The leading knowledge platform for the financial technology industry
The leading knowledge platform for the financial technology industry

A-Team Insight Blogs

How ‘Deep Learning’ Could Make A Deep Impact On Trading

As a machine learning technique, “deep learning” has evolved enough to be useful for trading operations, according to Elliot Noma, managing director at Garret Asset Management. Noma will be moderating a panel on the uses and limitations of the technique, at the Intelligent Trading Summit in New York on June 8.

Deep learning expands on neural nets, which simulate the levels of communication within the human brain — the same neural communications that lead to the decisions comprising consciousness, Noma explains. Neural nets previously only had one or two layers, while deep learning-capable neural nets can have as many as 100 layers, which make these networks better suited for working on large sets of data, he adds.

“The error rates for neural nets on classifying images had been around 30 percent,” says Noma. “Over the past few years, using the technology for deep learning, the error rates have come down to the same as human beings — no more than 5 percent.”

With each layer in a deep-learning network containing hundreds of simulated neurons, and 100 or more layers possible, such a network can “assess large amounts of data, and be trained on multiple different types of data sets,” says Noma. “Different results from different models can be connected together.”

New data sets keep arriving, including Twitter feeds, sentiment analysis, political and government statements, satellite data and other social media information. Deep learning can analyse all of these data sets, and compare the resulting analyses. Deep learning can also add analyses into multiple models that a firm is using.

“The key terms are boosting, bagging and stacking, which allow you take different large data sets, combine them in different ways, combine the analyses in different ways and adjust the analyses so if a previous analysis has mistakes, the neural nets catch and correct those mistakes,” says Noma.

For trading operations, deep learning networks can back-test new data sets and examine how they fit among all the available data. “You used to have to hire an analyst or assign an analyst to learn about the data, understand how to clean the data, and understand how the data fits with other data sets,” he says.

However, trading operations managers must put some guidance and care into implementation of deep learning technology, Noma explains. “With any powerful technique, you must have some idea of what it can do and what its limitations are,” he says. “You need access to someone who has that experience, whether that’s homegrown or external, to understand what the appropriate applications are for variations.”

Related content

WEBINAR

Recorded Webinar: How to turn data lineage from a regulatory response into a business advantage

Regulatory initiatives increasingly require firms to put in place robust data lineage frameworks to aid in understanding the workings behind reported values. But data lineage can add business value beyond regulatory compliance. As they move to automate data lineage processes by incorporating metadata management into their frameworks, firms can start to benefit from enhanced data...

BLOG

Refinitiv Teams with MarketPsych to Launch ESG Analytics

Continuing its focus on adding capabilities with an ESG theme, Refinitiv has teamed up with Los Angeles-based MarketPsych to launch Refinitiv MarketPsych ESG Analytics to monitor perceptions of sustainability and ESG risk. The analytical tool draws upon news and social media monitoring to offer analysts and portfolio managers numerical insights into companies and countries of...

EVENT

Data Management Summit London

The Data Management Summit Virtual explores how financial institutions are shifting from defensive to offensive data management strategies, to improve operational efficiency and revenue enhancing opportunities. We’ll be putting the business lens on data and deep diving into the data management capabilities needed to deliver on business outcomes.

GUIDE

Evaluated Pricing

Valuations and pricing teams are facing a much higher degree of scrutiny from both the regulatory community and the investor community in the glare of the post-crisis data transparency spotlight. Fair value price transparency requirements and the gradual move towards a more harmonised accounting standards environment is set within the context of the whole debate...