Evaluating LLM’s Understanding of Portfolio Theory: Risk Profiles and Optimal Portfolios
- 주제(키워드) Large language models , Benchmark dataset , Portfolio theory , Asset allocation , Risk profile
- 발행기관 고려대학교 기술경영전문대학원
- 지도교수 Jang Ho Kim
- 발행년도 2025
- 학위수여년월 2025. 8
- 학위명 석사
- 학과 및 전공 기술경영전문대학원 기술경영학과
- 원문페이지 77 p
- 실제URI http://www.dcollection.net/handler/korea/000000305625
- UCI I804:11009-000000305625
- DOI 10.23186/korea.000000305625.11009.0003240
- 본문언어 영어
초록/요약
This study designs a benchmark dataset based on portfolio theory to evaluate the investment decision-making capabilities of major large language models (LLMs) such as GPT-4, Gemini 1.5 Pro, and LLaMA 3.1-70B. The benchmark consists of multiple-choice questions derived from mathematically well-defined objective functions, allowing for the comparison of LLMs’selection accuracy across various investment goals and constraints. Through this framework, we quantitatively assess how rationally these LLMs can make investment decisions grounded in portfolio theory. Experimental results show that GPT achieves the highest accuracy on risk-based objective functions and maintains robust performance even under complex constraints. In contrast, Gemini demonstrates strength in return-focused tasks but exhibits significant performance degradation as the problem structure becomes more complex. LLaMA consistently shows the lowest accuracy across all scenarios. Furthermore, analysis of the inherent investment risk preferences of each model revealed clear differences between models and varying levels of sensitivity to the assigned hypothetical investor profiles.
more목차
ABSTRACT....................................................................................................ⅰ
초록 .................................................................................................................ⅲ
TABLE OF CONTENTS ................................................................................ⅴ
LIST OF TABLES...........................................................................................ⅶ
LIST OF FIGURES.........................................................................................ⅸ
CHAPTER 1. INTRODUCTION .....................................................................1
1.1 Research Background..............................................................................1
1.2 Research Objectives................................................................................2
1.2.1 Risk Profiles.....................................................................................2
1.2.2 Optimal Portfolios............................................................................3
CHAPTER 2. LITERATURE REVIEW...........................................................5
2.1 Portfolio Optimization.............................................................................5
2.1.1 Minimize Volatility..........................................................................6
2.1.2 Maximize Return..............................................................................6
2.1.3 Maximize Sharpe ratio .....................................................................7
2.1.4 Minimize Maximum Drawdown......................................................8
2.1.5 Minimize Conditional Value-at-Risk ...............................................9
2.2 Benchmark Dataset ...............................................................................12
2.2.1 General Benchmark Dataset...........................................................12
2.2.2 Financial Benchmark Dataset.........................................................13
CHAPTER 3. METHODOLOGY: RISK PROFILES....................................16
CHAPTER 4. METHODOLOGY: OPTIMAL PORTFOLIOS......................19
CHAPTER 5. RESULTS: RISK PROFILES..................................................27
5.1 Investor Profile of LLMs .....................................................................27
5.2 Assigning Profile of LLMs ...................................................................30
CHAPTER 6. RESULTS: OPTIMAL PORTFOLIOS ...................................44
6.1 Results by Investment Objectives.........................................................44
6.2 Results by Constraint Types..................................................................46
6.3 Results Analysis....................................................................................49
CHAPTER 7. CONCLUSION........................................................................58
REFERENCES................................................................................................61
