Bridging Human Expertise and AI: Evaluating the Role of Large Language Models in Retail Investors' Decision-Making

Bridging Human Expertise and AI: Evaluating the Role of Large Language Models in Retail Investors' Decision-Making

ABSTRACT

This study investigates the role of large language models in retail investors' decision-making processes from the perspective of the Theory of Planned Behaviour. It explores whether large language models can replace or change the role of financial experts and whether introducing large language models may lead to more informed retail investors' decisions. Qualitative interviews were conducted with experienced retail investors. Secondary data were gathered from YouTube recordings. Thematic analysis and Retrieval-Augmented Generation methodology was used for data extraction and analysis of the scripts. The findings indicate that while large language models have the potential to enhance accessibility to expert opinions and provide more informed investment decisions, they are unlikely to replace human experts. Retail investors show a preference for combining large language models insights with human expertise, recognising the limitations of large language models in managing complex and nuanced investment information. The study highlights the usefulness of the Theory of Planned Behaviour as a framework for the exploration of the topic. It also introduces a novel research method - advanced data extraction techniques on a vast unstructured dataset. Results contribute to the understanding of large language models potential in supporting retail investors and confirms the usefulness on the AESTIMA tool for data extraction and analysis.

Article history:

Introduction

Retail and individual investors' decisions to buy or sell financial market instruments are influenced by opinions from a range of sources, including information from financial experts, brokers, big investors, friends, and family, and expert opinions disseminated through media, including social media. Some investors, however, prefer to base their decisions on financial and fundamental analysis. Reliance on experts' opinions can create biases such as anchoring, where investors adjust their portfolios based on those opinions rather than an independent analysis, especially during times of market optimism. Biased, sensationalised news or viral social media posts can trigger emotional reactions and cause stock price volatility, leading to emotional decisions rather than in-depth analysis of data, resulting in impulsive investment choices. The fake or poorly verified news can lead to market instability, investors' losses and to prices of assets reaching a different value to their true value.

Large language models can be transformative tools in the finance sector, offering a range of applications that enhance efficiency, accuracy, and decision-making processes. Large language models can extract detailed insights from financial documents, conduct sentiment analysis, forecast market trends, provide a comprehensive assessment of financial health and market conditions to help investors and financial institutions make informed decisions. Large language models can also assist in interpreting complex regulatory documents, supporting compliance with legal requirements.

Large language models can support risk assessment and fraud detection due to their ability to identify patterns and anomalies in large sets of financial data, highlighting possible fraudulent activities, which is especially important in loan approval or money laundering detection. Large language models can optimise investment strategies by analysing market data and generating actionable insights. Their capability to understand natural language instructions may be particularly useful in customer service.

Despite the potential of large language models, there is still a lack of knowledge on their role in investment decisions. Investors need to critically evaluate the sources of information and consider a balanced approach that includes both external advice and independent analysis. The potential of large language models is still evolving, and in future, they might replace humans in some areas of professional practice. Understanding the motivations behind the introduction or resistance to introducing this technology into retail investors' practice is sparse. Gaining better insight into the issue would guide the development of large language models based tools for the industry and formulate strategies for their implementation. This research aims to investigate the role of large language models in the decision-making process of retail investors through answering two research questions: One. Are large language models seen as a potential replacement of human financial experts, or rather will the experts' role evolve alongside large language models introduction to the industry? Two. Does large language models use improve decision-making by enabling more informed, rational investments?

The second objective of this research is to validate the quality of data extraction and analysis from unstructured text sources, such as YouTube recordings scripts, using the Retrieval-Augmented Generation; using AESTIMA tool. This objective supports the primary research by ensuring that the data used to analyse the impact of large language models on retail investors' behaviour was robust.