Introduction

Supervision

• Jointly between MUFG1 and Surrey University
• MUFG: Dr Chris Kenyon, chris.kenyon@mufgsecurities.com
• Weekly progress calls

Project Parameters

• Prerequisites:
  • Good knowledge of Python
  • No knowledge of SWOT or investment analysis required, but an interest in finance/investing is beneficial
• More than one student can work on this project, but results must be written up separately
• Source code must be put on github.com
• Write-ups must use Overleaf

Objective

The aim of this project is to build methods for validating the scoring of massive language models (MLM) in the context of SWOT portfolio analysis based on publicly available documents following (Meskovskis and Kenyon 2024). In SWOT analysis (Meskovskis and Kenyon 2024), each strength, weakness, opportunity, and threat are scored on a low/medium/high scale by ChatPDF. The objective of this project is to validate the reasoning behind the scoring (low, medium, high) by ChatPDF of each strength, weakness, opportunity, and threat.

Methods

The student is free to use any method that provides numerical validation of ChatPDF scoring. Here we provide a suggested route:

• In SWOT analysis (Meskovskis and Kenyon 2024), each strength, weakness, opportunity, and threat are scored on a low/medium/high scale by ChatPDF.
• ChatPDF can be asked why a score was given, i.e., what makes the score high, medium, or low. ChatPDF will provide an answer.
• The task is then to analyze the answer to assign a score as to how valid the answer from ChatPDF is. It is expected that there are only a few rationales behind all of the scoring, e.g., X is critical to Y. It is critical to differentiate different scores.
• In some cases, it may be necessary to ask ChatPDF more questions to provide a thorough understanding of ChatPDF’s reasoning.

Success Criteria

The student must:

• Place their work in the context of current literature
• Provide a mathematical framework for the question(s) they address and justify their choices and construction
• Provide numerical examples
• Conclude by summarizing results and pointing to directions where the work could be extended

References

Meskovskis, Arturs, and Chris Kenyon. 2024. “SWOT Portfolio Construction and Validation Using Massive Language Models.” Available at SSRN 4784515.