Large Language Models are Intelligent Traders

Abstract

This paper introduces LLM-Traders, a novel approach to analyzing financial time series (FTS) through the use of fine-tuned large language models (LLMs) and prompt engineering. The research addresses three primary challenges inherent in FTS analysis: (1) pervasive noise, (2) complex, diverse range of models, and (3) constantly evolving dynamics. Our methodology initially concentrates on reducing overfitting, a prevalent issue caused by noisy data. This is achieved by meticulously fine-tuning the LLMs to recognize and interpret the unique attributes of FTS. Subsequently, we implement strategic prompt engineering within these models. This strategy enables effective navigation and adaptation to the multifaceted nature of FTS and accommodates the wide array of existing models. To adapt to the dynamic nature of FTS, we propose an innovative dynamic ensemble method. This approach combines multiple prompt responses in a synergistic manner, enhancing the versatility and accuracy of the analysis. Overall, our integrated approach provides a comprehensive, robust, and flexible framework for addressing the complexities of modern FTS analysis.