WallachBeth Cap Applies ‘Deep Learning' to Trading

WallachBeth Capital currently provides a range of bespoke agency-only services for developing and implementing trading strategies focusing on five main business areas: exchange-traded fund and portfolio trading, equity derivatives, electronic trading solutions, fixed income and research. The broker-dealer is now building deep learning into its products, or embedding it directly into its algorithms to help customers generate alpha.

To generate its deep learning analytics, WallachBeth sources data from a mixture of publicly-available market data feeds and unstructured data, such as news-based sentiment. The firm then provides bespoke reports, which can be incorporated into customer systems or traded electronically via its algorithms, as well as a strategic consulting service, dubbed Scout, during which it leverages the deep learning to help clients develop strategies for generating alpha, optimizing portfolios, managing market risk and creating optimized trade execution strategies.

"Because of the bespoke nature of the business, we can spend time talking about these groundbreaking analytics and how they can be used to invest and create products.... A lot of alpha is lost when you simply hand this to a buy-side trader rather than incorporating it efficiently," says Matthew Rowley, chief technology officer at WallachBeth.

For example, the firm has developed predictive pricing analytics that can provide customers with predictions for the percentage change in ETF prices between specific points in time. "We can predict whether the market will go up or down in the next minute, and we tend to be correct 77 percent of the time," Rowley says. "The whole thing is almost like a social experiment. We are always experimenting, and the deep learning field is evolving fast. The point is, we are using it for real decisions and putting it into algorithms today with success, and we are looking forward to expanding it across all of our products," he adds.

Another area where the broker plans to deploy deep learning techniques in the future is longer-term alpha generation. Rowley says portfolio managers will be able to use deep learning analytics for checking and scaling, reorganizing portfolios and generating investment predictions. "We do all of this today, but it's about more sophisticated techniques to supplement traditional techniques for alpha generation in the longer term," he adds.

WallachBeth has been "pleasantly surprised" by the response from customers across the board, from larger institutions to smaller hedge funds, Rowley says, adding that the firm is leveling the playing field by developing the analytics and applying them to real trading scenarios. "Because of some advances in this world of deep learning, we are able to create a mini-brain to extract rules.... Traders don't always know why they do what they do...but gradually the computer learns rules itself from the data," he says.