Utilizing Artificial Intelligence for Institutional Trading

There are many uses for artificial intelligence (AI), and the promise of what's to come is both exhilarating and scary. But before it becomes the source of a dystopian future, AI is being harnessed to help develop smarter techniques and predictive analytics for institutional trading.

Portware, a developer of automated trading solutions, and which was acquired by FactSet this past October, has been using AI and machine learning techniques since 2007 for trade execution services, according to Henri Waelbroeck, director of research at Portware.

"I think the basic concept is trading desks produce a lot of data and machine learning techniques can harvest this data to avoid repeating costly mistakes, by providing intelligent decisions at the right moment," he says.

Institutional trading can utilize AI to look at new orders that have just arrived at the trading desk and compare them to orders from the past and how it fits into the portfolio manager's order creation history.

"If you look at how technology works today, it really looks at data mostly for what it says about the past or in some cases just focuses on displaying the present in a user interface, but the point is, the decisions we make don't play themselves out in the past, they play out in the future," Waelbroeck says. "So to make intelligent decisions we really need to look at data not so much for what it says for the past or present, but what it says about the future."

The Limits of AI

Although AI is becoming more commonly used to extract meaning from data, it's still important to note that it still has a long way to go and that ultimately, traders are not replaceable. "You have to keep the traders engaged and technology is there to display to the traders what we think the optimal execution plan is and why. This way, the traders can evaluate this information and make their own decisions," Waelbroeck says.

The increased use of artificial intelligence in leveraging the power of data still has much fruit to bear and data is used abundantly in back-testing alpha signals.

"I think there is a lot of room to better harvest data in risk modeling, such as for portfolio optimization," he says. "Once, you get to that level of technology you do, of course, offer the ability to automate trading, and some traders find a segment of their flow is suitable for completely automatic execution where they don't actually see what the machine is thinking but they let it run. I think the service that is built entirely as a black-box automation service is unlikely to be the long-term solution. In our business, we need to keep the traders engaged in the process."

In trade execution, AI is becoming increasingly useful, Waelbroeck adds, as it's becoming clear that the institutional trading desks have a fiduciary obligation to use their data to optimize execution efficiency. "This is a trend that's developed not just in our industry, but in all industries to some extent," he says.