BST Panel: Algo Performance More Important Than Speed

Asset managers are more worried about the alpha performance of the algorithms they deploy than their speed, according to panelists speaking at yesterday's Buy-Side Technology European Summit, hosted in the suave surroundings of the Royal Horseguards Hotel in Westminster, London.

Moderator Peter van Kleef, managing director of inter-dealer broker Lakeview Capital Market Services, was joined on the panel by Neil J. Smith, vice president of State Street Global Advisors (SSGA); Mark Kontkowski, director at London-based investment manager, AHL; and Brendan McCarthy, director of global financial services firm, Knight Capital Group.

The panelists agreed that performance was of utmost importance to buy-side firms, saying asset managers tended to build algorithms themselves in the few instances that speed was considered of greater importance to them than performance.

"For some of the guys, speed is definitively important, but by and large on the buy side with the asset management community, they would give up speed for a couple of basis points of performance," said McCarthy. "What we are seeing across all clients, is a more focused approach to performance. People don't want to hear qualitative arguments for using a broker anymore ─ they want you to come into their office and justify quantitatively how your algos are performing. They want full disclosure in terms of what you're doing, how you're doing it, where the orders are routed, and they need to see exhaustive research and performance analysis to justify trading with you."

Understanding algorithms is the key to success, said AHL's Kontkowski. "You've got to know what the algo is doing ─ you can't just use the algo blindly and hope for the best. You've got to understand the algo, what it's supposed to do, and that not only helps you in your transaction-cost analysis and post-trade analysis, but it also lets you determine whether it did what it's supposed to do. That's your quick exit ─ if it didn't do what it's supposed to do then you'd reject that algo. You wouldn't reject an algo based on six trades that didn't perform well ─ you'd need a far larger data set than that."

By and large, the asset management community would give up speed for a couple of basis points of performance.

The panelists agreed that a one-month period is the optimum time for a proper evaluation of an algorithm. "You're looking at a month's data if something's not working. Doing it over a long period builds a picture that cannot lie," said SSGA's Smith.