Year: 2024
Type: dissertation
Abstract: Investment banks’ research departments play a crucial role in facilitating securities transactions mainly through analytical reports, which face growing demand for quality and frequency due to investment democratization and competition. The objective of this master’s thesis was to survey and synthesize a framework for the qualitative aspects of equity research, as well as to subsequently select a use case from the framework to design, develop and test an end-to-end natural language processing (NLP) solution leveraging state-of-the-art large language models (LLMs). This thesis was commissioned by a bank. \n \nApplying the design science research (DSR) methodology, the study first mapped the data sources, fundamental analysis, reporting guidelines and narrative building deployed by sellside analysts by conducting a literature review and interviews. Secondly, a theoretical overview of text-to-text NLP was performed using recent literature. Lastly, these areas were combined to program a GPT-based technological artifact in an iterative manner with a use case to analyze risks from corporate reports of publicly listed Finnish companies. \n \nFindings of the study note qualitative analysis and storytelling as being relatively unexplored areas especially from a process management perspective, as a part of equity research largely relies on tacit and private knowledge as well as nuanced reasoning and ways of working. Nonetheless, an abundance of data with established structure exists in its written materials. Retrieval-augmented generation (RAG) technology stood out as a promising addition to domain-specific LLM document processing tasks based on evaluation results. Numerous opportunities for further research exist for this research problem as the banking industry and said technology are undergoing a major transition.
Source: LUTPub (LUT University)
Author: Laurila, Roope
Institution: Lappeenranta-Lahti University of Technology
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W6996990847