AI-Driven Defensive Sector Rotation ETF for the Vietnamese Stock Market : An Empirical Investigation of Machine Learning Applications in Emerging Market Portfolio Management
베트남 주식시장을 위한 AI 기반 방어적 섹터 로테이션 ETF
- 주제어 (키워드) Exchange-Traded Funds , Sector Rotation , Reinforcement Learning , Model Optimization , Vietnamese Market , Portfolio Management
- 발행기관 서강대학교 일반대학원
- 지도교수 정재식
- 발행년도 2025
- 학위수여년월 2025. 8
- 학위명 석사
- 학과 및 전공 일반대학원 경제학과
- 실제 URI http://www.dcollection.net/handler/sogang/000000081914
- UCI I804:11029-000000081914
- 본문언어 영어
- 저작권 서강대학교 논문은 저작권 보호를 받습니다.
초록 (요약문)
This study investigates the development and systematic optimization of an intel- ligent sector rotation Exchange-Traded Fund (ETF) for the Vietnamese stock mar- ket using reinforcement learning techniques. We implement a three-tier AI frame- work integrating market regime detection, LSTM forecasting, and Proximal Policy Optimization (PPO) for dynamic sector allocation. Through systematic model im- provement addressing reward function calibration, cash management optimization, and dividend adjustments, our framework demonstrates substantial performance en- hancement. The final optimized strategy achieves 68.6% total return with a Sharpe ratio of 2.142, representing a 12.8× improvement in returns and 64× improvement in risk-adjusted performance compared to the initial model (5.37% return, 0.0335 Sharpe ratio). While initially underperforming the VN-Index benchmark signifi- cantly, the improved strategy achieves competitive performance while maintaining superior defensive characteristics, with maximum drawdown limited to -10.0% ver- sus -68.32% for the market index. Our research demonstrates the critical importance of systematic model development and reward function engineering in financial rein- forcement learning applications. Keywords: Exchange-Traded Funds, Sector Rotation, Reinforcement Learning, Model Optimization, Vietnamese Market, Portfolio Management JEL Classification: G11, G15, C61
more목차
1 Introduction 1
2 Literature Review 3
2.1 Sector Rotation Strategies and Emerging Markets . . . . . . . . . . . . . . 3
2.2 Market Regime Detection Methodologies . . . . . . . . . . . . . . . . . . 4
2.3 Reinforcement Learning for Portfolio Management . . . . . . . . . . . . . 5
2.4 Deep Learning in Financial Forecasting . . . . . . . . . . . . . . . . . . . 6
2.5 Research Gaps and Contributions . . . . . . . . . . . . . . . . . . . . . . 6
3 Methodology: Three-Tier AI Framework 7
3.1 Framework Overview and Integration Architecture . . . . . . . . . . . . . 7
3.1.1 Tier 1: Market Regime Detection . . . . . . . . . . . . . . . . . . 7
3.1.2 Tier 2: LSTM Sector Forecasting . . . . . . . . . . . . . . . . . . 8
3.1.3 Tier 3: Reinforcement Learning Portfolio Allocation . . . . . . . 8
3.2 Reinforcement Learning Framework: Design and Integration . . . . . . . . 8
3.2.1 Environment Design and State Space Construction . . . . . . . . 8
3.2.2 Action Space and Constraints . . . . . . . . . . . . . . . . . . . . 9
3.2.3 Reward Function: Evolution from Over-Conservative to Balanced 9
3.2.4 PPO Implementation and Training Strategy . . . . . . . . . . . . 11
3.3 Data Sources and Preprocessing . . . . . . . . . . . . . . . . . . . . . . . 12
4 Model Development and Iterative Improvements 12
iv
4.1 Initial Implementation and Performance Challenges . . . . . . . . . . . . 13
4.1.1 Problem Identification and Root Cause Analysis . . . . . . . . . . 13
4.2 Systematic Improvement Process . . . . . . . . . . . . . . . . . . . . . . 14
4.2.1 Stage 1: Reward Function Optimization . . . . . . . . . . . . . . 14
4.2.2 Stage 2: Cash Management and Regime Responsiveness . . . . . 15
4.2.3 Stage 3: Data Quality Enhancement - Dividend Adjustment . . . . 15
4.3 Cash Management Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 16
4.4 Validation and Performance Attribution . . . . . . . . . . . . . . . . . . . 17
4.4.1 Component Contribution Analysis . . . . . . . . . . . . . . . . . 17
4.4.2 Out-of-Sample Validation . . . . . . . . . . . . . . . . . . . . . . 18
4.5 Complete Performance Evolution . . . . . . . . . . . . . . . . . . . . . . 19
5 Empirical Results 19
5.1 Performance Evolution Through Development Stages . . . . . . . . . . . 19
5.2 Implementation Timeline and Improvement Factors . . . . . . . . . . . . 21
5.3 Final Model Performance vs. VN-Index Benchmark . . . . . . . . . . . . 21
5.4 Component Contribution Analysis . . . . . . . . . . . . . . . . . . . . . . 22
5.5 Regime-Specific Performance Analysis . . . . . . . . . . . . . . . . . . . 22
5.6 Risk-Return Characteristics Analysis . . . . . . . . . . . . . . . . . . . . 23
6 Discussion 24
6.1 Critical Role of Reward Function Engineering in Financial RL . . . . . . . 24
6.1.1 Mathematical Structure and Behavioral Consequences . . . . . . 24
6.1.2 Multi-Objective Optimization Benefits . . . . . . . . . . . . . . . 24
6.2 Three-Tier Framework Integration Effectiveness . . . . . . . . . . . . . . 25
6.2.1 Information Flow and Decision Quality . . . . . . . . . . . . . . 25
6.2.2 Adaptive Behavior and Market Responsiveness . . . . . . . . . . 25
v
6.3 Comparison with Financial RL Literature . . . . . . . . . . . . . . . . . . 26
6.4 Methodological Contributions and Transparency . . . . . . . . . . . . . . 26
6.4.1 Documentation of Development Process . . . . . . . . . . . . . . 26
6.4.2 Systematic Improvement Methodology . . . . . . . . . . . . . . . 26
6.5 Practical Implementation Considerations . . . . . . . . . . . . . . . . . . 27
6.5.1 Computational Requirements and Scalability . . . . . . . . . . . 27
6.5.2 Risk Management and Robustness . . . . . . . . . . . . . . . . . 27
6.6 Limitations and Areas for Future Research . . . . . . . . . . . . . . . . . 27
6.6.1 Data Dependencies and Overfitting Risks . . . . . . . . . . . . . 27
6.6.2 Market Impact and Scalability . . . . . . . . . . . . . . . . . . . 28
6.6.3 Regime Detection Stability . . . . . . . . . . . . . . . . . . . . . 28
7 Conclusion 28
7.1 Summary of Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . 28
7.2 Theoretical and Practical Implications . . . . . . . . . . . . . . . . . . . . 29
7.3 Limitations and Transparency . . . . . . . . . . . . . . . . . . . . . . . . 29
7.4 Future Research Directions . . . . . . . . . . . . . . . . . . . . . . . . . 30
7.4.1 Methodological Extensions . . . . . . . . . . . . . . . . . . . . . 30
7.4.2 Market Applications . . . . . . . . . . . . . . . . . . . . . . . . 30
7.4.3 Risk Management Enhancements . . . . . . . . . . . . . . . . . . 30
7.5 Final Remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
