Researchers from the University of Florida recently conducted a study to evaluate the predictive power of ChatGPT, a language model that can parse news headlines, and compare it to basic models like GPT-1, GPT-2, and BERT for predicting stock market returns. To their surprise, the researchers found that ChatGPT did a better job predicting the direction of returns than random guessing, leaving investors wondering why Wall Street hasn’t jumped on the AI train yet.
The ability to predict stock market returns is a relatively new capacity for language models and the study’s results suggest that incorporating ChatGPT or similar models into the decision-making process may improve the accuracy of quantitative trading strategies. This prompted Bloomberg to launch BloombergGPT, a new GPT-based language model trained on the FinPile dataset consisting of financial documents, news, filings, press releases, and social media. In theory, BloombergGPT should generally improve existing natural language processing tasks.
The study’s findings were met with some skepticism by Wall Street, who has yet to make use of AI models like ChatGPT in their trading operations. Investment management companies that rely on computer models for tasks such as trading and risk management have done so for decades. However, financial AI still has yet to replace traditional stock market analysis because AI has yet to outperform it.
Jonathan Larkin, a managing director from Columbia Investment Management Co, was quoted in The Wall Street Journal as saying, “Progress in applying AI to investing has been limited, though innovations in language modeling could change that in the years ahead.”
Bloomberg, founded in 1981 by Michael Bloomberg, is a well-established financial, media, and technology company. Bloomberg’s portfolio of services includes a stock and commodities exchange, financial website, and a newswire service suitable for international media organizations. Bloomberg also has the Bloomberg Terminal, a subscription-based service that uses natural language processing and machine learning models to provide financial data.
Jonathan Larkin is a managing director with Columbia Investment Management Co, where he is responsible for managing the $13 billion endowment fund for Columbia University. Larkin has over 18 years of experience in the financial industry and is an experienced investor and portfolio manager, aiming for long-term investments for his customers. He is also a guest lecturer at Columbia Business School, teaching finance and asset management to MBA students.
