Sentimental Signals: an Exploration of Large Language Model Enhanced Bond Trading Strategies

Abstract

In this thesis, we explore the application of ChatGPT, a Large Language Model (LLM) developed by OpenAI, in fixed income trading, with a particular focus on treasury bonds. Pure numerical data does not always perfectly describe the market: the movement of different financial instruments within the market is determined by quantitative data as well as oral, textual, visual, and other forms of qualitative data. We aim to study how market interpretability and trading performance could be enhanced through the incorporation of macroeconomic sentiment analyzed by ChatGPT. By integrating ChatGPT's capabilities in processing textual news and quantitative data, we generate predictive signals and trade sizes to trade across the bond yield curve. Our results across six distinct market phases demonstrate ChatGPT's potential to refine momentum trading strategies, with particularly effective results by combining ChatGPT-generated trading signals with generated trading-sizes. This thesis ultimately shows the potential of ChatGPT as a tool in fixed income algorithmic trading, incorporating qualitative insights into quantitative models.