LLMs Analyzing the Analysts: Do BERT and GPT Extract More Value from Financial Analyst Reports?

• DOI: 10.1145/3604237.3627721
• Publication Date: 2023-11-25T23:09:47Z
• Authors: Seonmi Kim; Seyoung Kim; Yejin Kim; Junpyo Park; Seongjin Kim; Moolkyeol Kim; Chang Hwan Sung; Joohwan Hong; Yongjae Lee
• Keywords: Investment; Sentiment Analysis; Code (set theory)

This paper examines the use of Large Language Models (LLMs), specifically BERT-based models and GPT-3.5, in sentiment analysis Korean financial analyst reports. Due to specialized language these reports, traditional natural processing techniques often prove insufficient, making LLMs a better alternative. These are capable understanding complexity subtlety language, allowing for more nuanced interpretation data. We focus our study on extraction scores from using them construct test investment strategies. Given that reports present unique linguistic challenges significant 'buy' recommendation bias, we employ fine-tuned texts. The aim this is investigate compare effectiveness enhancing subsequently utilize strategies, thereby evaluating models' potential extracting valuable insights code available at https://github.com/msraask3.