How LLMs are already being used to replace human traders
It's no secret that financial services companies are experimenting with AI technologies to enhance their trading performances, but it's still early days. Nonetheless, there are areas of promise, as illustrated by a recent paper from Columbia University researchers.
Click here to follow our new WhatsApp channel, and get instant news updates straight to your phone 📱
The paper, 'Large Language Model Agent in Financial Trading*', says that there are two broad ways in which LLMs have been tested in a trading capacity: as a 'trader' or as an 'alpha miner'.
As traders, LLMs operate in a variety of ways using different machine learning techniques. 'News-driven' agents will analyze swathes of news data to create an informed decision on whether to buy, sell or hold a stock. LLMs, naturally, are adept at analyzing textual data, from a variety of sources, with social media data from the likes of Twitter and Reddit being "an under studied field with great potential." Numerical and visual data isn't as reliable as textual data when analyzed by LLMs, but significant progress has been made in both areas.
Other trading agents use reinforcement learning, a trial-and-error focused technique appreciated by Goldman Sachs. These agents "have proven effective in aligning LLM outputs with expected behaviours," but a lack of high quality training data has proved a concern. Back testing can be a solution, but for markets with less available data, synthetic data may be the only solution.
Also with trading, 'Debate-driven' agents are more of an ecosystem. They comprise a number of different AI agents, each looking at different things such as mood, rhetoric and dependency, which will then debate each other in order to ensure the agent's final trading decision is well-reasoned and robust.
'Reflection-driven' agents are similar to news-driven ones, with the key difference being that data is summarized into "memories" upon discovery. These memories are then recalled and analyzed to create trading insights.
When it comes to alpha mining, LLMs function differently. A so-called "writer agent" will receive a 'general idea from a human trader' which it will use to generate a script for its implementation. Then, a "judge agent" will provide feedback and refine the loop. The code is then tested in a real world market, and the results are used to refine the script further. Alpha generation is a "resource intensive" job, so this application will be an area of great promise for firms looking to cut costs.
*Large Language Model Agent in Financial Trading: A Survey (arxiv.org) - Han Ding, Yinheng Li, Junhao Wang, Hang Chen
Have a confidential story, tip, or comment you’d like to share? Contact: Telegram: @AlexMcMurray, WhatsApp: (+1 269 237 3950). Click here to fill in our anonymous form, or email editortips@efinancialcareers.com. Signal also available.
Bear with us if you leave a comment at the bottom of this article: all our comments are moderated by human beings. Sometimes these humans might be asleep, or away from their desks, so it may take a while for your comment to appear. Eventually it will – unless it’s offensive or libelous (in which case it won’t.)
