This paper explores the nature of financial sentiment and how large language models (LLMs) contribute to its estimation. The authors trace the evolution of sentiment measurement in finance, from market-based and lexicon-based methods to advanced natural language processing techniques. They argue that financial sentiment is a complex, context-dependent concept that goes beyond simple polarity, requiring nuanced understanding of financial jargon, market conditions, and investor behavior. The emergence of LLMs has significantly enhanced sentiment analysis by providing deeper contextual understanding and greater accuracy in extracting sentiment from financial text.

The paper provides a comparative analysis of two major LLM architectures: BERT-based models (such as RoBERTa and FinBERT) and GPT-based models (including GPT-4, OPT, and LLaMA). BERT-based models, with their bidirectional attention, are optimized for structured sentiment classification tasks, capturing full context from both directions. In contrast, GPT-based models, using autoregressive generation, excel in financial text generation and real-time sentiment interpretation, making them suitable for dynamic market analysis. The study highlights how these models are applied in investor sentiment analysis, algorithmic trading, and financial decision-making.

The significance of this work lies in its conceptual clarification and practical guidance for leveraging LLMs in financial sentiment analysis. By exploring what financial sentiment is and how it is estimated within LLMs, the paper provides insights into the growing role of AI-driven sentiment analysis in finance. It underscores the importance of choosing the right model architecture based on the specific task—classification versus generation—and sets the stage for future research in this rapidly evolving field.