Integrating Financial Economics with AI in Stock Markets using Random Walk Hypothesis and Mean-Variance Utility Theory
랜덤 워크 가설과 평균-분산 효용 이론을 활용한 주식 시장에서의 인공지능과 금융경제학의 통합 연구
- 주제어 (키워드) AI in Financial Economics , Deep Learning , Stock Trend Prediction , Stock Investment Selection , Random Walk Hypothesis , Mean-Variance Utility Theory , 경제학에서의 인공지능 , 딥러닝 , 주가 트렌드 예측 , 주식 투자 선택 , 랜덤 워크 가설 , 평균-분산 효용 이론
- 발행기관 서강대학교 일반대학원
- 지도교수 양지훈
- 발행년도 2025
- 학위수여년월 2025. 2
- 학위명 박사
- 학과 및 전공 일반대학원 컴퓨터공학과
- 실제 URI http://www.dcollection.net/handler/sogang/000000079633
- UCI I804:11029-000000079633
- 본문언어 영어
- 저작권 서강대학교 논문은 저작권 보호를 받습니다.
목차
1 Introduction 1
1.1 Research Motivation and Goals 1
1.2 Chapter Abstract 3
1.2.1 DPUA: Decomposition of Price and Uncertainty Adjustment for
Stock Trend Prediction 3
1.2.2 MV-NUF: Mean-Variance Neural Utility Function for Stock Invest-
ment Selection 4
2 Related Work 5
2.1 Conventional Approach 5
2.1.1 Traditional Machine Learning-based Approach 5
2.1.2 Recurrent Neural Network-based Approach 6
2.2 Contemporary Approach 6
2.2.1 Attention-based Approach 7
2.2.2 Graph Convolutional Neural Network-based Approach 8
2.2.3 Hypergraph Neural Network-based Approach 9
2.3 Comparison to Our Approach 9
2.3.1 Comparison to DPUA 9
2.3.2 Comparison to MV-NUF 10
3 DPUA: Accurate Stock Trend Prediction through Decomposition of Price and Un-
certainty Adjustment 11
3.1 Introduction 11
3.2 Proposed Method: DPUA 15
3.2.1 Problem Formulation 15
3.2.2 Multi-Head Adaptive Exponential Smoothing with Attentive
Hawkes Process 15
3.2.3 Dual Multi-head Attention 19
3.2.4 Multi-graph Construction for STGRU 20
3.2.5 Predictor 23
3.2.6 Adjustment of Estimated Probability 24
3.2.7 Loss Function with Selective L2 Regularization 25
3.3 Experiment 26
3.3.1 Experimental Setup 26
3.3.2 Overall Performance Analysis 28
3.3.3 In-Depth Analysis 33
3.4 Conclusion 46
4 MV-NUF: Mean-Variance Neural Utility Function for Risk-Aware Stock Selection 47
4.1 Introduction 47
4.2 Prerequisite 50
4.2.1 Preference Representation by Utility Function 50
4.2.2 Axiom of Utility Function 51
4.2.3 Proof of Axioms for Mean-Variance Utility Function 51
4.2.4 Corollary of Utility Function 52
4.3 Proposed Method: MV-NUF 53
4.3.1 Problem Formulation 53
4.3.2 Design of λ1N 54
4.3.3 Proof of Quasi-Convexity 55
4.3.4 Overview of MV-NUF 56
4.3.5 Temporal Embedder 56
4.3.6 Return & Risk Predictor 59
4.3.7 Mean-Variance Utility Network 59
4.3.8 Training and Loss Functions 61
4.4 Experiment 63
4.4.1 Experimental Setup 63
4.4.2 Performance Comparison and Analysis 68
4.4.3 In-depth Analysis of Internal Processes 69
4.5 Discussion with The Financial Economics View 78
4.5.1 Interpretation of The Risk-Aversion Coefficient by Laplacian Matrix 78
4.5.2 Limitations of The Approximation of The Coefficient by Neural Net-
work 80
4.6 Conclusion 81
5 Conclusion 82
5.1 Summary 82
5.2 Boarder Impact 83
5.3 Future Work 84
