Knight Capital Sets the Record Straight on Algo Testing
Should something appear off, they would shut down the test, McCarthy says, adding that that's never actually happened. If everything goes OK with the limited testing and no changes are needed, they widen the scope of testing, he says.
"Maybe we run a larger order in a single name and then do some further analysis of what actually happened, what was expected to happen and just kind of begin to gradually expand," McCarthy says of the process.
When the firm is able to comfortably run multiple orders in a variety of securities across different marketplaces, and everything functions as intended, Knight then gets in touch with clients who are open to being beta users of a new algorithm, McCarthy says. Knight's internal trading desk can be among the beta testers and larger clients that are willing to run initial strategies with the understanding that Knight Capital will do a thorough review of their order placement, he says. "Everything is overly supervised, I guess is the easy way to describe it."
When these steps are completed, the algorithm is deployed as a new strategy, with the development team still performing real-time reviews of how it's functioning, McCarthy says. Once clients are using the algorithm as a strategy, McCarthy's team runs a algorithmic performance review, where they examine the strategy's real-time, end-of-day and end-of-month statistics.
[Are Algos Phasing Out Humans on Credit Desks?]
"That's one of the best ways to spot any type of aberration," he explains. "And then on top of all of that, we have client feedback."
Knight's message on its testing practices should play well to both regulators and the market following a year where public trust in the stock markets endured several staggering blows, all attributable to technology.
Knight Capital's Testing Regime
1. Initial development phase with no market interaction.
2. Production test phase at a limited range from the trading desk. New strategy could be tested on one order in one symbol.
3. Wider testing, possibly of a larger order in a single stock and then further analysis of what happened vs. what was expected.
4. Once the strategy can conduct multiple orders in a variety of securities across different markets, potential beta users are recruited.
5. When the algorithm has passed all tests, it's deployed as a new strategy and a performance review is done.
As the Senior Editor of Advanced Trading, Justin Grant plays a key role in steering the magazine's coverage of the latest issues affecting the buy-side trading community. Since joining Advanced Trading in 2010, Grant's news analysis has touched on everything from the latest ... View Full Bio
Tweet This