Early Warning System for an Emerging Market : Integrating Dynamic Connectedness, Regime Detection, and Cost-Sensitive Deep Learning
- 주제(키워드) Financial crises , Early warning systems , Dynamic connectedness
- 주제(DDC) 658
- 발행기관 아주대학교 일반대학원
- 지도교수 Ohsung Kwon
- 발행년도 2026
- 학위수여년월 2026. 8
- 학위명 석사
- 학과 및 전공 일반대학원 비즈니스애널리틱스학과
- 실제URI http://www.dcollection.net/handler/ajou/000000036423
- 본문언어 영어
- 저작권 아주대학교 논문은 저작권에 의해 보호받습니다.
초록/요약
This study develops a regime-aware early warning system for systemic financial stress in an emerging-market setting, using Indonesia as an empirical testbed. The framework integrates Time-Varying Parameter Vector Autoregression （TVP-VAR） connectedness measures, a Temporal Convolutional Network–Hidden Markov Model （TCN–HMM） for latent regime detection, and cost-sensitive predictive models to address structural instability and class imbalance. Using daily data from 2002 to 2024, connectedness analysis shows crude oil （WTI） is the dominant net transmitter of shocks, whereas the Indonesian equity market （JCI） primarily receives external spillovers. The TCN–HMM identified five distinct systemic regimes, with the crisis regime exhibiting the highest average connectedness. In forecasting, Random Forest achieved the highest Precision-Recall Area Under the Curve （PR-AUC）, whereas deep sequence models exhibited smaller generalization gaps but higher sensitivity to prediction horizons. Overall, combining network-based connectedness indicators with machine learning provides a useful early warning framework for non- stationary financial environments.
more목차
1. INTRODUCTION 1
1.1 Research Background 1
1.2 Statement of the Problem 4
1.3 Research Objectives 5
2. LITERATURE REVIEW 8
2.1 Spillover, Contagion, and Dynamic Systemic Risk 8
2.2 From Spillover Measurement to Regime-Sensitive Early Warning Signals 9
2.3 AI-Based Early Warning Systems and Generalization 12
2.4 Emerging Market Context and the Case of Indonesia 14
2.5 Conceptual Framework 16
3. DATA AND METHODOLOGY 18
3.1 Data and Variables 18
3.2 Methodology 21
4. RESULTS 26
4.1 Empirical Results 26
4.2 Discussion 45
4.3 Robustness Checks 50
5. CONCLUSION 55
5.1 Policy and Operational Implications 55
5.2 Conclusion and Future Research 56
REFERENCES 58
APPENDIX A. ADDITIONAL TABLES AND FIGURES 64
