Loading...
Cut the Chit-Chat: A New Framework for the Application of Generative Language Models for Portfolio Construction
Registration
Registration is now closed (this event already took place).
Details
The paper covers a new approach to using generative language models (GLMs) in portfolio construction. While current applications often rely on chat-based forecasts and classify model outputs into simple sentiment labels (e.g., positive, neutral, negative), this method overlooks the strength or intensity of the sentiment expressed. The paper introduces Logit Extraction, a technique that captures the model's internal probability estimates for each sentiment label. By doing so, it enables the creation of a continuous ranking variable, which proves more effective for cross-sectional portfolio construction than traditional discrete-label approaches. The results show that Logit Extraction significantly improves risk-adjusted returns. An open-source Python package, TokenProbs, accompanies the paper to support further exploration and use of the method.
If you're interested in the intersection of AI and finance, if you've attended any of our Quant Finance Bootcamp series, or if you're looking to explore cutting-edge methods for portfolio construction, please join us!
See the paper here.