algos are usually effective for around three weeks, or even sometimes only four days, before they must be retired. He suggested that this was one of the reasons why trading infrastructures have become so siloed and costly to maintain. “The faster you can adapt and develop algos, and reuse existing information on risk management side to this end, the better,” he said.

TraderServe’s technical director Nick Idelson, on the other hand, argued that some of his firm’s clients have been using the same algos for around four or five years. He indicated that this has been achieved “by using models that can be tweaked over time, rather than throwing algos away.”

This flexibility to alter algos in order to react to market developments is key, he said. However, both agreed that in terms of technology and investment, it is about firms picking their battles wisely. “Only do in-house development where really need to,” said Grant. “Vendors have a lot to offer as long as you define the benchmarks that you want to achieve.”

Louis Lovas, director of solutions at OneMarketData, added: “Hedge funds have limited resources and don't want to fall into sinkhole of software development. But you need to do your vendor homework first.”

A-Team’s own Pete Harris also provided a bit of background in terms of recent trends within the market and the requirement for firms to meet the “constant barrage of regulatory changes”. The technology underlying trading systems architecture therefore has to be sufficiently flexible and agile in order to meet both regulatory requirements and changing business needs, such as new data feeds, customers and technology upgrades.

The “good news”, however, is that technology such as that underlying complex event processing (CEP), pre-integrated stacks and tick databases has been developed to meet some of these needs (check out Pete’s commentary on the subject in his regular blogs here). Lovas, for instance, noted that his firm is focusing on the CEP and tick data space: “We are looking to empower quants by providing trading tools that enable them to deploy the strategies required by the market. The speed in which firms can respond to market developments is key.”